main.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"page-top\"></a>\n",
    "<div style=\"text-align: justify\">\n",
    "NOTE : for better user experience, this notebook uses the <span style=\"background-color: #e0e0e0\">&nbsp;Python Markdown&nbsp;</span> Jupyter Notebook extension (@see the corresponding <a href = \"https://jupyter-contrib-nbextensions.readthedocs.io/en/latest/nbextensions/python-markdown/readme.html\" >ReadTheDoc page</a> and the associated <a href = \"https://github.com/ipython-contrib/jupyter_contrib_nbextensions\" >GitHub repo</a> for details). Please make sure it is installed and activated on your environment and that the current notebook is set as \"Trusted\" for it to work.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Sentiment Analysis in French"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 0. Introduction"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "We will be performing sentiment analysis on Tweeter data from the open dataset by Charles MALAFOSSE which can be found on the following <a href=\"https://github.com/charlesmalafosse/open-dataset-for-sentiment-analysis?files=1\">GitHub page</a>. Sentiment is there categorized as either POSITIVE, NEGATIVE, or NEUTRAL.<br />\n",
    "As we're focusing on NLP Sentiment Classification in French language, we will only consider the 250,000 tweets in French language from that dataset.\n",
    "<br />\n",
    "<pre style='background-color: rgba(0, 255, 255, 0.01);'><code style='background-color: rgba(0, 255, 255, 0.001); word-break: keep-all;'>As we will explain in the upcoming sections, we will be training a dual bidirectional-GRU NLP model (@see chapter <a href=\"#gru\"><b>&sect;4.</b></a>) using fastText French word embeddings (@see chapter <a href=\"#embeddings\"><b>&sect;3.</b></a>) and transfer learning (@see chapter <a href=\"#transfer\"><b>&sect;2.</b></a>). It will be first done using review data that we have scrapped from the Web (@see chapter <a href=\"#scraping\"><b>&sect;2.1.</b></a>), and then again on the French Tweets dataset (@see chapters <a href=\"#dataset\"><b>&sect;1.</b></a> and <a href=\"#warming\"><b>&sect;5.2.a</b></a>).<br />\n",
    "This will translate in 3 steps as described in the below figure : i) pre-training, ii) warming-up and iii) fine-tuning.\n",
    "<img src=\"images/training_graph.png\" width=\"100%\" />\n",
    "</code></pre>\n",
    "One important note : when training our model following the above-depicted 3-steps, the only difference beween the step 1 on one hand (pre-training) and steps 2 &amp; 3 on the other (warming-up &amp; fine tuning) consists in the model output layer.<br />\n",
    "It is important to note that the entire data pipeline otherwise remains identical. Neither the pre-processing nor the text tokenization, nor the dimensions of the model input layer (max number of tokens in the imput text) are modified.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Table of contents\n",
    "1. <a href=\"#dataset\">The Dataset</a><br />\n",
    "    1.1. <a href=\"#preprocessing\">Pre-processing</a><br />\n",
    "    1.2. <a href=\"#specificity\">Domain-specificity</a><br />\n",
    "    1.3. <a href=\"#imbalance\">High Imbalance Characteristic</a><br />\n",
    "<br />\n",
    "2. <a href=\"#transfer\">Transfer Learning</a><br />\n",
    "    2.1. <a href=\"#scraping\">Web Scraping</a><br />\n",
    "    2.2. <a href=\"#tokenizer\">Custom Text Tokenizer</a><br />\n",
    "<br />\n",
    "3. <a href=\"#embeddings\">Word Embeddings</a><br />\n",
    "    3.1. <a href=\"#fasttext\">fastText</a><br />\n",
    "    3.2. <a href=\"#ukn\">The &lt;UKN&gt; Token</a><br />\n",
    "<br />\n",
    "4. <a href=\"#gru\">Dual bi-GRU NLP Model</a><br />\n",
    "    4.1. <a href=\"#architecture\">Architecture</a><br />\n",
    "    4.2. <a href=\"#optimization\">Hyperparameters optimization</a><br />\n",
    "<br />\n",
    "5. <a href=\"#training\">NLP Model Training</a><br />\n",
    "    5.1. <a href=\"#pretraining\">Pre-training</a><br />\n",
    "    5.2. <a href=\"#warming\">Warming-Up</a><br />\n",
    "    5.3. <a href=\"#tuning\">Fine-tuning</a><br />\n",
    "<br />\n",
    "6. <a href=\"#conclusion\">Conclusion</a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"dataset\"></a>\n",
    "<h2>1. The Dataset <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os, gc, datetime\n",
    "import base64, cv2\n",
    "\n",
    "import numpy as np\n",
    "np.set_printoptions(suppress=True) #prevent numpy exponential \n",
    "                                   #notation on print\n",
    "import pandas as pd\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "plt.rcParams.update({'font.size': 12})\n",
    "%matplotlib inline\n",
    "\n",
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "from my_NLP_RNN_fr_lib.tweet_utils import \\\n",
    "    TWEET_MAX_CHAR_COUNT, html_escape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style type='text/css'>\n",
       ".CodeMirror{\n",
       "font-size: 8.5pt;\n",
       "}</style>\n"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%%html\n",
    "<style type='text/css'>\n",
    ".CodeMirror{\n",
    "font-size: 8.5pt;\n",
    "}</style>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "First, let us download French Tweets dataset, which consists of 3 zip files :\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "     100%|██████████████████████████████████████████████████| 2320/2320 [01:48<00:00, 21.37it/s]\n",
      "     100%|██████████████████████████████████████████████████| 3221/3221 [01:52<00:00, 28.58it/s]\n",
      "     100%|██████████████████████████████████████████████████| 3328/3328 [00:06<00:00, 543.04it/s]\n"
     ]
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.tweet_utils import download_tweet_fr ; download_tweet_fr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 451,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded the dataset in 2.6395 seconds\n"
     ]
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.tweet_utils import load_tweet_fr ; tweets = load_tweet_fr()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"preprocessing\"></a>\n",
    "<h3>1.1. Pre-processing <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "As far as data cleaning goes, we ensure that :\n",
    "    <ul style=\"margin: 0 5px 0 50px; /* top right bottom left*/\">\n",
    "        <li>no '@' mention,</li>\n",
    "        <li>no URL link,</li>\n",
    "        <li>no html tag,</li>\n",
    "        <li>no special character (such as '&oelig;', which we replace with 'oe') and</li>\n",
    "        <li>no duplicate posts</li>\n",
    "    </ul>\n",
    "remain in the body of the French Tweets we feed our model. This information does indeed not add value to build a Sentiment Analysis model.\n",
    "<br />\n",
    "<br />\n",
    "In addition to that, we also deal with hashtags (#) in a manner such that we get the better out of them. Below we can see 2 examples of how we handle them&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 452,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <!DOCTYPE html><html><head>\n",
       "        <style>\n",
       "        div.centered_table_container {\n",
       "            text-align: center;\n",
       "        }\n",
       "        table {\n",
       "            display: inline-block;\n",
       "        }\n",
       "        </style></head>\n",
       "        <body>\n",
       "        <div class=\"centered_table_container\"><style  type=\"text/css\" >\n",
       "    #T_27345b2c_f042_11ea_9a5b_309c23ff89f2 th {\n",
       "          text-align: center;\n",
       "    }    #T_27345b2c_f042_11ea_9a5b_309c23ff89f2row0_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_27345b2c_f042_11ea_9a5b_309c23ff89f2row0_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_27345b2c_f042_11ea_9a5b_309c23ff89f2row1_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_27345b2c_f042_11ea_9a5b_309c23ff89f2row1_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }</style><table id=\"T_27345b2c_f042_11ea_9a5b_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" >original</th>        <th class=\"col_heading level0 col1\" >converted</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_27345b2c_f042_11ea_9a5b_309c23ff89f2row0_col0\" class=\"data row0 col0\" >Allez #tousAvecLesBleus, g&eacute;nial!</td>\n",
       "                        <td id=\"T_27345b2c_f042_11ea_9a5b_309c23ff89f2row0_col1\" class=\"data row0 col1\" >Allez tous Avec Les Bleus, g&eacute;nial!</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_27345b2c_f042_11ea_9a5b_309c23ff89f2row1_col0\" class=\"data row1 col0\" >#AllezLePSG</td>\n",
       "                        <td id=\"T_27345b2c_f042_11ea_9a5b_309c23ff89f2row1_col1\" class=\"data row1 col1\" >Allez Le PSG</td>\n",
       "            </tr>\n",
       "    </tbody></table></div></body></html>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.tweet_utils import hashtag_converter_tweet_fr\n",
    "from my_NLP_RNN_fr_lib.display_helper import dataframe_pretty_print_center\n",
    "\n",
    "dummy_tweet_df = pd.DataFrame({'tweet_text':\n",
    "                               [html_escape('Allez #tousAvecLesBleus, génial!'), html_escape(\"#AllezLePSG\")]})\n",
    "result_df = pd.concat(\n",
    "    [dummy_tweet_df, hashtag_converter_tweet_fr(dummy_tweet_df[ 'tweet_text' ])], axis = 1 )\n",
    "result_df.columns = [ 'original', 'converted' ]\n",
    "dataframe_pretty_print_center(result_df) ; del result_df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Lets apply this data pre-processing to the whole dataset, then :\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 453,
   "metadata": {},
   "outputs": [],
   "source": [
    "from my_NLP_RNN_fr_lib.tweet_utils import clean_tweet_fr ; tweets = clean_tweet_fr(tweets)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "\"{:,}\".format(tweets.shape[0])": "219,599"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "The dataset is made up of {{\"{:,}\".format(tweets.shape[0])}}  records at this stage.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 454,
   "metadata": {},
   "outputs": [],
   "source": [
    "tweets_count = tweets.shape[0]\n",
    "tweets.drop( tweets[tweets[ 'tweet_text' ].map(\n",
    "    lambda txt: len(str(txt))) > TWEET_MAX_CHAR_COUNT].index, inplace=True )\n",
    "tweets.reset_index(drop=True, inplace=True)\n",
    "dropped_tweets_count = tweets_count - tweets.shape[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "\"{:,}\".format(dropped_tweets_count)": "287",
     "\"{:2.2f}%\".format(dropped_tweets_count/tweets_count*100)": "0.13%",
     "TWEET_MAX_CHAR_COUNT": "280"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "After the pre-processing of the French Twitter dataset, {{\"{:,}\".format(dropped_tweets_count)}} entries unfortunately ended up being more than {{TWEET_MAX_CHAR_COUNT}} characters long, mostly as a result of hashtags words splitting. In an effort to keeping things simple, we dropped those tweets (which barely accounted for {{\"{:2.2f}%\".format(dropped_tweets_count/tweets_count*100)}} of the entire dataset).\n",
    "<br />\n",
    "<p style=\"text-align: justify; color: darkgray;\">\n",
    "<u>REMARK&nbsp;</u>: any real-life entry that would exceed this max length would be truncated (as part of the <em><b><small>my_NLP_RNN_fr_lib.fastText.sentences_to_indices</small></b></em> input procedure).\n",
    "</p>\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"specificity\"></a>\n",
    "<h3>1.2. Domain-specificity <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "tweets_sample_size = 3 ; tweets_sample_size": "3"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "By design, the French Tweets dataset we're using is <u>specific to soccer</u>. To that extent, it can be considered to be <span style=\"background-color: #e0e0e0\">&nbsp;domain-specific&nbsp;</span>.\n",
    "<br />\n",
    "<br />\n",
    "Below are displayed {{tweets_sample_size = 3 ; tweets_sample_size}} random tweets from that dataset :\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "<div style=\"margin-left:45px;text-align: justify;\"><span style=\"color:darkred;padding-left:45px\"> - ?? Allez les bleus?? Nice Magnifique France Nice 06 Nice Moments Nissa La Bella frenchriviera I Lovenice Coted Azur France Metropole NCA</span></div>\n"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/markdown": [
       "<div style=\"margin-left:45px;text-align: justify;\"><span style=\"color:darkred;padding-left:45px\"> - On regrette que tu n'aies pas &eacute;t&eacute; l&agrave; pour ce match, quoique, c'est pas tout &agrave; fait s&ucirc;r ?. Rendez -vous au Parc ?</span></div>\n"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/markdown": [
       "<div style=\"margin-left:45px;text-align: justify;\"><span style=\"color:darkred;padding-left:45px\"> - Encore 8 ?? 5 ?? jours ou je vais mentionner ? pour qu'il m'envoie son maillot ! ?? Mendy Manchester City</span></div>\n"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import random\n",
    "from IPython.display import Markdown, display\n",
    "def printmd(string):\n",
    "    # @see 'https://stackoverflow.com/questions/23271575#46934204'\n",
    "    # (printing-bold-colored-etc-text-in-ipython-qtconsole)\n",
    "    display(Markdown(string))\n",
    "\n",
    "dummy = [printmd('<div style=\"margin-left:45px;text-align: justify;\">' +\n",
    "                 '<span style=\"color:darkred;padding-left:45px\">' +\n",
    "                 \" - \" + html_escape(tweet['tweet_text']) + \"</span></div>\\n\" )\n",
    "         for i, tweet in\n",
    "         tweets.iloc[random.sample(range(0, tweets.shape[0]-1), tweets_sample_size)].iterrows()]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"imbalance\"></a>\n",
    "<h3>1.3. High Imbalance Characteristic <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "highly inbalanced => downsampling => small dataset => need for transfer learning (from \"generic\" domain [resto+movies+appliances] to domain-specific [soccer])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Lets have a look at how tweets are distributed among the three classes that are \"NEGATIVE\", \"NEUTRAL\" and \"POSITIVE\"&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 488,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "<!DOCTYPE html>\n",
       "<html>\n",
       "<head>\n",
       "<style>\n",
       "\n",
       "* {\n",
       "    box-sizing: border-box;\n",
       "}\n",
       "\n",
       ".left_column {\n",
       "    float: left;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "}\n",
       "\n",
       ".right_column {\n",
       "    float:right;\n",
       "    text-align; right;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "    #background: yellow;\n",
       "}\n",
       "\n",
       "/* Clearfix (clear floats) */\n",
       ".row::after {\n",
       "    content: \"\";\n",
       "    clear: both;\n",
       "    display: table;\n",
       "}\n",
       "\n",
       "</style>\n",
       "</head>\n",
       "<body>\n",
       "\n",
       "<h2></h2>\n",
       "\n",
       "<div class=\"row\" style=\"width: 100%;\">\n",
       "  <div class=\"left_column\" style=\"width: 50.00%;\">\n",
       "      <center>\n",
       "          <style  type=\"text/css\" >\n",
       "    #T_d961e058_f045_11ea_8719_309c23ff89f2 th {\n",
       "          text-align: center;\n",
       "    }    #T_d961e058_f045_11ea_8719_309c23ff89f2row0_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_d961e058_f045_11ea_8719_309c23ff89f2row0_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_d961e058_f045_11ea_8719_309c23ff89f2row0_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_d961e058_f045_11ea_8719_309c23ff89f2row1_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_d961e058_f045_11ea_8719_309c23ff89f2row1_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_d961e058_f045_11ea_8719_309c23ff89f2row1_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_d961e058_f045_11ea_8719_309c23ff89f2row2_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_d961e058_f045_11ea_8719_309c23ff89f2row2_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_d961e058_f045_11ea_8719_309c23ff89f2row2_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }</style><table id=\"T_d961e058_f045_11ea_8719_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" >sentiment</th>        <th class=\"col_heading level0 col1\" >count</th>        <th class=\"col_heading level0 col2\" >percentage</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_d961e058_f045_11ea_8719_309c23ff89f2row0_col0\" class=\"data row0 col0\" >NEGATIVE</td>\n",
       "                        <td id=\"T_d961e058_f045_11ea_8719_309c23ff89f2row0_col1\" class=\"data row0 col1\" >6,069</td>\n",
       "                        <td id=\"T_d961e058_f045_11ea_8719_309c23ff89f2row0_col2\" class=\"data row0 col2\" >2.76%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_d961e058_f045_11ea_8719_309c23ff89f2row1_col0\" class=\"data row1 col0\" >NEUTRAL</td>\n",
       "                        <td id=\"T_d961e058_f045_11ea_8719_309c23ff89f2row1_col1\" class=\"data row1 col1\" >182,049</td>\n",
       "                        <td id=\"T_d961e058_f045_11ea_8719_309c23ff89f2row1_col2\" class=\"data row1 col2\" >82.90%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_d961e058_f045_11ea_8719_309c23ff89f2row2_col0\" class=\"data row2 col0\" >POSITIVE</td>\n",
       "                        <td id=\"T_d961e058_f045_11ea_8719_309c23ff89f2row2_col1\" class=\"data row2 col1\" >31,481</td>\n",
       "                        <td id=\"T_d961e058_f045_11ea_8719_309c23ff89f2row2_col2\" class=\"data row2 col2\" >14.34%</td>\n",
       "            </tr>\n",
       "    </tbody></table>\n",
       "      </center>\n",
       "  </div>\n",
       "  <div class=\"right_column\" style=\"width: 50.00%;\">\n",
       "      <center>\n",
       "          <img src=\"data:image/jpeg;charset=utf-8;base64, /9j/4AAQSkZJRgABAQEASABIAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQoKCgoKBggLDAsKDAkKCgr/2wBDAQICAgICAgUDAwUKBwYHCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgr/wAARCAEFAZUDASIAAhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAAAgEDAwIEAwUFBAQAAAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkKFhcYGRolJicoKSo0NTY3ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWGh4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl5ufo6erx8vP09fb3+Pn6/8QAHwEAAwEBAQEBAQEBAQAAAAAAAAECAwQFBgcICQoL/8QAtREAAgECBAQDBAcFBAQAAQJ3AAECAxEEBSExBhJBUQdhcRMiMoEIFEKRobHBCSMzUvAVYnLRChYkNOEl8RcYGRomJygpKjU2Nzg5OkNERUZHSElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6goOEhYaHiImKkpOUlZaXmJmaoqOkpaanqKmqsrO0tba3uLm6wsPExcbHyMnK0tPU1dbX2Nna4uPk5ebn6Onq8vP09fb3+Pn6/9oADAMBAAIRAxEAPwD9/KKKKACiiigAooooAKKKKAPF4f8AgpD/AME77iJZ4P29fgu6OoZHT4paQQwPIIIuORTv+Hjf/BPT/o/D4M/+HQ0n/wCSKP8AgnJ/yj0+A/8A2Rnwv/6abavZqAPGf+Hjf/BPT/o/D4M/+HQ0n/5Io/4eN/8ABPT/AKPw+DP/AIdDSf8A5Ir2aigDxn/h43/wT0/6Pw+DP/h0NJ/+SKP+Hjf/AAT0/wCj8Pgz/wCHQ0n/AOSK9mooA8Z/4eN/8E9P+j8Pgz/4dDSf/kij/h43/wAE9P8Ao/D4M/8Ah0NJ/wDkivZqKAPGf+Hjf/BPT/o/D4M/+HQ0n/5Io/4eN/8ABPT/AKPw+DP/AIdDSf8A5Ir2aigDxn/h43/wT0/6Pw+DP/h0NJ/+SKP+Hjf/AAT0/wCj8Pgz/wCHQ0n/AOSK9mooA8Z/4eN/8E9P+j8Pgz/4dDSf/kij/h43/wAE9P8Ao/D4M/8Ah0NJ/wDkivZqKAPGf+Hjf/BPT/o/D4M/+HQ0n/5Io/4eN/8ABPT/AKPw+DP/AIdDSf8A5Ir2aigDxn/h43/wT0/6Pw+DP/h0NJ/+SKP+Hjf/AAT0/wCj8Pgz/wCHQ0n/AOSK9mrj/iD+0N8APhLrMXhz4q/HLwf4Z1Ge1W5hsPEHia1sppISzKJVSaRWKFkdQwGCUYdjWlOlVrS5acXJ9km39yTf4GdWrSow5qklFd20l97aX4nE/wDDxv8A4J6f9H4fBn/w6Gk//JFH/Dxv/gnp/wBH4fBn/wAOhpP/AMkVs/8ADa/7Gn/R23wx/wDC907/AOPUf8Nr/saf9HbfDH/wvdO/+PVv/Z+Yf8+Z/wDgE/8A5E5/7Ry7/n/D/wADh/8AJmN/w8b/AOCen/R+HwZ/8OhpP/yRR/w8b/4J6f8AR+HwZ/8ADoaT/wDJFbP/AA2v+xp/0dt8Mf8AwvdO/wDj1H/Da/7Gn/R23wx/8L3Tv/j1H9n5h/z5n/4BP/5EP7Ry7/n/AA/8Dh/8mY3/AA8b/wCCen/R+HwZ/wDDoaT/APJFH/Dxv/gnp/0fh8Gf/DoaT/8AJFbP/Da/7Gn/AEdt8Mf/AAvdO/8Aj1H/AA2v+xp/0dt8Mf8AwvdO/wDj1H9n5h/z5n/4BP8A+RD+0cu/5/w/8Dh/8mY3/Dxv/gnp/wBH4fBn/wAOhpP/AMkUf8PG/wDgnp/0fh8Gf/DoaT/8kV6T8Pvij8M/i1o0viP4VfETQvE2nQXTW01/4f1eG9hjmCqxiZ4WZQ4V0YqTkB1PcVu1zThOnJxmmmujTT+52f4HVCcKkFKDTT2aaafzTa/E8Z/4eN/8E9P+j8Pgz/4dDSf/AJIo/wCHjf8AwT0/6Pw+DP8A4dDSf/kivZqKko8Z/wCHjf8AwT0/6Pw+DP8A4dDSf/kij/h43/wT0/6Pw+DP/h0NJ/8AkivZqKAPGf8Ah43/AME9P+j8Pgz/AOHQ0n/5Io/4eN/8E9P+j8Pgz/4dDSf/AJIr2aigDxn/AIeN/wDBPT/o/D4M/wDh0NJ/+SKP+Hjf/BPT/o/D4M/+HQ0n/wCSK9mooA8Z/wCHjf8AwT0/6Pw+DP8A4dDSf/kij/h43/wT0/6Pw+DP/h0NJ/8AkivZqKAPGf8Ah43/AME9P+j8Pgz/AOHQ0n/5Io/4eN/8E9P+j8Pgz/4dDSf/AJIr2aigDxn/AIeN/wDBPT/o/D4M/wDh0NJ/+SKP+Hjf/BPT/o/D4M/+HQ0n/wCSK9mooA8Z/wCHjf8AwT0/6Pw+DP8A4dDSf/kimzf8FIf+Cd9vE08/7evwXREUs7v8UtIAUDkkk3HAr2ivGf8Ago3/AMo9Pjx/2RnxR/6abmgD2aiiigAooooAKKKKACiiigAooooA8Z/4Jyf8o9PgP/2Rnwv/AOmm2r2avGf+Ccn/ACj0+A//AGRnwv8A+mm2r2agAooooAKKKKACiiigAooooAKKKKACiiigAooooAK/Hz/g4H/5PL8M/wDZMbL/ANOOo1+wdfj5/wAHA/8AyeX4Z/7JjZf+nHUa+48O/wDkpo/4J/kj4LxK/wCSWl/jp/nI+FqKKK/fD+eAooooAKKKKAP2E/4N+P8AkzTxN/2U69/9N2nV90V8L/8ABvx/yZp4m/7Kde/+m7Tq+6K/m3i3/kpsX/jf5RP6f4O/5JbB/wCBfnIKKKK+dPpQooooAKKKKACiiigAooooAKKKKACiiigArxn/AIKN/wDKPT48f9kZ8Uf+mm5r2avGf+Cjf/KPT48f9kZ8Uf8AppuaAPZqKKKACvMPB/7Y37PXjv45eJf2cPDXjDUJPGHhLTJNR1zTrrwtqVtClrHMIZJIrqa3W3uQsjBSIZJCCeRXp9fGXiz40aLB/wAFYLi8u/g/8W7rSLf4K3nhW41ux+C3iSTTJdTOpx3fkR6gLD7I6tCrETiXyS2E8zcQKAPcf2eP24v2ef2pPC58efCG/wDF58PHQIdbt/Evin4Ya/4f0y706VBJHc295qtjbQXMbRkSAxO/yEP93mrH7PH7av7Nn7VOo32kfBHx7dahd2Om22pG11Pw3qOlSXWn3BdYL+1W/t4TeWchjcJdQeZC5XhzXxZ4Z/Z68TeM/h78Rf2TP2BdF+Nug/CTVf2V/EfhJdP+NEHiKztdN8SyQW9locWlL4kC3cKJbNficW4Fmqra7cMK7nwL8a5YvjHB+3H4h/Z++IXgvwT8Dv2V9a03xrZeJPBlzpN1calJc6ZfHTrGK8WIXxtotHuVE8LNbM15GI5m3NgA+lPjf+3P+y3+zl46tvhx8YfiZJpeqzWEF/drBoF/eW+k2U07W8N3qNzbQSQ6ZbSTI8aT3bwxs0bgMdjYPin+3P8Ast/Bb4q2/wAF/iR8TJLHXpfsH2sQ6Bf3NlpX22YwWX9oXsED2um+fKpSL7VLF5hGF3V8n/ta6f8AFGx1b9pzw5B+zJ8QPEN3+0v8FNJ0r4cJpvhK4vbeLUW0u/06TStUuLdZINKEElzHcPLcukBS4l2SO6MtYH7Sfwb+NXhXwj+1B+yNH8IPGvijxJ+0FpHh+0+HHi/RvC13eaWJP+Ec07Qp5Ly/jRoNN+x3NjNfN9qeLck4MXmOSoAPs/xr+3P+y38PPjZD+z34u+Jklt4mkvrCxuEj0C/msLC8vsfYbS71COBrOyuLjcnkwTzRyS+YmxW3rn1qvzr+P/w6+LcbftC/sgWPwP8AGmp+IvjV8cdA8TeCPF+neFrqfRBpjW/h+Ka8u9TRDbWUlgdLuf3M8iSuIYPJSQyAD70+Jfg3xH478Lv4f8LfFjxB4Lu2mRxrnhm206W6VVPMYXULS6g2t0OYi3oRQB5z/wAE5P8AlHp8B/8AsjPhf/0021ezV4x/wTgUp/wTx+AqNIXI+DHhYFmxk/8AEpteTjivZ6ACiiigAooooAKKKKACiiigAooooAKKKKACiiigAr8fP+Dgf/k8vwz/ANkxsv8A046jX7B1+Pn/AAcD/wDJ5fhn/smNl/6cdRr7jw7/AOSmj/gn+SPgvEr/AJJaX+On+cj4Wooor98P54CiiigAooooA/YT/g34/wCTNPE3/ZTr3/03adX3RXwv/wAG/H/Jmnib/sp17/6btOr7or+beLf+Smxf+N/lE/p/g7/klsH/AIF+cgooor50+lCiiigAooooAKKKKACiiigAooooAKKKKACvGf8Ago3/AMo9Pjx/2RnxR/6abmvZq8Z/4KN/8o9Pjx/2RnxR/wCmm5oA9mooooAKKKKACiiigAooooAKKKKAPGf+Ccn/ACj0+A//AGRnwv8A+mm2r2avGf8AgnJ/yj0+A/8A2Rnwv/6abavZqACiiigAooooAKKKKACiiigAooooAKKKKACiiigAr8fP+Dgf/k8vwz/2TGy/9OOo1+wdfj5/wcD/APJ5fhn/ALJjZf8Apx1GvuPDv/kpo/4J/kj4LxK/5JaX+On+cj4Wooor98P54CiiigAooooA/YT/AIN+P+TNPE3/AGU69/8ATdp1fdFfC/8Awb8f8maeJv8Asp17/wCm7Tq+6K/m3i3/AJKbF/43+UT+n+Dv+SWwf+BfnIKKKK+dPpQooooAKKKKACiiigAooooAKKKKACiiigArxn/go3/yj0+PH/ZGfFH/AKabmvZq8Z/4KN/8o9Pjx/2RnxR/6abmgD2aiiigAooooAKKKKACiiigAooooA8Z/wCCcn/KPT4D/wDZGfC//pptq9mrxn/gnJ/yj0+A/wD2Rnwv/wCmm2r2agAooooAKKKKACiiigAooooAKKKKACiiigAooooAK/Hz/g4H/wCTy/DP/ZMbL/046jX7B1+Pn/BwP/yeX4Z/7JjZf+nHUa+48O/+Smj/AIJ/kj4LxK/5JaX+On+cj4Wooor98P54CiiigAooooA/YT/g34/5M08Tf9lOvf8A03adX3RXwv8A8G/H/Jmnib/sp17/AOm7Tq+6K/m3i3/kpsX/AI3+UT+n+Dv+SWwf+BfnIKKKK+dPpQooooAKKKKACiiigAooooAKKKKACiiigArxn/go3/yj0+PH/ZGfFH/ppua9mrxn/go3/wAo9Pjx/wBkZ8Uf+mm5oA9mooooAK+L/H+seOf2dP2/fiV4l8H/ABH8b+ILef8AZq1jxjF4S8QeLr2+0uLU7fUVEItbJ5PJtRtHl4iRWKsQSa+0K8al/YI/Zxm/aS/4a2mh8dP452tF9tb4veJTY/Zywc2v9nf2h9i+ylwHNt5Hk7gG2buaAPi6/wDjL8cv2NvhJ4T+M/w/+PPjP4gaz8QP2OfGvxB8R2vjDxNc6vbXHiLSdM0rULXU7K3ndo9PhaW/mga1tVit2WWACMMgJ9w/ZX07xd8Bv2yfBfwP0z49eNvHWgeP/wBnu88WeIZvGvi661lk1qw1HSbdb63a5d/saXceqXBNtB5duPsiGOJMNn2b4J/sFfsnfs86rqOr/Cr4Ti1fUtFbRTb6nrl9qdvY6Uz+Y2mWUF7PLFp9iXwxtLZYoCVXKfKuMPR/+Caf7J3hD4VeOvhL8PfC2t6Fa/EDwTL4S1XVF8VX2o31hozwzRJY2EupS3IsbaJZ5TFbQqsEbNkR8CgD59/bX8T/ABO8XeN/2qviDZftBeM/Btz+zr8IdO1z4a2XhzxTc6fp8eoHTtQ1OTUNQtYnWLVIpZLeK2MN0ssIjtpAiK7s9cp+0p8Y/jX4r8I/tQftcxfF/wAa+F/Ef7PukeH7v4ceD9G8U3dnpZk/4RzTtdnS8sI3WDUvtlzfTWLfakl2pABF5bgsfsL4m/8ABPj9kL4wa1pGu+PvhH9ok0fRLPRktrLX9QsrXUNMtJDLbWGoW1tcRw6naxSMzJBeJNGpdyFG9s3fil+wz+y38aPirb/Gj4kfDSS+16L7B9rMPiC/trLVfsUxnsv7QsYJ0tdS8iVi8X2qKXyycrtoA+Rf2gPiL8XZG/aF/a/sPjj400zxD8FfjjoHhnwR4Q07xTdQaIdLWDw/LNZ3emI4tr2S/OqXP76aN5UE0HkvGYwT96fEvxl4j8CeF38QeFvhP4g8aXazIg0Pwzc6dFdMrHmQNqF3awbV6nMob0BrhfGv7DH7LfxD+NkP7Qfi74aSXPiaO+sL64aPxBfw2F/eWOPsN3d6fFOtne3FvtTyZ54ZJIvLj2MuxcetUAeMf8E4GL/8E8fgK7RlCfgx4WJVsZH/ABKbXg44r2evGf8AgnJ/yj0+A/8A2Rnwv/6abavZqACiiigAooooAKKKKACiiigAooooAKKKKACiiigAr8fP+Dgf/k8vwz/2TGy/9OOo1+wdfj5/wcD/APJ5fhn/ALJjZf8Apx1GvuPDv/kpo/4J/kj4LxK/5JaX+On+cj4Wooor98P54CiiigAooooA/YT/AIN+P+TNPE3/AGU69/8ATdp1fdFfC/8Awb8f8maeJv8Asp17/wCm7Tq+6K/m3i3/AJKbF/43+UT+n+Dv+SWwf+BfnIKKKK+dPpQooooAKKKKACiiigAooooAKKKKACiiigArxn/go3/yj0+PH/ZGfFH/AKabmvZq8Z/4KN/8o9Pjx/2RnxR/6abmgD2aiiigAooooAKKKKACiiigAooooA8Z/wCCcn/KPT4D/wDZGfC//pptq9mrxn/gnJ/yj0+A/wD2Rnwv/wCmm2r2agAooooAKKKKACiiigAooooAKKKKACiiigAooooAK/Hz/g4H/wCTy/DP/ZMbL/046jX7B1+Pn/BwP/yeX4Z/7JjZf+nHUa+48O/+Smj/AIJ/kj4LxK/5JaX+On+cj4Wooor98P54CiiigAooooA/YT/g34/5M08Tf9lOvf8A03adX3RXwv8A8G/H/Jmnib/sp17/AOm7Tq+6K/m3i3/kpsX/AI3+UT+n+Dv+SWwf+BfnIKKKK+dPpQooooAKKKKACiiigAooooAKKKKACiiigArxn/go3/yj0+PH/ZGfFH/ppua9mrxn/go3/wAo9Pjx/wBkZ8Uf+mm5oA9mooooAKKKKACivkn4Kf8ABTmf4s/Hj4jfDvVdI+EOieH/AIY+IvEdh4kQ/GyS58XRWektKjaidAXSQEgkaNTk3fyoxYFyAjdF+zB+3l4++M/xC8HeC/jF+zivgGD4peALnxn8L7uPxcuqS3+nW72fnW+oRC2hFhfLHqFpL5Eb3MZV5AJi0TCgD6Uor5x/ad/bl+Inwg+Jnib4a/BD9nFPiDP8PPh5B42+I8k/i9dKaw0yeW8jtreyU204vb+UadfOsEjW0YWFd06mRRXNfF7/AIKf3vhxPEvjT4D/AADHj7wN8O/hvpfjr4k+JG8VDTriz0m+gmu4k060NrL/AGhdpZW73TwySWyhHiVZHd9gAPrOivlP44f8FMdQ+GXibx3r3gX4FQ+KPhr8IbbSJ/ix43Pi0Wdzp0d9bQ3pawsfssgvxb2Fxb3c3mT22I5gI/NcFa+rKAPGf+Ccn/KPT4D/APZGfC//AKabavZq8Z/4Jyf8o9PgP/2Rnwv/AOmm2r2agAooooAKKKKACiiigAooooAKKKKACiiigAooooAK/Hz/AIOB/wDk8vwz/wBkxsv/AE46jX7B1+Pn/BwP/wAnl+Gf+yY2X/px1GvuPDv/AJKaP+Cf5I+C8Sv+SWl/jp/nI+FqKKK/fD+eAooooAKKKKAP2E/4N+P+TNPE3/ZTr3/03adX3RXwv/wb8f8AJmnib/sp17/6btOr7or+beLf+Smxf+N/lE/p/g7/AJJbB/4F+cgooor50+lCiiigAooooAKKKKACiiigAooooAKKKKACvGf+Cjf/ACj0+PH/AGRnxR/6abmvZq8Z/wCCjf8Ayj0+PH/ZGfFH/ppuaAPZqKKKACiiigD5J/aE/Y//AGi/2yPiH4c8J/GzwV8MfCfgHwj4s1bVbfxF4S8S3t/4g1i2udPv7AW3kTadbxaaZY74tcMtxdB9hQDD71zfhB+yz+2v4C1Dwh8SPi9D4E13UvgP8GNX8K/DLTvCGq3Ly+MdSuIrFFv75byK2j05mTTIYhbJNMga7mZrkKqgfZFFAHx98Vf2V/26tY8Qa98UvhfrHwvTxT8Xvgho/gr4qrq97qFpZ6DqtkNQI1fS444blrxAdWvF+xzSQbhDbk3AO/OF8S/+CcP7QnhHw946+Cn7Kvinwangj4s/CHRfAPijUvF+oXcWpeGlsNPm0o6lZQwW0sWoSSWEsai3lktlSa3V/MdXZB9vUUAfFHx2/wCCdXx68Qn4vfA34L634Mh+Fnx/sNHs/HN9r2p3ceseHYrbS7XRr0WVvFbSQ35uNNsbZE8ye28mUO584EIPrf4l/Cb4V/Gjwu/gf4xfDTw/4s0WSZJn0fxNo0F/atIhyjmKdGQsp6HGR2roKKAPGP8AgnBGkX/BPH4CxRIFVfgx4WCqowAP7JteBXs9eM/8E5P+UenwH/7Iz4X/APTTbV7NQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFfj5/wcD/APJ5fhn/ALJjZf8Apx1Gv2Dr8fP+Dgf/AJPL8M/9kxsv/TjqNfceHf8AyU0f8E/yR8F4lf8AJLS/x0/zkfC1FFFfvh/PAUUUUAFFFFAH7Cf8G/H/ACZp4m/7Kde/+m7Tq+6K+F/+Dfj/AJM08Tf9lOvf/Tdp1fdFfzbxb/yU2L/xv8on9P8AB3/JLYP/AAL85BRRRXzp9KFFFFABRRXH/EH9ob4AfCXWYvDnxV+OXg/wzqM9qtzDYeIPE1rZTSQlmUSqk0isULI6hgMEow7GtKdKrWly04uT7JNv7km/wM6tWlRhzVJKK7tpL720vxOworzL/htf9jT/AKO2+GP/AIXunf8Ax6j/AIbX/Y0/6O2+GP8A4Xunf/Hq3/s/MP8AnzP/AMAn/wDInP8A2jl3/P8Ah/4HD/5M9NorzL/htf8AY0/6O2+GP/he6d/8eo/4bX/Y0/6O2+GP/he6d/8AHqP7PzD/AJ8z/wDAJ/8AyIf2jl3/AD/h/wCBw/8Akz02ivMv+G1/2NP+jtvhj/4Xunf/AB6j/htf9jT/AKO2+GP/AIXunf8Ax6j+z8w/58z/APAJ/wDyIf2jl3/P+H/gcP8A5M9NorC+H3xR+Gfxa0aXxH8KviJoXibToLpraa/8P6vDewxzBVYxM8LMocK6MVJyA6nuK3a5pwnTk4zTTXRpp/c7P8DqhOFSClBpp7NNNP5ptfiFeM/8FG/+Uenx4/7Iz4o/9NNzXs1eM/8ABRv/AJR6fHj/ALIz4o/9NNzUlHs1FFFABRRRQAUUUUAFFFFABRRRQB4z/wAE5P8AlHp8B/8AsjPhf/0021ezV4z/AME5P+UenwH/AOyM+F//AE021ezUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABX4+f8HA//ACeX4Z/7JjZf+nHUa/YOvx8/4OB/+Ty/DP8A2TGy/wDTjqNfceHf/JTR/wAE/wAkfBeJX/JLS/x0/wA5HwtRRRX74fzwFFFFABRRRQB+wn/Bvx/yZp4m/wCynXv/AKbtOr7or4X/AODfj/kzTxN/2U69/wDTdp1fdFfzbxb/AMlNi/8AG/yif0/wd/yS2D/wL85BRRRXzp9KFFFFABX4+f8ABwP/AMnl+Gf+yY2X/px1Gv2Dr8fP+Dgf/k8vwz/2TGy/9OOo19x4d/8AJTR/wT/JHwXiV/yS0v8AHT/OR8LUUUV++H88BRRRQAUUUUAfsJ/wb8f8maeJv+ynXv8A6btOr7or4X/4N+P+TNPE3/ZTr3/03adX3RX828W/8lNi/wDG/wAon9P8Hf8AJLYP/AvzkFeM/wDBRv8A5R6fHj/sjPij/wBNNzXs1eM/8FG/+Uenx4/7Iz4o/wDTTc186fSns1FFFABRRRQAUUUUAFFFFAHk/wC0F+2/+zF+y3r9p4Z+N3xGn0u9udLk1SaKy8PahqI07To5BG+oXz2cEq6fZq52m6uTFCCGG/5Wx6rb3EF1Al1azJJFIgeOSNgVZSMggjqCO9fEv7XSeOvg7+0p8bvGNx+z/wCN/Hlh8ZP2fNH8LeCD4Q8J3WrJ/bFhN4gEul3bwI6afFMNXtZFubkxW3E++QFAD9D/AA0+BfxE8P8A7JPgL4Ey/GnXvC3iDw14P0fTNS8TeFodPnuZJ7W0ihl2jUbS6hKuyEkmLdjoVoAz/wDgnJ/yj0+A/wD2Rnwv/wCmm2r2avn34Y/sM/EL4P8Aw28PfCTwD/wUL+NNpoXhbQ7TSNFtZNK8GStDaW0KQwoXfw8WchEUbmJJxkkmtz/hl/43f9JG/jN/4I/BX/zO0AezV4X+2n/wUE+DX7Cn/CNf8Lc8M+J9R/4Sn7Z/Z/8AwjllbzeX9m8jzPM86eLGftCbcbs4bOMDN7/hl/43f9JG/jN/4I/BX/zO15l+0b/wSY8Pftaf2N/w0F+2r8ZvEH/CP/aP7I/deFrTyPP8rzf+PbQo927yY/vZxt4xk57stll0cbF46MnS1uou0tna3zt8jgzSOZTwM1gJRjV05XJXjur3Xpe3mcV/xED/ALGn/RM/id/4JtO/+T6P+Igf9jT/AKJn8Tv/AATad/8AJ9Y3/EOj+yf/ANHH/Gb/AMGWg/8Ayoo/4h0f2T/+jj/jN/4MtB/+VFfVfWPDz/nzX/8AAl/mfI/V/Ej/AJ/4f/wF/wCRs/8AEQP+xp/0TP4nf+CbTv8A5Po/4iB/2NP+iZ/E7/wTad/8n1jf8Q6P7J//AEcf8Zv/AAZaD/8AKiuR/Z//AODfL4NeLvgP4J8V/Gj42fGHR/GOp+EdNu/FmkQXuhxpY6nJaxvdQKraU5UJMzqAWYgLjcetH1jw8/581/8AwJf5h9X8SP8An/h//AX/AJHo/wDxED/saf8ARM/id/4JtO/+T6P+Igf9jT/omfxO/wDBNp3/AMn1jf8AEOj+yf8A9HH/ABm/8GWg/wDyoo/4h0f2T/8Ao4/4zf8Agy0H/wCVFH1jw8/581//AAJf5h9X8SP+f+H/APAX/kfVX7Hn7Yfwz/bZ+Gd98VfhVoeu6fp2n67LpM0PiC2himaaOGGYsohmlUptnQAlgchuOhPq9fL/AOz7/wAEyn/ZZ8GXPw++A/7c/wAZtC0i81N9RubT7J4Tut9y8ccbSb7jQZGGUhjGAcfLnGSSe6/4Zf8Ajd/0kb+M3/gj8Ff/ADO18ljngpYybwiapX91S1dtN/Pc+xwEcdHBU1jGnVt7zjpFvXby2PZqK8Z/4Zf+N3/SRv4zf+CPwV/8ztH/AAy/8bv+kjfxm/8ABH4K/wDmdrkOw9morxn/AIZf+N3/AEkb+M3/AII/BX/zO0f8Mv8Axu/6SN/Gb/wR+Cv/AJnaAPZq/Hz/AIOB/wDk8vwz/wBkxsv/AE46jX6O/wDDL/xu/wCkjfxm/wDBH4K/+Z2vHv2gv+CM3w6/am8Z23xB+PH7Ynxm13V7PTE062u8+GbXZbJJJIsey30SNTh5pDkjPzYzgAD6PhXN8NkecLFV03FRkvdtfVK27XY+Z4uyXFZ/krwmHaUnKL966Vott7J9+x+LVFfrX/xDo/sn/wDRx/xm/wDBloP/AMqKP+IdH9k//o4/4zf+DLQf/lRX6d/xEvIv+fdT7o//ACR+Vf8AELc//wCftL75f/IH5KUV+nv7QH/Bvl8GvCPwH8beK/gv8bPjDrHjHTPCOpXfhPSJ73Q5EvtTjtZHtYGVdKQsHmVFIDKSGxuHWuu/4h0f2T/+jj/jN/4MtB/+VFH/ABEvIv8An3U+6P8A8kH/ABC3P/8An7S++X/yB+SlFfrX/wAQ6P7J/wD0cf8AGb/wZaD/APKij/iHR/ZP/wCjj/jN/wCDLQf/AJUUf8RLyL/n3U+6P/yQf8Qtz/8A5+0vvl/8gbP/AAb8f8maeJv+ynXv/pu06vuivl/9n3/gmU/7LPgy5+H3wH/bn+M2haReam+o3Np9k8J3W+5eOONpN9xoMjDKQxjAOPlzjJJPdf8ADL/xu/6SN/Gb/wAEfgr/AOZ2vyPPMdSzPOK+KpJqM5XV99lva/Y/ZMgwFbK8loYSq05QjZ22vdvS9u/Y9morxn/hl/43f9JG/jN/4I/BX/zO0f8ADL/xu/6SN/Gb/wAEfgr/AOZ2vKPXPZqK8Z/4Zf8Ajd/0kb+M3/gj8Ff/ADO0f8Mv/G7/AKSN/Gb/AMEfgr/5naAPZq/Hz/g4H/5PL8M/9kxsv/TjqNfo7/wy/wDG7/pI38Zv/BH4K/8Amdrx79oL/gjN8Ov2pvGdt8Qfjx+2J8Ztd1ez0xNOtrvPhm12WySSSLHst9EjU4eaQ5Iz82M4AA+j4VzfDZHnCxVdNxUZL3bX1Stu12PmeLslxWf5K8Jh2lJyi/eulaLbeyffsfi1RX61/wDEOj+yf/0cf8Zv/BloP/yoo/4h0f2T/wDo4/4zf+DLQf8A5UV+nf8AES8i/wCfdT7o/wDyR+Vf8Qtz/wD5+0vvl/8AIH5KUV+nv7QH/Bvl8GvCPwH8beK/gv8AGz4w6x4x0zwjqV34T0ie90ORL7U47WR7WBlXSkLB5lRSAykhsbh1rrv+IdH9k/8A6OP+M3/gy0H/AOVFH/ES8i/591Puj/8AJB/xC3P/APn7S++X/wAgfkpRX61/8Q6P7J//AEcf8Zv/AAZaD/8AKij/AIh0f2T/APo4/wCM3/gy0H/5UUf8RLyL/n3U+6P/AMkH/ELc/wD+ftL75f8AyBs/8G/H/Jmnib/sp17/AOm7Tq+6K+X/ANn3/gmU/wCyz4Mufh98B/25/jNoWkXmpvqNzafZPCd1vuXjjjaTfcaDIwykMYwDj5c4yST3X/DL/wAbv+kjfxm/8Efgr/5na/I88x1LM84r4qkmozldX32W9r9j9kyDAVsryWhhKrTlCNnba929L279j2avGf8Ago3/AMo9Pjx/2RnxR/6abmj/AIZf+N3/AEkb+M3/AII/BX/zO1ifEz9hr4hfF/4b+IPhL4+/4KFfGi80LxRod3pGtWiaV4NiM9pcwvDMgdPDwZCyOw3KQRnIINeUeufQNFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQB5hrH7XHwl0GTxemqHU4x4J+J/h/wFrDCyDBtX1ldFNl5eGy0R/t6wDucbT5vBCAt0nw7+MXhX4m+L/Hngrw9b3qXfw78Vw+Htda6hVUku5NJ07VVaEhiXj+z6nbglgp3rIMYAZvhr9tD9n6HS/wBovx74r+KmiftR6B8OvFvjHQPFk/in4L6pput6R/a2l2ekrBe3OlQ2M2q2kkEuk2ZzHFcxN9kEjMgcxj6P/wCCdehfs3wfDPxZ4+/Zu/ai1f4vW/jHxu+qeKvFuv67bX16uqx6bp9gbScW8EH2Z4rWys1MEkayJnLD5hQB9A0UUUAFFFFABRRRQAUUUUAFFFFABRRRQAVwHwa+OsXxd+IPxW8Bx+GGsD8MPH8Hhh7o3nm/2kZdA0fWPtAXYvlADVhDsy+fIL7vn2r8T/tm+M/2APiV+3V4j/Zm8JfsVP8AF39oM2Vhda9Zap4zXwzp8UT2cX2eR7q7uUa4UQCEE6fbXZUja4DhgPbv+COPgpNJ/Yo0f4t2/wALPA/gu3+LEtv43svDfgZL547K2vdNslhju7i8mke8u1jhRXmVYY9qxosY8su4B9U0UUUAFFFFABRRRQAUUUUAFFFFABRRRQB5D4T/AGy/h34rj8HNB4b1qFvG/wAXvFHw70kPFCRFqOhtr4uZ5j5nywSDw9dmMrucmaAMi5cp1nwI+NGi/HvwRfeONA0i6soLHxn4j8OSQ3hUu0+j61e6RNKNpI2SS2LyJ32OuQDkV8E/Fz9lHQPhP8Xtb8X/ALQX7NP7RVt4PtviNrvizwx41+A/x68Qaxp+k3Go3F/LPfvodrPb3umXEkeoXglWxtrqMfaZ1Em1+fsX9gSP9mA/so+HNS/Y3+IFz4q+Hup3uralpXiO+1y61K41C6utUu7m/lnubxjPJMb6W68zzTvV9ynBXAAPY6KKKACiiigAooooAKKKKACiiigAooooAKKKKACiiigAooooAKKKKACiiigD4a/a11/4g+NP2kfFPgfx3c/tTeIPAGlTWcFt4R+Bvhu00rTZS9jbzSpc6xb3SancsWlYlYpbRFEmwpJtEje9/sIeHvgp4W+DN3o3wI/ZM174O6RHr8zXXh/xL4bi0691G6aGEyahLslla5aQFENxK7SOYSCSEFeOaN8JP24NO/aP/aT+Inwz/aY8K/C3wbf/ABT0+/th44+EM2qLfxx+CvDME2oQ3japZobYSQSQHCMqy2kwMhIZU9k/YY+L03xl+GGta3cftd+BfjVJpviy406Txb8O/DX9mabA0cFuzWYAvbxJ5EaQs0qS7f3oj2ho2yAe00UUUAFFFFABRRRQAUUUUAFFFFABRRRQB8I/8FAP2gdF+IniPxV8M/iJ/wAEzvDPxr+G3gPx1oPgvW9U1rxPZx6pF4m1qHTHsI9Psrm1KiMSaxpMcl39qgeM3LOqsIGJ9u/4Jufsw/Eb9kr4AXPws8fSadY2reIJLvwr4O0fxVquuWfhDSTbW0UOjwX2qMZ7iOJoZXBCQxr52xIlVMt8t/FnS/2CPiJ+2X+0Bp/7U37CPjnx3rei/EvR7e117wH8J/EmswXEC+EvDl3CbqfTUeA3ccsrAA7XSKO344Vj9cfsG6R+zto3wg1K1/Zm+BfjD4faC3iSZ7vRvGvhLVdGuprv7PbhrhINUVZmiMYiQSKPLLRuB8ytQB7ZRRRQAUUUUAFFFFABRRRQAUUUUAFFFFAH53/ELwB4s+OXxX8Xa/8AGf8AYg/aK/aK8LQeNNXs9E0uf4reBYfBccVrqE9sIItIg160S6WIxNERqkM1yrI4lEcm9F+2v2dHDfBfQo0/Z2uvhPHFDLFD8Pbz+y9+jxpNIiJjSriezCuqiVRFKwCygNtcMq/JH7PvwM/aS0HQfGMA/wCClGofDCG8+NHxB1Gy8Ejwt4ZuhY2t14v1e5gkEt3bvMwnimjuBvYkCfHAAA+wfgbpPiTQ/hbpel+LvjTJ8RNRi8/7R4xlsbO2bUMzyFcx2SJAvlqVi+RRny8nLFiQDrKKKKACiiigAooooAKKKKACiiigAooooAKKKKACiiigAooooAKKKKACiiigD4K/b/8A2CNQ8bfHrWf2udZ/aA+HvjPQ9MttOe5+Cv7SV9cv4L0RgqQRz2vkXCwWEk8sYYy3VnelpWbZtGFHr/8AwSJuPFunfsA/D34XeMvBem6XdfD3QbPwmuo6B4psNY0vXhYWkELajZXNlI4MEjh12SrFMkkUitGAFd/Mv2h7fw98LfjR8W/CX7Uf7KHxO+IHgL4n+OfCnjPw1r/w38CXXiSMSaNbaQI9GvreySWeFEvtH+1DzIvs8seoSLvB8wV67/wTp8B+JfDfg34h/EPVfgrP8NdI+IXxLn8Q+Efh9eWcFrc6JpY0vTbBBPbwM0dtNcS2E980KklDe7XxIHAAPoaiiigAooooAKKKKACiiigAooooAKKKKAPnXxZ8A/24PCvxj8deOf2W/jT8FfDOh+Odetda1Kx8XfCfW9Y1Ca+j0qw01pZbmLxHbQ8xafAqrDbwqqIuVeQySydX+x/8bPih8WdJ8ceDfjbYeHV8XfDbx3J4X17UvCAmXS9Uk/s6w1GO5t452eSEeRqMMckTSSGOaKVd7gA18y/tJeFv2RviB8T/AIufEr4w/Aa11TxB4a/aA8C/Dxpv+E/1WxN/a6xb+EojetFDcrGjQJrsmFRArizXcQXZq+h/2I7j4Y+HJvip8APgt8L/AA94V8KfC34lR6Do1v4dTCXoufD2i61NdTHJ3ztcatOrOSS3lhmJYsSAe5UUUUAFFFFABRRRQAUUUUAFFFFABRRRQB+ZXxH0j9m79o3xpa/CT4a/sjfsww/G/wCIHx98aaFrWqeJvg/p+tXPh/RdHu9SafXNRtS8c91c3SW1k6s8kaO+sQOSR9/7C/4JvazY6t+yFoVlY/DrwZ4W/sLxB4j0C60v4eaGumaHJc6brt/p895ZWqkiCC6mtpLpUyxH2jlnOWPjPgj4aftR/tb/ABW+IX7TvwV/aU8KfB1rDxrrfgvTLTQ/hHpWrapqEWi6hLp7zazfXf7+ZZp7V5Y7aFoAlu0B3szFh7r+wF8Tb34s/sraD4m1fwh4d0TULHVtb0PVbPwhbeTpMt7pmr3mm3N1ZJ1W2uJrSS4jBLHZMuWY5YgHslFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQB+Zn7Yf7ZXxC/Z8+Ofxl+C9v4akvrnVP2gPAXjPRdUh+OvgrRxBoVjaeEJtQ09rXVNetLy3aZdL1JBE8CxS/a1YsY5mevpb/gmb8XNa/aDi+Nnx7vNFGl6Z4t+MMdz4e0k+ONE157O0g8K+HrJ0ebRb+9tYS1za3Mnk+dvCyK7KvmAnyL9urwr+3n8V/2ldb8BeA/2EJT8LYYLQSfEnwLe+EZfE/ilmtojLGk2t3cX9kxRMWt932eeZxDvilg+U19A/8ABOz4OeD/AIGfAa68DeC/2M9U+CNv/wAJFcXNzoGu+ILDVr/Wbh4YN+q3N5Z3d39ollIEZeeUzH7OMgII8gHvNFFFABRRRQAUUUUAFFFFABRRRQAUUUUAfll/wUx+Bf7JnxG/bI8UWWg6j4N8efF/VYNPlv8A4f2n7P3gbWbrSEFjBFBLq+ua1ZMLONo40kUXdwZjEyi3hkRUSvqv/gkh+zF41/ZU/Zi1DwR8Qdc+FF1qmreMLrV57T4O+DNM0bStNWS2tYltX/s62tYr64VYQXu2t4XcOibAsaZ9O+I/7C/7Enxj8aXvxI+Lv7HXwr8VeItS8v8AtHXvEnw902+vbry41ij8yeaBnfbGiIu4nCoqjgAV1Xwi+BfwS/Z+8NT+DPgL8HfCvgjR7m+a9udK8IeHrbTbaW5ZERp2it0RWkKRxqXI3ERqM4UYAOqooooAKKKKACiiigAooooAKKKKACiiigD8WviF8R/2NvFH7fep/FP9pP8AbS0jwppnjP4y+LfA/wATvhL4P8d6v4SOm2WjDVraw1rVZdN1KH7bJcnRraN2uIsMurQIpAjTd+tH7L+u/s6eIfgF4Yu/2SdR8O3Xw4t9PNj4Uk8Jsh05be2ka3McJT5cI8TocfxI2ec13tFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAV8mfFL/gqNpfwl/aC+IPwx8Z+EvBmjeHfhlptxqXiC48R/EhrHxNqmnW+kJqdxqOkaIbBl1KzjR2iMq3aHzIZgUGwbvrOvn/48f8E8fh9+1B40F/8AtBfGPx14r8Fpq41OL4U6nJpY0KO6Fs9uGEkdguoFNskhMRvDESxyhHy0AeBaJ/wXd8Aah8OviN4qXwh8Otd1jwf8D9Z+JujaD8OvjhZeIvtFlpqw+fp2py21qBpN7uubYBVW6iYNKY5ZPJcVe/bC/wCCiX7Xnw/+CHxb8LaJ8C9E8AfEzwn4J0Dxb4Xmk8bpqVvNpGoapJZOZi2mMsN9C1u6SW4jnh/eoY7l8Ej2fVv+CeWn+LvgV43/AGa/iV+1r8XfFXgzxp4BvfB66Vrl/o5fRtPuYfJZ7aeHTI5prhY/lWa8e6bkltxZid749fsH/Bn9ovxD4x8SeP8AV/EMc3jb4aW/gnUo9NvooktrOC8nvYbmDMTMtyk85YMxaP8AdoDGfm3AHknx2/b48dfswX/xB8X/ABa/Z1huPGPgT4CW3jHVNC8K/Fi+vtJu0bWL61jsrdbixt4hMVhEjXhtUkPmCEgpEjmp8Rf+Cmv7Tfwqu/ipo3jL9hbSRffBrwPa+O/Gy2HxaEsH/CMXEd68b2sh01Wn1P8A4luoA2bJHBm1OLwh0Lej+P8A/gm98Pvi9ovia0+MPxx+IHifVPF/wzi8D694jvpNJt7uewi1G5v45glpp8NulwHuWj3LCEMcaZQvukbqPil+xN8Kvi3qvxh1fxJ4g8QQS/G34V2fgHxUtjdwKtpptsurrHLaboW8u4I1q6y0nmIfLhwg2vvAPn/9pr/guF8Bf2dPiz4t8E3N78P5dJ+HQ0tvGv8AwkfxbstH8QXC3lpb32dG0aWF5NV8q0uoJH3S2+9i8UXmyIyjqfFH/BS34l+GPiN46Wb9li1f4dfDj41aL8OvE/jRvHoW9afVF0dYLy2077ERNHFNrNusyvcRlUG+MzHdGno2o/sN6VZePr7x98Jf2jPiR8Pn15LD/hMNM8J3OlNa+IZbO1is4bi4F9p9y8E/2WCCBpbR7d2SCMEkopEvif8AYK+EHivwj8QvBmo+JPEqWvxK+Kmk+P8AXZIby3EkGpae+kPDFATAQtux0W13q4dyJJsOu5NgByP/AAT3+O/7Xfxq8V/F21+P3hjwiPDvhn4teIdD8N6tpPiuS4voo7W5jSKye0GlW0fkpGxIuWnklZvlZP46+m68x+DX7L+kfAz4oeNfH3gv4peKn0nxzrVxrWoeBr46fJpVnqk/lfaLy3cWi3qvJ5WTG9y8ILuVjUkEenUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQAUUUUAFFFFABRRRQB//9k=\" />\n",
       "      </center>\n",
       "  </div>\n",
       "</div>\n",
       "</body>\n",
       "</html>\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.display_helper import two_columns_display__classes_distribution\n",
    "\n",
    "two_columns_display__classes_distribution( tweets[['sentiment']] )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "\"{:,}\".format(round(int(3*(df[ df.index == 'NEGATIVE' ]['count'][0] + df[ df.index == 'POSITIVE' ]['count'][0])/2), -4))": "50,000",
     "\"{:,}\".format(tweets.shape[0])": "219,599",
     "\"{:2.2f}%\".format(df[ df.index == 'NEUTRAL' ].percentage[0])": "45.60%"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "As can be observed above, the \"NEUTRAL\" sentiment is over-represented in the French Tweets dataset with a staggering {{\"{:2.2f}%\".format(df[ df.index == 'NEUTRAL' ].percentage[0])}}. If a model predicts every tweet to be \"NEUTRAL\", then that model will have a {{\"{:2.2f}%\".format(df[ df.index == 'NEUTRAL' ].percentage[0])}} accuracy. This gives high of an incentive for any model trained on such a dataset to be biased towards tagging a tweet as \"NEUTRAL\".<br />\n",
    "To avoid that, we will use <span style=\"background-color: #e0e0e0\">&nbsp;downsampling&nbsp;</span> when training our NLP model, i.e. we will not use all the \"NEUTRAL\" tweets but randomly pick some of them only. That way, we will train our NLP model on a dataset that is more evenly balanced between the three classes ; even if not perfectly.<br />\n",
    "One of the adverse direct consequences of downsampling however is that our training dataset will shrink from {{\"{:,}\".format(tweets.shape[0])}} records down to around {{\"{:,}\".format(round(int(3*(df[ df.index == 'NEGATIVE' ]['count'][0] + df[ df.index == 'POSITIVE' ]['count'][0])/2), -4))}}.\n",
    "<br />\n",
    "<br />\n",
    "However still of non-negligible size, this smaller dataset will not allow for the highest predictive power of an NLP model. To circumvent that bereavement, we will resort to <span style=\"background-color: #e0e0e0\">&nbsp;transfer learning&nbsp;</span>. We will first train our model on a different dataset, larger and more generic in nature, and then fine-tune such a model using our French tweet dataset. That way, our model will have gained insight into how to classify french-language text in general and will use that knowledge when starting to learn how to do so on our more specific content.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"transfer\"></a>\n",
    "<h2>2. Transfer Learning <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h2>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"scraping\"></a>\n",
    "<h3>2.1. Web Scraping <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "In order to train a model with the highest predictive power, we turn towards transfer learning from a domain as generic as possible.<br />\n",
    "For that purpose, we scrapped the web using the <span style=\"background-color: #e0e0e0\">&nbsp;BeautifullSoup&nbsp;</span> python library (<a href=\"https://pypi.org/project/beautifulsoup4/\">https://pypi.org/project/beautifulsoup4/</a>), which basically provides convenience methods to scan thru web pages via an HTML/XML parser.\n",
    "<br />\n",
    "<br />\n",
    "We collected review data from the web on the three below broad product families :\n",
    "    <ol style=\"list-style-type: square; margin: 0 5px 0 50px; /* top right bottom left*/\">\n",
    "        <li>bars & restaurants (across french cities)\n",
    "            <ol style=\"list-style-type: disc;\">\n",
    "                <li>casual dining</li>\n",
    "                <li>family style</li>\n",
    "                <li>cafeteria</li>\n",
    "                <li>brasserie and bistro</li>\n",
    "                <li>coffee house</li>\n",
    "                <li>wine bar</li>\n",
    "                <li>etc.</li>\n",
    "            </ol>\n",
    "        </li>\n",
    "        <li>movies (french and international)\n",
    "            <ol style=\"list-style-type: disc;\">\n",
    "                <li>action</li>\n",
    "                <li>anime</li>\n",
    "                <li>adventure</li>\n",
    "                <li>...</li>\n",
    "                <li>historical film</li>\n",
    "                <li>psychological movie</li>\n",
    "                <li>romantic movie.</li>\n",
    "                <li>...</li>\n",
    "            </ol>\n",
    "        </li>\n",
    "        <li>appliances &amp; electronics (french market)\n",
    "            <ol style=\"list-style-type: disc;\">\n",
    "                <li>large appliances (washing machine, fridge, oven, etc)</li>\n",
    "                <li>kitchen and cooking</li>\n",
    "                <li>video-game console</li>\n",
    "                <li>tablet, smartphone, connected object</li>\n",
    "                <li>computers</li>\n",
    "                <li>video &amp; stereo</li>\n",
    "            </ol>\n",
    "        </li>\n",
    "    </ol>\n",
    "Lets load and apply the same pre-processing as before to it&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded the dataset in 1.7199 seconds\n"
     ]
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.tweet_utils import load_review_fr\n",
    "\n",
    "reviews = load_review_fr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "reviews = clean_tweet_fr(reviews, col_name = 'comment')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "<!DOCTYPE html>\n",
       "<html>\n",
       "<head>\n",
       "<style>\n",
       "\n",
       "* {\n",
       "    box-sizing: border-box;\n",
       "}\n",
       "\n",
       ".left_column {\n",
       "    float: left;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "}\n",
       "\n",
       ".right_column {\n",
       "    float:right;\n",
       "    text-align; right;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "    #background: yellow;\n",
       "}\n",
       "\n",
       "/* Clearfix (clear floats) */\n",
       ".row::after {\n",
       "    content: \"\";\n",
       "    clear: both;\n",
       "    display: table;\n",
       "}\n",
       "\n",
       "</style>\n",
       "</head>\n",
       "<body>\n",
       "\n",
       "<h2></h2>\n",
       "\n",
       "<div class=\"row\" style=\"width: 100%;\">\n",
       "  <div class=\"left_column\" style=\"width: 100.00%;\">\n",
       "      <center>\n",
       "          <style  type=\"text/css\" >\n",
       "    #T_7222f7e8_ee92_11ea_913a_309c23ff89f2 th {\n",
       "          text-align: center;\n",
       "    }    #T_7222f7e8_ee92_11ea_913a_309c23ff89f2row0_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_7222f7e8_ee92_11ea_913a_309c23ff89f2row0_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_7222f7e8_ee92_11ea_913a_309c23ff89f2row0_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_7222f7e8_ee92_11ea_913a_309c23ff89f2row1_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_7222f7e8_ee92_11ea_913a_309c23ff89f2row1_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_7222f7e8_ee92_11ea_913a_309c23ff89f2row1_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_7222f7e8_ee92_11ea_913a_309c23ff89f2row2_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_7222f7e8_ee92_11ea_913a_309c23ff89f2row2_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_7222f7e8_ee92_11ea_913a_309c23ff89f2row2_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }</style><table id=\"T_7222f7e8_ee92_11ea_913a_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" >type</th>        <th class=\"col_heading level0 col1\" >count</th>        <th class=\"col_heading level0 col2\" >percentage</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_7222f7e8_ee92_11ea_913a_309c23ff89f2row0_col0\" class=\"data row0 col0\" >appliances_n_electronics</td>\n",
       "                        <td id=\"T_7222f7e8_ee92_11ea_913a_309c23ff89f2row0_col1\" class=\"data row0 col1\" >              37,096</td>\n",
       "                        <td id=\"T_7222f7e8_ee92_11ea_913a_309c23ff89f2row0_col2\" class=\"data row0 col2\" >21.30%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_7222f7e8_ee92_11ea_913a_309c23ff89f2row1_col0\" class=\"data row1 col0\" >bars_n_restaurants</td>\n",
       "                        <td id=\"T_7222f7e8_ee92_11ea_913a_309c23ff89f2row1_col1\" class=\"data row1 col1\" >              48,664</td>\n",
       "                        <td id=\"T_7222f7e8_ee92_11ea_913a_309c23ff89f2row1_col2\" class=\"data row1 col2\" >27.94%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_7222f7e8_ee92_11ea_913a_309c23ff89f2row2_col0\" class=\"data row2 col0\" >movies</td>\n",
       "                        <td id=\"T_7222f7e8_ee92_11ea_913a_309c23ff89f2row2_col1\" class=\"data row2 col1\" >              88,418</td>\n",
       "                        <td id=\"T_7222f7e8_ee92_11ea_913a_309c23ff89f2row2_col2\" class=\"data row2 col2\" >50.76%</td>\n",
       "            </tr>\n",
       "    </tbody></table>\n",
       "      </center>\n",
       "  </div>\n",
       "  <div class=\"right_column\" style=\"width: 0.00%;\">\n",
       "      <center>\n",
       "          \n",
       "      </center>\n",
       "  </div>\n",
       "</div>\n",
       "</body>\n",
       "</html>\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.display_helper import two_columns_display\n",
    "\n",
    "df = reviews[[ 'type', 'rating' ]].groupby('type').count().rename(columns = {'rating':'count'})\n",
    "df[ 'percentage' ] = (df / len( reviews ) * 100)['count']\n",
    "df_html = df.reset_index(level=0).style.hide_index() \\\n",
    "    .set_properties(**{'width':'20em', 'text-align':'center'}) \\\n",
    "    .format({\"count\": \"{:20,.0f}\", \"percentage\": \"{0:.2f}%\"}) \\\n",
    "    .set_table_styles([dict(selector=\"th\", props=[('text-align', 'center')])]).render()\n",
    "two_columns_display(df_html, '', 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Contrarily to the French Tweets dataset which has 3 classes (NEGATIVE/NEUTRAL/POSITIVE), the French Reviews dataset has 5 classes. The commented reviews are indeed here accompanied by a score between 1 and 5&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "<!DOCTYPE html>\n",
       "<html>\n",
       "<head>\n",
       "<style>\n",
       "\n",
       "* {\n",
       "    box-sizing: border-box;\n",
       "}\n",
       "\n",
       ".left_column {\n",
       "    float: left;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "}\n",
       "\n",
       ".right_column {\n",
       "    float:right;\n",
       "    text-align; right;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "    #background: yellow;\n",
       "}\n",
       "\n",
       "/* Clearfix (clear floats) */\n",
       ".row::after {\n",
       "    content: \"\";\n",
       "    clear: both;\n",
       "    display: table;\n",
       "}\n",
       "\n",
       "</style>\n",
       "</head>\n",
       "<body>\n",
       "\n",
       "<h2></h2>\n",
       "\n",
       "<div class=\"row\" style=\"width: 100%;\">\n",
       "  <div class=\"left_column\" style=\"width: 100.00%;\">\n",
       "      <center>\n",
       "          <div style=\"text-align: center;\"><img style=\"display: inline-block;\" src=\"data:image/jpeg;charset=utf-8;base64, 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\" /></div>\n",
       "      </center>\n",
       "  </div>\n",
       "  <div class=\"right_column\" style=\"width: 0.00%;\">\n",
       "      <center>\n",
       "          \n",
       "      </center>\n",
       "  </div>\n",
       "</div>\n",
       "</body>\n",
       "</html>\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib.ticker import FuncFormatter\n",
    "\n",
    "hist1, bins = np.histogram(\n",
    "    reviews[(reviews['type'] == 'bars_n_restaurants')]['rating']\n",
    "    , bins = 5, range = (1, 6))\n",
    "hist2, bins = np.histogram(\n",
    "    reviews[(reviews['type'] == 'movies')]['rating']\n",
    "    , bins = 5, range = (1, 6))\n",
    "hist3, bins = np.histogram(\n",
    "    reviews[(reviews['type'] == 'appliances_n_electronics')]['rating']\n",
    "    , bins = 5, range = (1, 6))\n",
    "\n",
    "normalized_hist = \\\n",
    "    (hist1+hist2+hist3).astype(np.float32) / (hist1.sum()+hist2.sum()+hist3.sum())\n",
    "\n",
    "plt.ioff() ; fig = plt.figure(figsize=(10, 3))\n",
    "width = .7 ; center = list(range(1, 6, 1))\n",
    "bar1 = plt.bar(center, hist1, align='center', width=width)\n",
    "bar2 = plt.bar(center, hist2, align='center', width=width, bottom=hist1)\n",
    "bar3 = plt.bar(center, hist3, align='center', width=width, bottom=hist1+hist2)\n",
    "ax = plt.gca()\n",
    "ax.get_yaxis().set_major_formatter(\n",
    "    FuncFormatter(lambda y, p: str(int(y/1000)) ))\n",
    "\n",
    "plt.ylabel( \"reviews $\\it{(thousands)}$\" )\n",
    "plt.xlabel( \"rating\" )\n",
    "plt.legend((bar1[0], bar2[0], bar3[0])\n",
    "           , ('bars_n_restaurants', 'movies', 'appliances_n_electronics'))\n",
    "\n",
    "for i in range(0, 5):\n",
    "    x = center[i]\n",
    "    y = hist1[i]+hist2[i]+hist3[i]\n",
    "    ax.annotate( \"{0:.0%}\".format( normalized_hist[i] ), (x-.1, y+1000) )\n",
    "\n",
    "ax.set_ylim([0, ax.get_ylim()[1]*1.05]) ; #plt.show()\n",
    "fig_url = os.path.join('tmp', \"reviews_dataset.jpg\")\n",
    "plt.savefig(fig_url, bbox_inches='tight') ; plt.close(fig)\n",
    "\n",
    "# use base64 encoding to circumvent image (external file) caching (webbrowser)\n",
    "with open(fig_url, \"rb\") as image_file:\n",
    "    encoded_string = base64.b64encode(image_file.read())\n",
    "encoded_string = \"data:image/jpeg;charset=utf-8;base64, \" + encoded_string.decode()\n",
    "os.remove(fig_url)\n",
    "\n",
    "two_columns_display(\n",
    "    '<div style=\"text-align: center;\"><img style=\"display: inline-block;\" src=\"' +\n",
    "    encoded_string + '\" /></div>', '', 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "We will treat this Sentiment Classification problem as a regression problem (we treat the \"rating\" variable as a continuous variable). That way, our model will be highly penalized if it classifies a \"5 stars\" product review as a \"1 star\" product review (even more so than if the matter was dealt with as a pure classification problem) since the objective function will bring to minimizing the distance between the prediction and the real rating (as opposed to simply categorizing a rating in the proper class). \n",
    "<br />\n",
    "<br />\n",
    "Note that, however imbalanced to a certain extend, the French Reviews is way less so than the French Tweets dataset.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"tokenizer\"></a>\n",
    "<h3>2.2. Custom Text Tokenizer <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "We will be using a custom tokenizer for our datasets. It uses a <span style=\"background-color: #e0e0e0\">&nbsp;regex pattern&nbsp;</span> which has a few specificities to better take advantage of the raw text comments we are encountering in real life.<br />\n",
    "Most of those specificities have to do with quantities, units, and whether or not the two are space-separated. We deam those specificities important since many times in the French Reviews dataset, prices, volumes, weights and distances do count in the satisfaction one has with regards to a product that is being rated and commented. We want to optimize the chances for our model to grab those relations not less than any other.\n",
    "<br />\n",
    "<br />\n",
    "Lets apply this tokenizer to our French Tweets dataset and look at the tokens distribution accros our records&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "from my_NLP_RNN_fr_lib.tokenizer_utils import tokenize_tweet_fr\n",
    "\n",
    "tokenized_tweets = tokenize_tweet_fr(tweets)\n",
    "#del tweets ; silent = gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.tweet_utils import plt_hist_bin_width\n",
    "\n",
    "tweet_length_data = tokenized_tweets.map( lambda words_list: len( words_list ) )\n",
    "binwidth = 50\n",
    "ax = plt_hist_bin_width( data = tweet_length_data, binwidth = binwidth\n",
    "                         , ylog = True, rotate_text = False )\n",
    "ax.set_ylabel('tweets count\\n($\\it{log scale}$)')\n",
    "ax.set_xlabel('words count')\n",
    "plt.show()\n",
    "\n",
    "max_length = np.arange(min(tweet_length_data), max(tweet_length_data) + binwidth, binwidth)[-1]\n",
    "tweets_count = tweet_length_data.map(\n",
    "    lambda tweet_length: (tweet_length <= max_length) & (tweet_length > max_length-binwidth)\n",
    ").sum()\n",
    "del tokenized_tweets, tweet_length_data ; dummy = gc.collect()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "TWEET_MAX_CHAR_COUNT": "280",
     "max_length": "200",
     "max_length-binwidth": "150",
     "tweets_count": "219886"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "NLP models allow for a fixed number of tokens per input text. This maximum number of words (of tokens) is to be determined from the start.\n",
    "<br />\n",
    "<br />\n",
    "As can be observed on the above chart, with a character limit of {{TWEET_MAX_CHAR_COUNT}} characters for each tweet (imposed by the Twiter platform), {{tweets_count}} still manage to count between {{max_length-binwidth}} and {{max_length}} words. To remain on the safe side and not constrain ourselves too much with regards to maximum allowable length of input text, we then chose to take a limit of no less than {{TWEET_MAX_CHAR_COUNT}} tokens max for the input layer to our sentiment analysis NLP model.\n",
    "<br />\n",
    "<br />\n",
    "Lets then filter out comments that are more than {{TWEET_MAX_CHAR_COUNT}} tokens long in our pre-training French Reviews dataset&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenized_reviews = tokenize_tweet_fr(reviews, col_name='comment')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "reviews_length_data = \\\n",
    "    tokenized_reviews.map( lambda sentence_tokens_list: len( sentence_tokens_list ) )\n",
    "reviews_count = tokenized_reviews.shape[0]\n",
    "reviews.drop(\n",
    "    reviews_length_data[\n",
    "        (reviews_length_data > TWEET_MAX_CHAR_COUNT) | (reviews_length_data == 0)\n",
    "    ].index\n",
    "    , inplace=True )\n",
    "reviews.reset_index(drop=True, inplace=True)\n",
    "dropped_reviews_count = reviews_count - tokenized_reviews.shape[0]\n",
    "del tokenized_reviews, reviews_length_data ; dummy = gc.collect()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "\"{:,}\".format(dropped_reviews_count)": "0",
     "\"{:2.2f}%\".format(dropped_reviews_count/reviews_count*100)": "0.00%"
    }
   },
   "source": [
    "<div style=\"text-align: justify\">\n",
    "{{\"{:,}\".format(dropped_reviews_count)}} entries had to be dropped. They accounted for {{\"{:2.2f}%\".format(dropped_reviews_count/reviews_count*100)}} of the dataset.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# data sanity check ; ensure no \"rating\" is malformed =>\n",
    "assert (not reviews['rating'].isnull().any() and\n",
    "        np.issubdtype(reviews['rating'].dtype, np.number)) \\\n",
    "    , 'we have a \"rating is malformed\" issue with the training data !'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"embeddings\"></a>\n",
    "<h2>3. Word Embeddings <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h2>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"fasttext\"></a>\n",
    "<h3>3.1. fastText <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "We're using word embeddings from fastText. They have been generated by training fastText NLP models on text gethered from <a href=\"https://commoncrawl.org/\">Common Crawl</a> and <a href=\"https://www.wikipedia.org/\">Wikipedia</a>. For more on this, please visit <a href=\"https://fasttext.cc/docs/en/crawl-vectors.html\">https://fasttext.cc/docs/en/crawl-vectors.html</a>\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████████████████████████████████████████████████| 314395/314395 [11:17<00:00, 464.09it/s]\n"
     ]
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.fastText import download_fastText_fr_300_vec ; download_fastText_fr_300_vec()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded the Vocabulary in 138.4681 seconds [1,234,042 word embeddings of 300 features]\n"
     ]
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.fastText import read_fastText_vecs\n",
    "\n",
    "word_to_index, index_to_word, word_to_vec_map = \\\n",
    "    read_fastText_vecs( os.path.join('data', 'fastText_french', 'cc.fr.300.vec') )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "'{:,}'.format(len( word_to_vec_map ))": "1,234,043",
     "word_to_vec_map[ 'bonjour' ].shape": "(300,)",
     "word_to_vec_map[ 'bonjour' ].shape[0]": "300"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Our vocabulary is comprised of {{'{:,}'.format(len( word_to_vec_map ))}} words and the shape of a word vector is \"{{word_to_vec_map[ 'bonjour' ].shape}}\", meaning each token is codified in <u>a {{word_to_vec_map[ 'bonjour' ].shape[0]}} dimensions space</u>.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"ukn\"></a>\n",
    "<h3>3.2. The &lt;UKN&gt; Token <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Below, we calculate the dimensions for the vector associated to the \"&lt;UKN&gt;\" token. This \"&lt;UKN&gt;\" token will serve as a placeholder for any word encountered that does not belong to our vocabulary. For instance, when a word is mis-spelled, it probably doesn't belong to our vocabulary. We then associate it to the \"&lt;UKN&gt;\" (\"unknown\") token prior to feeding it to our model.<br />\n",
    "For our \"&lt;UKN&gt;\" token vector, we take the average of all word embeddings, and we add that vector to our vocab in the last position.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "avg_vec = np.zeros( word_to_vec_map[ 'bonjour' ].shape )\n",
    "\n",
    "for k, v in word_to_vec_map.items() :\n",
    "    avg_vec += v\n",
    "\n",
    "avg_vec /= len( word_to_vec_map )\n",
    "\n",
    "w = \"<UNK>\"\n",
    "word_to_vec_map[ w ] = avg_vec ; del avg_vec\n",
    "word_to_index[ w ] = len( word_to_index )\n",
    "index_to_word[ len( index_to_word ) ] = w"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<hr style=\"height:2px;border-width:0;color:gray;background-color:gray;width:80%\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "max_len = 7 ; max_len": "7"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Lets visualize how the <small><b><em>sentences_to_indices</em></b></small> procedure of the <small><b><em>my_NLP_RNN_fr_lib.fastText</em></b></small> module <a href='.\\my_NLP_RNN_fr_lib\\fastText.py' target='_blank'><img href='.' src='./images/target_blank.png' style='vertical-align: baseline; display: inline;' /></a> works with an example of 4 dummy input texts where we arbitrarily fix a maximum length to {{max_len = 7 ; max_len}} tokens per text for illustration&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                      INPUT SENTENCE\n",
      "0                                     trop drole mdr\n",
      "1                                  on joue au foot ?\n",
      "2                                  le repas est pret\n",
      "3           ceci est un mot inconnu : klugkjygjlyffl\n",
      "4  cette phrase est bien plus longue que la limit...\n",
      "\n",
      "text_vectors =\n",
      " [[1184231.  925728. 1040209.       0.       0.       0.       0.]\n",
      " [1069401. 1011262.  847095.  964350.   61594.       0.       0.]\n",
      " [1022097. 1123299.  950048. 1098621.       0.       0.       0.]\n",
      " [ 878293.  950048. 1188986. 1050709.  998286.   61589. 1234043.]\n",
      " [ 879671. 1086497.  950048.  858923. 1091407. 1028989. 1110576.]]\n"
     ]
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.fastText import sentences_to_indices\n",
    "\n",
    "X1 = np.array([\"trop drole mdr\", \"on joue au foot ?\", \"le repas est pret\"\n",
    "               , \"ceci est un mot inconnu : klugkjygjlyffl\"\n",
    "               , \"cette phrase est bien plus longue que la limite autorisee\"])\n",
    "X1_indices = sentences_to_indices(X1,word_to_index, max_len = max_len)\n",
    "print(pd.DataFrame({\"INPUT SENTENCE\":X1}))\n",
    "print(chr(10) + \"text_vectors =\\n\", X1_indices)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "As illustrated above, each input sentence is \"padded\" so that each sequence is of the same length. As for the last sentence, which is longuer than the limit we used for this example, it has been truncated.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"gru\"></a>\n",
    "<h2>4. Dual bi-GRU NLP Model <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h2>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Compared to <u>L</u>ong <u>S</u>hort <u>T</u>erm <u>M</u>emory networks, the <u>G</u>ated <u>R</u>ecurrent <u>U</u>nit model is simpler. It consists of only the <u>update gate</u> and the <u>reset gate</u>. GRU counts one less gate than LSTM, so there are fewer parameters and faster convergence time during training.\n",
    "</div>\n",
    "</div>\n",
    "<hr style=\"height:0pt; visibility:hidden; margin-bottom: 10px; margin-top:0px;\" />\n",
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2); padding-top:0px'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) have been widely used in the field of text sentiment analysis and have achieved good results. However, although CNN can extract local information between consecutive words of a sentence, it ignores the contextual semantic information between words. Bidirectional GRU can make up for the shortcomings that CNN can't extract contextual semantic information of long text, but it can't extract the local features of the text as well as CNN. Therefore, we consider a multi-channel model that combines the power of the two.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"architecture\"></a>\n",
    "<h3>4.1. Architecture <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.python.util import deprecation\n",
    "deprecation._PRINT_DEPRECATION_WARNINGS = False\n",
    "from keras.utils import plot_model\n",
    "import keras.backend as K\n",
    "\n",
    "from my_NLP_RNN_fr_lib.model.architecture import build_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Built the model in 12.7192 seconds\n"
     ]
    }
   ],
   "source": [
    "model = build_model( (TWEET_MAX_CHAR_COUNT, ), word_to_vec_map, word_to_index ) # default hyperparameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "from my_NLP_RNN_fr_lib.model.plot import plot_model\n",
    "\n",
    "architecture_plot = plot_model(\n",
    "    model\n",
    "    , bgcolor = '#123456', forecolor = '#a18a12'\n",
    "    , framecolor = '#e6c619', watermark_text = \"NLP / sentiment of French text\"\n",
    "    , fillcolors = {'InputLayer': '#654987', 'Embedding': '#995511'\n",
    "                    , 'Bidirectional(GRU)': '#0e0e0e', 'Conv1D': '#444444', 'GlobalMaxPooling1D': '#263340'\n",
    "                    , 'Concatenate': '#654321', 'Dense': '#556699'}\n",
    "    , fontcolors = {'Embedding': '#ccae0c', 'Bidirectional(GRU)': '#ffffff'\n",
    "                    , 'Concatenate': '#000000', 'Dense': '#333333'}\n",
    ")\n",
    "silent = cv2.imwrite(os.path.join('tmp', 'architecture.png'), architecture_plot)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "<a href=\"./tmp/architecture.png\" target=\"_blank\">\n",
    "<table border=0 width=\"300px;\" style=\"float: left;\"><tr><td>\n",
    "<img alt=\"NLP / sentiment of French text\" src=\"./tmp/architecture.png?uncache=1456148618\" />\n",
    "<br /><center><em>click to enlarge</em></center>\n",
    "</td></tr></table></a>\n",
    "For each input (in <span style=\"background-color:#987db9; color: #e6c619;\">&nbsp;<b>purple</b>&nbsp;</span> on the left figure), the sequence of words is branched-out into two bi-directionnal GRU cells (in <span style=\"background-color:#0e0e0e; color: #ffffff;\">&nbsp;<b>black</b>&nbsp;</span> on the left figure). Each generated signal is then fed to two distinct 1D convolutional units for feature extraction (in <span style=\"background-color:#999999; color: #e6c619;\">&nbsp;<b>grey</b>&nbsp;</span> on the left figure). At this stage, we have four distinct signals, which we combine for treatment in a 3-steps densely-connected network (in <span style=\"background-color:#a4b4e2; color: #777777;\">&nbsp;<b>light-blue</b>&nbsp;</span> on the left figure). The remaining of the model is mostly there for regulation (to avoid over-fitting).\n",
    "<br />\n",
    "<br />\n",
    "These different units all have hyperparameters that need tuning for our NLP model to deliver the best predictions it can. For a complete view of which hyperparameter is concerned for each of the units constitutive of our NLP model, please refere to the <small><b><em>build_model</em></b></small> method of the <small><b><em>my_NLP_RNN_fr_lib.model.architecture</em></b></small> module <a href='.\\my_NLP_RNN_fr_lib\\model\\architecture.py' target='_blank'><img href='.' src='./images/target_blank.png' style='vertical-align: baseline; display: inline;' /></a>.\n",
    "<br clear=\"left\" />\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [],
   "source": [
    "K.clear_session()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"optimization\"></a>\n",
    "<h3>4.2. Hyperparameters optimization <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "To optimize the set of hyperparameters values of our NLP model, we first relate to random search and, in a second time, we add <span style=\"background-color: #e0e0e0\">&nbsp;XGBoost&nbsp;</span> + <span style=\"background-color: #e0e0e0\">&nbsp;Scikit-Learn&nbsp;</span> to the mix to get to an even better performing model instance.\n",
    "<br />\n",
    "Complete details on how it has all been done can be found in <a href=\"https://github.com/aurelienmorgan/french_text_sentiment/blob/master/model/optimization/README.md\" target=\"_blank\">this autonomous step-by-step Notebook with full running code <img href='.' src='./images/target_blank.png' style='vertical-align: baseline; display: inline;' /></a>.\n",
    "<br />\n",
    "<br />\n",
    "In short, the whole procedure has taken place <u>on a small subset of the French Reviews training data</u>&nbsp;:\n",
    "<ul style=\"margin: 0 5px 0 50px; /* top right bottom left*/\">\n",
    "    <li>First, we trained a bunch of model instances, each using a different set of randomly-generated hyperparameter values.</li>\n",
    "    <li>Second, with this information, we trained an XGBoost regressor to predict NLP model performance based on the set of hyperparameters values</li>\n",
    "    <li>Third, we feed the XGBoost regressor with tens millions of sets of randomly-generated hyperparameters values and select the ones predicted to bring the best performance for our NLP model architecture.</li>\n",
    "</ul>\n",
    "The first step gave us one NLP model instance which we call the \"<b>best random search</b>\" model instance. The third step gave us one NLP model instance which we call the \"<b>best XGBoost-identified</b>\" model instance.\n",
    "<br />\n",
    "We then fitted few most promissing sets of hyperparameters values on the entire training dataset, and this finally gave us one NLP model instance which we call the \"<b>best refitted</b>\" model instance.\n",
    "<br />\n",
    "For comparison purposes, to highlight the gain in predictive power made at each step of the optimization procedure, we plot the training history (loss and performance metric over training epoch) for the 3 above-mentioned NLP model instances&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1008x576 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.model.training import train_models\n",
    "from my_NLP_RNN_fr_lib.model.plot import plot_trained_models_history\n",
    "\n",
    "csv_full_path = os.path.join('model', 'trained_instances.csv')\n",
    "\n",
    "trained_models_df = pd.read_csv(csv_full_path)\n",
    "\n",
    "batch_size = 512 ; epochs = 120\n",
    "\n",
    "best_random_20th_idx = \\\n",
    "    trained_models_df[\n",
    "        (trained_models_df[ \"batch_size\" ] == batch_size) &\n",
    "        (trained_models_df[ \"epochs\" ] == epochs) &\n",
    "        (trained_models_df['is_random_set'] == 1) &\n",
    "        (trained_models_df['training records'] == 7352) &\n",
    "        (trained_models_df['validation records'] == 1298)][\"val_rmse__last_10_avg\"].idxmin()\n",
    "best_xgboost_20th_idx = \\\n",
    "    trained_models_df[\n",
    "        (trained_models_df[ \"batch_size\" ] == batch_size) &\n",
    "        (trained_models_df[ \"epochs\" ] == epochs) &\n",
    "        (trained_models_df['is_random_set'] == 0) &\n",
    "        (trained_models_df['training records'] == 7352) &\n",
    "        (trained_models_df['validation records'] == 1298)][\"val_rmse__last_10_avg\"].idxmin()\n",
    "best_xgboost_idx = \\\n",
    "    trained_models_df[\n",
    "        (trained_models_df[ \"batch_size\" ] == batch_size) &\n",
    "        (trained_models_df[ \"epochs\" ] == epochs) &\n",
    "        (trained_models_df['is_random_set'] == 0) &\n",
    "        (trained_models_df['training records'] == 147061) &\n",
    "        (trained_models_df['validation records'] == 25952)][\"val_rmse__last_10_avg\"].idxmin()\n",
    "\n",
    "plot_trained_models_history(\n",
    "    trained_models_df.loc[[\n",
    "        best_random_20th_idx, best_xgboost_20th_idx, best_xgboost_idx]].reset_index(drop=True)\n",
    "    , validation_only = True\n",
    "    , labels = [r' / $\\bf{best}$ $\\bf{random}$ $\\bf{search}$'\n",
    "                , r' / $\\bf{best}$ $\\bf{XGBoost-identified}$'\n",
    "                , r' / $\\bf{best}$ $\\bf{refitted}$']\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "These performances have respectively been obtained when using the below-listed sets of hyperparameters values&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <!DOCTYPE html><html><head>\n",
       "        <style>\n",
       "        .rotate4853958390 div {\n",
       "            position:   absolute;\n",
       "            white-space:  nowrap;\n",
       "            left:       5px;                        /* rotated header bottom margin */\n",
       "            top:        185px;   /* height + rotated header left margin */\n",
       "            /* attempt at right-alligning, short on large columns (and/or large 'min_column_width_px') => */\n",
       "            transform: translateY(100%);\n",
       "            #background-color: yellow;\n",
       "        }\n",
       "\n",
       "        .rotate4853958390 {\n",
       "            transform:          rotate(-90deg);\n",
       "            transform-origin:          0% 100%;\n",
       "            height:                      180px;\n",
       "            width: 45px;\n",
       "            #background-color: red;\n",
       "        }\n",
       "        </style></head>\n",
       "        <body><center>\n",
       "        <table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th class=\"rotate4853958390\"><div>model</div></th>\n",
       "      <th class=\"rotate4853958390\"><div><p style=\"color: darkgray;\">training records</p></div></th>\n",
       "      <th class=\"rotate4853958390\"><div><p style=\"color: darkgray;\">validation records</p></div></th>\n",
       "      <th class=\"rotate4853958390\"><div>spatial_dropout_prop</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>recurr_units</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>recurrent_regularizer_l1_factor</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>recurrent_regularizer_l2_factor</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>recurrent_dropout_prop</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>conv_units</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>kernel_size_1</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>kernel_size_2</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>dense_units_1</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>dense_units_2</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>dropout_prop</div></th>\n",
       "      <th class=\"rotate4853958390\"><div>lr</div></th>\n",
       "      <th class=\"rotate4853958390\"><div><p style=\"color: darkgray;\">valiadation RMSE</p></div></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>best random search</td>\n",
       "      <td>7352</td>\n",
       "      <td>1298</td>\n",
       "      <td>0.65</td>\n",
       "      <td>115</td>\n",
       "      <td>0.04136</td>\n",
       "      <td>0.00843</td>\n",
       "      <td>0.00356</td>\n",
       "      <td>25</td>\n",
       "      <td>3</td>\n",
       "      <td>2</td>\n",
       "      <td>49</td>\n",
       "      <td>42</td>\n",
       "      <td>0.22</td>\n",
       "      <td>0.0348</td>\n",
       "      <td>0.650973</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>best XGBoost-identified</td>\n",
       "      <td>7352</td>\n",
       "      <td>1298</td>\n",
       "      <td>0.62</td>\n",
       "      <td>68</td>\n",
       "      <td>0.03626</td>\n",
       "      <td>0.00635</td>\n",
       "      <td>0.00616</td>\n",
       "      <td>28</td>\n",
       "      <td>2</td>\n",
       "      <td>3</td>\n",
       "      <td>46</td>\n",
       "      <td>32</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.0362</td>\n",
       "      <td>0.621018</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>best refitted</td>\n",
       "      <td>147061</td>\n",
       "      <td>25952</td>\n",
       "      <td>0.62</td>\n",
       "      <td>68</td>\n",
       "      <td>0.03626</td>\n",
       "      <td>0.00635</td>\n",
       "      <td>0.00616</td>\n",
       "      <td>28</td>\n",
       "      <td>2</td>\n",
       "      <td>3</td>\n",
       "      <td>46</td>\n",
       "      <td>32</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.0362</td>\n",
       "      <td>0.549117</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table></center></body></html>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.display_helper import format_vertical_headers\n",
    "\n",
    "best_models_df = \\\n",
    "    trained_models_df.loc[\n",
    "        [best_random_20th_idx, best_xgboost_20th_idx, best_xgboost_idx]\n",
    "        , trained_models_df.columns.isin([\n",
    "            'training records', 'validation records', 'spatial_dropout_prop', 'recurr_units',\n",
    "            'recurrent_regularizer_l1_factor', 'recurrent_regularizer_l2_factor',\n",
    "            'recurrent_dropout_prop', 'conv_units', 'kernel_size_1',\n",
    "            'kernel_size_2', 'dense_units_1', 'dense_units_2', 'dropout_prop', 'lr',\n",
    "            'val_rmse__last_10_avg'\n",
    "        ])]\n",
    "best_models_df.index = ['best random search', 'best XGBoost-identified', 'best refitted']\n",
    "best_models_df = \\\n",
    "    best_models_df.reset_index(level=0) \\\n",
    "    .rename(columns = {'index': 'model', 'val_rmse__last_10_avg':'valiadation RMSE'})\n",
    "format_vertical_headers(\n",
    "    best_models_df\n",
    "    , greyed_out_colnames = ['training records', 'validation records', 'valiadation RMSE']\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<hr style=\"height:2px;border-width:0;color:gray;background-color:gray;width:80%\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "However tedious, the process followed to get to the optimized set of hyperparameters values for our NLP model architecture represents a dramatic enhancement in performance compared to the best we could come up with using pure random search.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"training\"></a>\n",
    "<h2>5. NLP Model Training <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h2>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "As explained in the summary figure seen in <a href=\"#page-top\"><b>&sect;0.</b></a>, we follow a 3-Steps process to train our Fench Sentiment NLP model&nbsp;:\n",
    "    <ul style=\"margin: 0 5px 0 50px; /* top right bottom left*/\">\n",
    "        <li>Step 1&nbsp;: pre-training</li>\n",
    "        <li>Step 2&nbsp;: warming-up</li>\n",
    "        <li>Step 3&nbsp;: fine-tuning</li>\n",
    "    </ul>\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"pretraining\"></a>\n",
    "<h3>5.1. Pre-training <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "For the pre-training of our French Sentiment NLP model, we work with the French Reviews dataset (that we web-scrapped, as explained <a href=\"#scraping\"><b>&sect;2.1.</b></a>). We detail here what we basically did to train the \"<b>best-refitted</b>\" model instance mentioned in <a href=\"#optimization\"><b>&sect;4.2.</b></a> above.\n",
    "<br />\n",
    "<br />\n",
    "Below, we start by splitting the French Reviews dataset into 'training' and 'validation' subsets&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 745,
   "metadata": {},
   "outputs": [],
   "source": [
    "reviews_X = \\\n",
    "    sentences_to_indices(reviews['comment'], word_to_index, max_len = TWEET_MAX_CHAR_COUNT)\n",
    "reviews_Y = \\\n",
    "    reviews['rating']\n",
    "#del reviews ; silent = gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 746,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "test_prop = .15\n",
    "X_train, X_valid, y_train, y_valid = \\\n",
    "    train_test_split(reviews_X, reviews_Y,\n",
    "                     test_size=test_prop)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "\"{:,}\".format(X_train.shape[0])": "147,061",
     "\"{:,}\".format(X_valid.shape[0])": "25,952"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Our training sample is made up of {{\"{:,}\".format(X_train.shape[0])}} records and we have {{\"{:,}\".format(X_valid.shape[0])}} additional ones for validation purposes. We use the <small><b><em>keras.optimizers.RMSprop</em></b></small> optimizer., We employ learning rate decay. Our loss function is the MeanSquarredError function. And our performance metric is the RootMeanSquaredError metric.\n",
    "<br />\n",
    "It has all been wrapped-up within the <b>add_trained_model</b> convenience method that is part of the <small><b><em>my_NLP_RNN_fr_lib.model.training</em></b></small> module <a href='.\\my_NLP_RNN_fr_lib\\model\\training.py' target='_blank'><img href='.' src='./images/target_blank.png' style='vertical-align: baseline; display: inline;' /></a>.\n",
    "<br />\n",
    "<br />\n",
    "Information pertaining to each of the NLP model trained instances are stored on-the-fly in <a href=\"./model/trained_instances.csv\" target=\"_blank\">a dedicated global csv file</a>.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "To save on space, saved NLP model instances are not made available here but running the below cell would retrain the best performing one of them and save it locally&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<center><div style=\"background-color: #f0f0f0; width: 700px; text-align: left; font-size: 8.5pt;\">\n",
    "<code>from my_NLP_RNN_fr_lib.model.training import add_trained_model\n",
    "\n",
    "csv_full_path = os.path.join('model', 'trained_instances.csv')\n",
    "save_root = os.path.join('model', 'trained_instances')\n",
    "\n",
    "trained_model_timestamp = add_trained_model(\n",
    "    (X_train, X_valid, y_train, y_valid)\n",
    "    , word_to_vec_map, word_to_index\n",
    "    , **trained_models_df.loc[best_xgboost_idx, trained_models_df.columns.isin([\n",
    "        'spatial_dropout_prop', 'recurr_units',\n",
    "        'recurrent_regularizer_l1_factor', 'recurrent_regularizer_l2_factor',\n",
    "        'recurrent_dropout_prop', 'conv_units', 'kernel_size_1',\n",
    "        'kernel_size_2', 'dense_units_1', 'dense_units_2', 'dropout_prop', 'lr',\n",
    "       'lr_decay_rate', 'lr_decay_step', 'batch_size', 'epochs'\n",
    "    ])]\n",
    "    , local_path = save_root, csv_full_path = csv_full_path\n",
    ")\n",
    "</code></div></center>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Any trained instance of our NLP model can then be reloaded as shown below&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.models import load_model\n",
    "from my_NLP_RNN_fr_lib.model.training import root_mean_squared_error\n",
    "\n",
    "if 'trained_model_timestamp' in locals() :\n",
    "    if not trained_model_timestamp :\n",
    "        trained_model_timestamp = trained_models_df.loc[best_xgboost_idx, 'timestamp']\n",
    "else :\n",
    "    trained_model_timestamp = trained_models_df.loc[best_xgboost_idx, 'timestamp']\n",
    "\n",
    "custom_objects={'root_mean_squared_error': root_mean_squared_error}\n",
    "trained_model = load_model(\n",
    "    os.path.join('model', 'trained_instances', trained_model_timestamp\n",
    "                 , 'model_' + trained_model_timestamp + '_last.h5')\n",
    "    , custom_objects = custom_objects\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "reviews_sample_size = 10 ; reviews_sample_size": "10"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Now that we have an optimum trained instance of our NLP model, we can start looking into some French Reviews ratings predictions. Lets randomly pick {{reviews_sample_size = 10 ; reviews_sample_size}} from our training dataset&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "import random\n",
    "\n",
    "rows = random.sample(list(reviews.index), reviews_sample_size)\n",
    "\n",
    "predicted_ratings = trained_model.predict(reviews_X[rows])\n",
    "#[(predicted_rating[0], int(round(predicted_rating[0]))) for predicted_rating in predicted_ratings]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 185,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "<!DOCTYPE html>\n",
       "<html>\n",
       "<head>\n",
       "<style>\n",
       "\n",
       "* {\n",
       "    box-sizing: border-box;\n",
       "}\n",
       "\n",
       ".left_column {\n",
       "    float: left;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "}\n",
       "\n",
       ".right_column {\n",
       "    float:right;\n",
       "    text-align; right;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "    #background: yellow;\n",
       "}\n",
       "\n",
       "/* Clearfix (clear floats) */\n",
       ".row::after {\n",
       "    content: \"\";\n",
       "    clear: both;\n",
       "    display: table;\n",
       "}\n",
       "\n",
       "</style>\n",
       "</head>\n",
       "<body>\n",
       "\n",
       "<h2></h2>\n",
       "\n",
       "<div class=\"row\" style=\"width: 100%;\">\n",
       "  <div class=\"left_column\" style=\"width: 55.00%;\">\n",
       "      <center>\n",
       "          <style  type=\"text/css\" >\n",
       "    #T_03493992_e9ed_11ea_8459_309c23ff89f2 th {\n",
       "          text-align: center;\n",
       "    }</style><table id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" >rating</th>        <th class=\"col_heading level0 col1\" >review</th>        <th class=\"col_heading level0 col2\" >predicted rating (rounded)</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row0_col0\" class=\"data row0 col0\" >3</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row0_col1\" class=\"data row0 col1\" >Un restaurant de sp&eacute;cialit&eacute;s s&eacute;n&eacute;galaise sans pr&eacute;tention mais c'est bon et correct au niveau tarif.</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row0_col2\" class=\"data row0 col2\" >2</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row1_col0\" class=\"data row1 col0\" >5</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row1_col1\" class=\"data row1 col1\" >Tout &eacute;tait excellent! Les d&eacute;cors, les acteurs, l'histoire. En bref, c'est un tr&egrave;s bon film. Ce qui ne m'a pas plu: Rien du tout.</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row1_col2\" class=\"data row1 col2\" >4</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row2_col0\" class=\"data row2 col0\" >2</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row2_col1\" class=\"data row2 col1\" >l'innocence, la magie, la nostalgie, la fausse naivet&eacute; Ce qui ne m'a pas plu: un peu cucul par moment et d&eacute;mod&eacute;</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row2_col2\" class=\"data row2 col2\" >4</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row3_col0\" class=\"data row3 col0\" >3</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row3_col1\" class=\"data row3 col1\" >la transposition &agrave; l'&eacute;cran d'un fait divers qui tend &agrave; se banaliser Ce qui ne m'a pas plu: quelques longueurs et certains roles qui ne sonnent pas justes</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row3_col2\" class=\"data row3 col2\" >3</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row4_col0\" class=\"data row4 col0\" >4</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row4_col1\" class=\"data row4 col1\" >tr&egrave;s romantique</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row4_col2\" class=\"data row4 col2\" >3</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row5_col0\" class=\"data row5 col0\" >4</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row5_col1\" class=\"data row5 col1\" >Joli cadre moderne avec une terrasse dans le m&ecirc;me ton et une qualit&eacute; de galette surprenante.</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row5_col2\" class=\"data row5 col2\" >3</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row6_col0\" class=\"data row6 col0\" >5</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row6_col1\" class=\"data row6 col1\" >Couteau au top-Coupe bien, se nettoie bien, tr&egrave;s jolie sur le plan de travail</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row6_col2\" class=\"data row6 col2\" >3</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row7_col0\" class=\"data row7 col0\" >4</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row7_col1\" class=\"data row7 col1\" >Pas d&eacute;&ccedil;ue, le livre est bien respect&eacute;. je doutais du choix des acteur et au final ils sont parfaits. j'ai pass&eacute; un agr&eacute;able moment.</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row7_col2\" class=\"data row7 col2\" >4</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row8_col0\" class=\"data row8 col0\" >4</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row8_col1\" class=\"data row8 col1\" >j'ai tout appr&eacute;ci&eacute; dans ce film, la mise en sc&egrave;ne, lesmagnifiques paysages d' Egypte tout comme le Jardin des Plantes, ainsi que le jeux des acteurs souvent difficiles &agrave; reconnaitre (Mr Almaric). Avec une speciale dedicace &agrave; Louise Bourgoin qui excelle d'intelligence dans son role et qui illumine la pellicule. Merci Mr Besson ! Ce qui ne m'a pas plu: Tout m'a plu!</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row8_col2\" class=\"data row8 col2\" >4</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row9_col0\" class=\"data row9 col0\" >5</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row9_col1\" class=\"data row9 col1\" >Tr&egrave;s beau film et admirablement interpr&eacute;t&eacute;, mais surtout n'oubliez pas votre mouchoir</td>\n",
       "                        <td id=\"T_03493992_e9ed_11ea_8459_309c23ff89f2row9_col2\" class=\"data row9 col2\" >4</td>\n",
       "            </tr>\n",
       "    </tbody></table>\n",
       "      </center>\n",
       "  </div>\n",
       "  <div class=\"right_column\" style=\"width: 45.00%;\">\n",
       "      <center>\n",
       "          <img src=\"data:image/jpeg;charset=utf-8;base64, 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\" />\n",
       "      </center>\n",
       "  </div>\n",
       "</div>\n",
       "</body>\n",
       "</html>\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# table\n",
    "sampled_results = \\\n",
    "    pd.concat((pd.DataFrame(reviews.loc[rows, ['rating', 'comment']].values)\n",
    "              , pd.DataFrame({2: [int(round(pred[0])) for pred in predicted_ratings]}))\n",
    "             , axis = 1)\n",
    "sampled_results = \\\n",
    "    sampled_results.apply(lambda x: (x[0], html_escape(x[1]), x[2])\n",
    "                          , axis=1, result_type='expand') \\\n",
    "    .rename(columns = {0:'rating', 1:'review', 2: 'predicted rating (rounded)'})\n",
    "\n",
    "# plot\n",
    "plt.ioff() ; fig = plt.figure(figsize=(7, 8))\n",
    "ax = plt.gca()\n",
    "x_index = np.arange(reviews_sample_size)\n",
    "bar_width = 0.35\n",
    "ax.bar(x_index, reviews_Y[rows]\n",
    "       , bar_width, label=\"actual (training dataset sample)\", color='black', alpha=.4)\n",
    "ax.bar(x_index + bar_width, predicted_ratings.reshape((reviews_sample_size,))\n",
    "       , bar_width, label=\"predicted\", color='y', alpha=.4)\n",
    "ax.set_title(\"French Reviews ratings\")\n",
    "ax.legend()\n",
    "ax.get_xaxis().set_visible(False)\n",
    "#plt.show()\n",
    "\n",
    "if not os.path.isdir(os.path.join('tmp')) : os.makedirs('tmp')\n",
    "title = \"reviews_predictions\"\n",
    "fig_url = os.path.join('tmp', title+\".jpg\")\n",
    "plt.savefig(fig_url, bbox_inches='tight') ; plt.close(fig)\n",
    "\n",
    "# use base64 encoding to circumvent image (external file) caching (webbrowser)\n",
    "with open(fig_url, \"rb\") as image_file:\n",
    "    encoded_string = base64.b64encode(image_file.read())\n",
    "encoded_string = \"data:image/jpeg;charset=utf-8;base64, \" + encoded_string.decode()\n",
    "os.remove(fig_url)\n",
    "\n",
    "two_columns_display(sampled_results.style.hide_index().set_table_styles(\n",
    "            [dict(selector=\"th\", props=[('text-align','center')])]\n",
    "        ).render(), '<img src=\"' + encoded_string + '\" />', .55)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<hr style=\"height:2px;border-width:0;color:gray;background-color:gray;width:80%\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "That's all for the pre-training of our NLP model. We're now in a position to transfer that knowledge our model has acquired from the rather generic domain that our web-scrapped French Reviews dataset consitutes and use it as a learning starting point for our (smaller ; soccer-specific) target French Tweets dataset.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"warming\"></a>\n",
    "<h3>5.2. Warming-Up <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"subsetting\"></a>\n",
    "<h4>5.2.a. Training / Validation / Holdout data subsetting <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h4>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "In the pre-training phase (<a href=\"#pretraining\"><b>&sect;5.1.</b></a> above), transfer Learning took place on the French Reviews dataset. The warming-up phase here, on the other hand, takes place on our target French Tweets dataset. As seen in <a href=\"#imbalance\"><b>&sect;1.3.</b></a> , this dataset is made up of three sentiment classes which are \"NEGATIVE\", \"NEUTRAL\" and \"POSITIVE\".\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 438,
   "metadata": {},
   "outputs": [],
   "source": [
    "tweets_X = \\\n",
    "    sentences_to_indices(tweets['tweet_text'], word_to_index, max_len = TWEET_MAX_CHAR_COUNT)\n",
    "tweets_Y = \\\n",
    "    tweets['sentiment']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "If we recall from <a href=\"#imbalance\"><b>&sect;1.3.</b></a> above, the French Tweets dataset being highly imbalanced, it needs <b>downsampling</b> to address the fact that the \"NEUTRAL\" sentiment class is way over-represented. But prior to doing that, in order to evaluate fairly the performance of our model, we need splitting our real-life dataset into <u>training</u> / <u>validation</u> / <u>holdout</u> subsets.\n",
    "<br />\n",
    "<br />\n",
    "This operation has to take place now, since we indeed want to respect the elementary rules of model training which are&nbsp;:\n",
    "    <ul style=\"list-style-type: disc; margin: 0px 200px 0px 50px; /* top right bottom left*/\">\n",
    "        <li>we measure the final performance of a model on&nbsp;:\n",
    "            <ol style=\"list-style-type: circle; padding-bottom: 0;\">\n",
    "                <li>data it had never been presented before (the holdout subset),</li>\n",
    "                <li>data drawn from the real life distribution it will have to predict from when deployed into production (no alteration of the resampling kind),</li>\n",
    "            </ol>\n",
    "        <li>the validation and holdout datasets MUST be drawn from identical distributions,</li>\n",
    "        <li>it is acceptable to <b>train</b> a model on manipulated data (using techinques such as oversampling/undersampling or data augmentation)</li>\n",
    "    </ul>\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 439,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "test_prop = .15 ; holdout_size = 3000\n",
    "tweets_X_train, tweets_X_valid, tweets_y_train, tweets_y_valid = \\\n",
    "    train_test_split(tweets_X, tweets_Y,\n",
    "                     test_size=test_prop)\n",
    "tweets_X_valid, tweets_X_holdout, tweets_y_valid, tweets_y_holdout = \\\n",
    "    train_test_split(tweets_X_valid, tweets_y_valid,\n",
    "                     test_size=holdout_size)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "\"{:,}\".format(holdout_size)": "3,000",
     "\"{:,}\".format(len(tweets_y_train))": "186,659",
     "\"{:,}\".format(len(tweets_y_valid))": "29,940",
     "\"{:g}%\".format(100*(1-test_prop))": "85%",
     "\"{:g}%\".format(100*test_prop)": "15%"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "We thus take {{\"{:g}%\".format(100*(1-test_prop))}} of our French Tweets and will downsample these {{\"{:,}\".format(len(tweets_y_train))}} records and will use the result of that for training. Save the {{\"{:,}\".format(holdout_size)}} records which constitute our holdout data, the remaing {{\"{:g}%\".format(100*test_prop)}} are to be used for validation (making it {{\"{:,}\".format(len(tweets_y_valid))}} validation records).\n",
    "<br />\n",
    "<br />\n",
    "It is on the training dataset (and on the training dataset only) that we operate resampling. We employ <b>downsampling</b> on the French Tweets dataset to address the fact that the \"NEUTRAL\" sentiment class is way over-represented there. This results in the below classes distribution for our resized training dataset&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 490,
   "metadata": {},
   "outputs": [],
   "source": [
    "from imblearn.under_sampling import RandomUnderSampler\n",
    "\n",
    "rus = RandomUnderSampler( sampling_strategy={'NEUTRAL':tweets_y_train[tweets_y_train == 'POSITIVE'].count()})\n",
    "tweets_X_train_resampled, tweets_y_train_resampled = rus.fit_sample(tweets_X_train, tweets_y_train )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 492,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "<!DOCTYPE html>\n",
       "<html>\n",
       "<head>\n",
       "<style>\n",
       "\n",
       "* {\n",
       "    box-sizing: border-box;\n",
       "}\n",
       "\n",
       ".left_column {\n",
       "    float: left;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "}\n",
       "\n",
       ".right_column {\n",
       "    float:right;\n",
       "    text-align; right;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "    #background: yellow;\n",
       "}\n",
       "\n",
       "/* Clearfix (clear floats) */\n",
       ".row::after {\n",
       "    content: \"\";\n",
       "    clear: both;\n",
       "    display: table;\n",
       "}\n",
       "\n",
       "</style>\n",
       "</head>\n",
       "<body>\n",
       "\n",
       "<h2></h2>\n",
       "\n",
       "<div class=\"row\" style=\"width: 100%;\">\n",
       "  <div class=\"left_column\" style=\"width: 50.00%;\">\n",
       "      <center>\n",
       "          <style  type=\"text/css\" >\n",
       "    #T_fb49386e_f045_11ea_b73b_309c23ff89f2 th {\n",
       "          text-align: center;\n",
       "    }    #T_fb49386e_f045_11ea_b73b_309c23ff89f2row0_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_fb49386e_f045_11ea_b73b_309c23ff89f2row0_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_fb49386e_f045_11ea_b73b_309c23ff89f2row0_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_fb49386e_f045_11ea_b73b_309c23ff89f2row1_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_fb49386e_f045_11ea_b73b_309c23ff89f2row1_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_fb49386e_f045_11ea_b73b_309c23ff89f2row1_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_fb49386e_f045_11ea_b73b_309c23ff89f2row2_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_fb49386e_f045_11ea_b73b_309c23ff89f2row2_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_fb49386e_f045_11ea_b73b_309c23ff89f2row2_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }</style><table id=\"T_fb49386e_f045_11ea_b73b_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" >sentiment</th>        <th class=\"col_heading level0 col1\" >count</th>        <th class=\"col_heading level0 col2\" >percentage</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_fb49386e_f045_11ea_b73b_309c23ff89f2row0_col0\" class=\"data row0 col0\" >NEGATIVE</td>\n",
       "                        <td id=\"T_fb49386e_f045_11ea_b73b_309c23ff89f2row0_col1\" class=\"data row0 col1\" >5,164</td>\n",
       "                        <td id=\"T_fb49386e_f045_11ea_b73b_309c23ff89f2row0_col2\" class=\"data row0 col2\" >8.80%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_fb49386e_f045_11ea_b73b_309c23ff89f2row1_col0\" class=\"data row1 col0\" >NEUTRAL</td>\n",
       "                        <td id=\"T_fb49386e_f045_11ea_b73b_309c23ff89f2row1_col1\" class=\"data row1 col1\" >26,758</td>\n",
       "                        <td id=\"T_fb49386e_f045_11ea_b73b_309c23ff89f2row1_col2\" class=\"data row1 col2\" >45.60%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_fb49386e_f045_11ea_b73b_309c23ff89f2row2_col0\" class=\"data row2 col0\" >POSITIVE</td>\n",
       "                        <td id=\"T_fb49386e_f045_11ea_b73b_309c23ff89f2row2_col1\" class=\"data row2 col1\" >26,758</td>\n",
       "                        <td id=\"T_fb49386e_f045_11ea_b73b_309c23ff89f2row2_col2\" class=\"data row2 col2\" >45.60%</td>\n",
       "            </tr>\n",
       "    </tbody></table>\n",
       "      </center>\n",
       "  </div>\n",
       "  <div class=\"right_column\" style=\"width: 50.00%;\">\n",
       "      <center>\n",
       "          <img src=\"data:image/jpeg;charset=utf-8;base64, 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\" />\n",
       "      </center>\n",
       "  </div>\n",
       "</div>\n",
       "</body>\n",
       "</html>\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "two_columns_display__classes_distribution( tweets_y_train_resampled )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "The resulting dataset is still too widely biased against NEGATIVE tweets, though. If we were to train a model on it, it would neglect to learn to detect NEGATIVE instances altogether. To adress that, we now choose to use <b>upsampling</b>. There are indeed too few NEGATIVE records for us to use downsampling again. The resulting training dataset would become way too small for any learning to happen otherwise.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 493,
   "metadata": {},
   "outputs": [],
   "source": [
    "from imblearn.over_sampling import RandomOverSampler\n",
    "\n",
    "ros = RandomOverSampler( 'not majority' )\n",
    "tweets_X_train_resampled, tweets_y_train_resampled = ros.fit_sample(tweets_X_train_resampled, tweets_y_train_resampled)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Following the principles listed above, we end up with the below 3 data subsets&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 494,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m\u001b[4mTraining\u001b[0m\u001b[1m dataset\u001b[0m\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "<!DOCTYPE html>\n",
       "<html>\n",
       "<head>\n",
       "<style>\n",
       "\n",
       "* {\n",
       "    box-sizing: border-box;\n",
       "}\n",
       "\n",
       ".left_column {\n",
       "    float: left;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "}\n",
       "\n",
       ".right_column {\n",
       "    float:right;\n",
       "    text-align; right;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "    #background: yellow;\n",
       "}\n",
       "\n",
       "/* Clearfix (clear floats) */\n",
       ".row::after {\n",
       "    content: \"\";\n",
       "    clear: both;\n",
       "    display: table;\n",
       "}\n",
       "\n",
       "</style>\n",
       "</head>\n",
       "<body>\n",
       "\n",
       "<h2></h2>\n",
       "\n",
       "<div class=\"row\" style=\"width: 100%;\">\n",
       "  <div class=\"left_column\" style=\"width: 50.00%;\">\n",
       "      <center>\n",
       "          <style  type=\"text/css\" >\n",
       "    #T_0b7949ae_f046_11ea_a721_309c23ff89f2 th {\n",
       "          text-align: center;\n",
       "    }    #T_0b7949ae_f046_11ea_a721_309c23ff89f2row0_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b7949ae_f046_11ea_a721_309c23ff89f2row0_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b7949ae_f046_11ea_a721_309c23ff89f2row0_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b7949ae_f046_11ea_a721_309c23ff89f2row1_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b7949ae_f046_11ea_a721_309c23ff89f2row1_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b7949ae_f046_11ea_a721_309c23ff89f2row1_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b7949ae_f046_11ea_a721_309c23ff89f2row2_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b7949ae_f046_11ea_a721_309c23ff89f2row2_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b7949ae_f046_11ea_a721_309c23ff89f2row2_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }</style><table id=\"T_0b7949ae_f046_11ea_a721_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" >sentiment</th>        <th class=\"col_heading level0 col1\" >count</th>        <th class=\"col_heading level0 col2\" >percentage</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_0b7949ae_f046_11ea_a721_309c23ff89f2row0_col0\" class=\"data row0 col0\" >NEGATIVE</td>\n",
       "                        <td id=\"T_0b7949ae_f046_11ea_a721_309c23ff89f2row0_col1\" class=\"data row0 col1\" >26,758</td>\n",
       "                        <td id=\"T_0b7949ae_f046_11ea_a721_309c23ff89f2row0_col2\" class=\"data row0 col2\" >33.33%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_0b7949ae_f046_11ea_a721_309c23ff89f2row1_col0\" class=\"data row1 col0\" >NEUTRAL</td>\n",
       "                        <td id=\"T_0b7949ae_f046_11ea_a721_309c23ff89f2row1_col1\" class=\"data row1 col1\" >26,758</td>\n",
       "                        <td id=\"T_0b7949ae_f046_11ea_a721_309c23ff89f2row1_col2\" class=\"data row1 col2\" >33.33%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_0b7949ae_f046_11ea_a721_309c23ff89f2row2_col0\" class=\"data row2 col0\" >POSITIVE</td>\n",
       "                        <td id=\"T_0b7949ae_f046_11ea_a721_309c23ff89f2row2_col1\" class=\"data row2 col1\" >26,758</td>\n",
       "                        <td id=\"T_0b7949ae_f046_11ea_a721_309c23ff89f2row2_col2\" class=\"data row2 col2\" >33.33%</td>\n",
       "            </tr>\n",
       "    </tbody></table>\n",
       "      </center>\n",
       "  </div>\n",
       "  <div class=\"right_column\" style=\"width: 50.00%;\">\n",
       "      <center>\n",
       "          <img src=\"data:image/jpeg;charset=utf-8;base64, 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\" />\n",
       "      </center>\n",
       "  </div>\n",
       "</div>\n",
       "</body>\n",
       "</html>\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m\u001b[4mValidation\u001b[0m\u001b[1m dataset\u001b[0m\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "<!DOCTYPE html>\n",
       "<html>\n",
       "<head>\n",
       "<style>\n",
       "\n",
       "* {\n",
       "    box-sizing: border-box;\n",
       "}\n",
       "\n",
       ".left_column {\n",
       "    float: left;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "}\n",
       "\n",
       ".right_column {\n",
       "    float:right;\n",
       "    text-align; right;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "    #background: yellow;\n",
       "}\n",
       "\n",
       "/* Clearfix (clear floats) */\n",
       ".row::after {\n",
       "    content: \"\";\n",
       "    clear: both;\n",
       "    display: table;\n",
       "}\n",
       "\n",
       "</style>\n",
       "</head>\n",
       "<body>\n",
       "\n",
       "<h2></h2>\n",
       "\n",
       "<div class=\"row\" style=\"width: 100%;\">\n",
       "  <div class=\"left_column\" style=\"width: 50.00%;\">\n",
       "      <center>\n",
       "          <style  type=\"text/css\" >\n",
       "    #T_0b88cf76_f046_11ea_a922_309c23ff89f2 th {\n",
       "          text-align: center;\n",
       "    }    #T_0b88cf76_f046_11ea_a922_309c23ff89f2row0_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b88cf76_f046_11ea_a922_309c23ff89f2row0_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b88cf76_f046_11ea_a922_309c23ff89f2row0_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b88cf76_f046_11ea_a922_309c23ff89f2row1_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b88cf76_f046_11ea_a922_309c23ff89f2row1_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b88cf76_f046_11ea_a922_309c23ff89f2row1_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b88cf76_f046_11ea_a922_309c23ff89f2row2_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b88cf76_f046_11ea_a922_309c23ff89f2row2_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b88cf76_f046_11ea_a922_309c23ff89f2row2_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }</style><table id=\"T_0b88cf76_f046_11ea_a922_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" >sentiment</th>        <th class=\"col_heading level0 col1\" >count</th>        <th class=\"col_heading level0 col2\" >percentage</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_0b88cf76_f046_11ea_a922_309c23ff89f2row0_col0\" class=\"data row0 col0\" >NEGATIVE</td>\n",
       "                        <td id=\"T_0b88cf76_f046_11ea_a922_309c23ff89f2row0_col1\" class=\"data row0 col1\" >843</td>\n",
       "                        <td id=\"T_0b88cf76_f046_11ea_a922_309c23ff89f2row0_col2\" class=\"data row0 col2\" >2.82%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_0b88cf76_f046_11ea_a922_309c23ff89f2row1_col0\" class=\"data row1 col0\" >NEUTRAL</td>\n",
       "                        <td id=\"T_0b88cf76_f046_11ea_a922_309c23ff89f2row1_col1\" class=\"data row1 col1\" >24,821</td>\n",
       "                        <td id=\"T_0b88cf76_f046_11ea_a922_309c23ff89f2row1_col2\" class=\"data row1 col2\" >82.90%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_0b88cf76_f046_11ea_a922_309c23ff89f2row2_col0\" class=\"data row2 col0\" >POSITIVE</td>\n",
       "                        <td id=\"T_0b88cf76_f046_11ea_a922_309c23ff89f2row2_col1\" class=\"data row2 col1\" >4,276</td>\n",
       "                        <td id=\"T_0b88cf76_f046_11ea_a922_309c23ff89f2row2_col2\" class=\"data row2 col2\" >14.28%</td>\n",
       "            </tr>\n",
       "    </tbody></table>\n",
       "      </center>\n",
       "  </div>\n",
       "  <div class=\"right_column\" style=\"width: 50.00%;\">\n",
       "      <center>\n",
       "          <img src=\"data:image/jpeg;charset=utf-8;base64, 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\" />\n",
       "      </center>\n",
       "  </div>\n",
       "</div>\n",
       "</body>\n",
       "</html>\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m\u001b[4mHoldout\u001b[0m\u001b[1m dataset\u001b[0m\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "<!DOCTYPE html>\n",
       "<html>\n",
       "<head>\n",
       "<style>\n",
       "\n",
       "* {\n",
       "    box-sizing: border-box;\n",
       "}\n",
       "\n",
       ".left_column {\n",
       "    float: left;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "}\n",
       "\n",
       ".right_column {\n",
       "    float:right;\n",
       "    text-align; right;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "    #background: yellow;\n",
       "}\n",
       "\n",
       "/* Clearfix (clear floats) */\n",
       ".row::after {\n",
       "    content: \"\";\n",
       "    clear: both;\n",
       "    display: table;\n",
       "}\n",
       "\n",
       "</style>\n",
       "</head>\n",
       "<body>\n",
       "\n",
       "<h2></h2>\n",
       "\n",
       "<div class=\"row\" style=\"width: 100%;\">\n",
       "  <div class=\"left_column\" style=\"width: 50.00%;\">\n",
       "      <center>\n",
       "          <style  type=\"text/css\" >\n",
       "    #T_0b98ac5c_f046_11ea_8de0_309c23ff89f2 th {\n",
       "          text-align: center;\n",
       "    }    #T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row0_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row0_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row0_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row1_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row1_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row1_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row2_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row2_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row2_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }</style><table id=\"T_0b98ac5c_f046_11ea_8de0_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" >sentiment</th>        <th class=\"col_heading level0 col1\" >count</th>        <th class=\"col_heading level0 col2\" >percentage</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row0_col0\" class=\"data row0 col0\" >NEGATIVE</td>\n",
       "                        <td id=\"T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row0_col1\" class=\"data row0 col1\" >62</td>\n",
       "                        <td id=\"T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row0_col2\" class=\"data row0 col2\" >2.07%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row1_col0\" class=\"data row1 col0\" >NEUTRAL</td>\n",
       "                        <td id=\"T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row1_col1\" class=\"data row1 col1\" >2,491</td>\n",
       "                        <td id=\"T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row1_col2\" class=\"data row1 col2\" >83.03%</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row2_col0\" class=\"data row2 col0\" >POSITIVE</td>\n",
       "                        <td id=\"T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row2_col1\" class=\"data row2 col1\" >447</td>\n",
       "                        <td id=\"T_0b98ac5c_f046_11ea_8de0_309c23ff89f2row2_col2\" class=\"data row2 col2\" >14.90%</td>\n",
       "            </tr>\n",
       "    </tbody></table>\n",
       "      </center>\n",
       "  </div>\n",
       "  <div class=\"right_column\" style=\"width: 50.00%;\">\n",
       "      <center>\n",
       "          <img src=\"data:image/jpeg;charset=utf-8;base64, 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\" />\n",
       "      </center>\n",
       "  </div>\n",
       "</div>\n",
       "</body>\n",
       "</html>\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('\\033[1m' + '\\033[4m' + 'Training' + '\\033[0m' + '\\033[1m' + ' dataset' + '\\033[0m')\n",
    "two_columns_display__classes_distribution( tweets_y_train_resampled )\n",
    "print('\\033[1m' + '\\033[4m' + 'Validation' + '\\033[0m' + '\\033[1m' + ' dataset' + '\\033[0m')\n",
    "two_columns_display__classes_distribution( tweets_y_valid )\n",
    "print('\\033[1m' + '\\033[4m' + 'Holdout' + '\\033[0m' + '\\033[1m' + ' dataset' + '\\033[0m')\n",
    "two_columns_display__classes_distribution( tweets_y_holdout )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<hr style=\"height:2px;border-width:0;color:gray;background-color:gray;width:80%\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"training\"></a>\n",
    "<h4>5.2.b. model training <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h4>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "tweets_y_train_resampled_1_hot = \\\n",
    "    pd.get_dummies( tweets_y_train_resampled ) #One-Hot encoded\n",
    "tweets_y_valid_1_hot = \\\n",
    "    pd.get_dummies( tweets_y_valid ) #One-Hot encoded\n",
    "tweets_y_holdout_1_hot = \\\n",
    "    pd.get_dummies( tweets_y_holdout ) #One-Hot encoded"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "During the pre-training phase, we had 5 (ordered) classes. Thus, as introduced in <a href=\"#scraping\"><b>&sect;2.1.</b></a>, we legitimately handeled the Sentiment Analysis problem as a regression problem. It translated in having <u>a relu-activated 1 Node Dense output layer</u>.\n",
    "<br />\n",
    "For the remaining of the training process of our NLP model, the target dataset is made-up of 3 classes where we can't really consider any hierarchy. We then now use <u>a softmax-activated 3 Nodes Dense output layer</u>&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.models import Model\n",
    "from keras.layers import Dense\n",
    "\n",
    "o = Dense(3, activation = \"softmax\", name = \"tweets_sentiment\")(trained_model.layers[-2].output) # ignore original last layer\n",
    "\n",
    "tweets_model = Model( inputs = trained_model.input, outputs = [o])\n",
    "#tweets_model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "TWEET_MAX_CHAR_COUNT": "280"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Remember that we are using the architecture depicted in <a href=\"#architecture\"><b>&sect;4.1.</b></a> above. With our optimum set of hyperparameter values, for each input sentence (of maximum length {{TWEET_MAX_CHAR_COUNT}}), the inner tensors of our NLP model are of the shapes listed in the below table. Note that, for the warming-up of our model with its new output, we freeze all layers except the (new) very last one. This represents the \"<b>Step 2</b>\" in the summary figure seen in <a href=\"#page-top\"><b>&sect;0.</b></a> &nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <!DOCTYPE html><html><head>\n",
       "        <style>\n",
       "        div.centered_table_container {\n",
       "            text-align: center;\n",
       "        }\n",
       "        table {\n",
       "            display: inline-block;\n",
       "        }\n",
       "        </style></head>\n",
       "        <body>\n",
       "        <div class=\"centered_table_container\"><style  type=\"text/css\" >\n",
       "    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2 th {\n",
       "          text-align: center;\n",
       "    }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row0_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row0_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row0_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row0_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row1_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row1_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row1_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row1_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row2_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row2_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row2_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row2_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row3_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row3_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row3_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row3_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row4_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row4_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row4_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row4_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row5_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row5_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row5_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row5_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row6_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row6_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row6_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row6_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row7_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row7_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row7_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row7_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row8_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row8_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row8_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row8_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row9_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row9_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row9_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row9_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row10_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row10_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row10_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row10_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row11_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row11_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row11_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row11_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row12_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row12_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row12_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row12_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row13_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row13_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row13_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row13_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row14_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row14_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row14_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row14_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row15_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row15_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row15_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row15_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row16_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row16_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row16_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row16_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row17_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row17_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row17_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row17_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row18_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row18_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row18_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row18_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row19_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row19_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row19_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row19_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row20_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row20_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row20_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row20_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }</style><table id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" >class</th>        <th class=\"col_heading level0 col1\" >input_shape</th>        <th class=\"col_heading level0 col2\" >output_shape</th>        <th class=\"col_heading level0 col3\" >trainable</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row0_col0\" class=\"data row0 col0\" >InputLayer</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row0_col1\" class=\"data row0 col1\" >(280)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row0_col2\" class=\"data row0 col2\" >(280)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row0_col3\" class=\"data row0 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row1_col0\" class=\"data row1 col0\" >Embedding</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row1_col1\" class=\"data row1 col1\" >(280)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row1_col2\" class=\"data row1 col2\" >(280, 300)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row1_col3\" class=\"data row1 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row2_col0\" class=\"data row2 col0\" >SpatialDropout1D</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row2_col1\" class=\"data row2 col1\" >(280, 300)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row2_col2\" class=\"data row2 col2\" >(280, 300)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row2_col3\" class=\"data row2 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row3_col0\" class=\"data row3 col0\" >Bidirectional</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row3_col1\" class=\"data row3 col1\" >(280, 300)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row3_col2\" class=\"data row3 col2\" >(280, 136)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row3_col3\" class=\"data row3 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row4_col0\" class=\"data row4 col0\" >Bidirectional</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row4_col1\" class=\"data row4 col1\" >(280, 300)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row4_col2\" class=\"data row4 col2\" >(280, 136)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row4_col3\" class=\"data row4 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row5_col0\" class=\"data row5 col0\" >Conv1D</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row5_col1\" class=\"data row5 col1\" >(280, 136)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row5_col2\" class=\"data row5 col2\" >(279, 28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row5_col3\" class=\"data row5 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row6_col0\" class=\"data row6 col0\" >Conv1D</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row6_col1\" class=\"data row6 col1\" >(280, 136)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row6_col2\" class=\"data row6 col2\" >(278, 28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row6_col3\" class=\"data row6 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row7_col0\" class=\"data row7 col0\" >Conv1D</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row7_col1\" class=\"data row7 col1\" >(280, 136)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row7_col2\" class=\"data row7 col2\" >(279, 28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row7_col3\" class=\"data row7 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row8_col0\" class=\"data row8 col0\" >Conv1D</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row8_col1\" class=\"data row8 col1\" >(280, 136)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row8_col2\" class=\"data row8 col2\" >(278, 28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row8_col3\" class=\"data row8 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row9_col0\" class=\"data row9 col0\" >GlobalMaxPooling1D</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row9_col1\" class=\"data row9 col1\" >(279, 28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row9_col2\" class=\"data row9 col2\" >(28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row9_col3\" class=\"data row9 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row10_col0\" class=\"data row10 col0\" >GlobalMaxPooling1D</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row10_col1\" class=\"data row10 col1\" >(278, 28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row10_col2\" class=\"data row10 col2\" >(28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row10_col3\" class=\"data row10 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row11_col0\" class=\"data row11 col0\" >GlobalMaxPooling1D</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row11_col1\" class=\"data row11 col1\" >(279, 28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row11_col2\" class=\"data row11 col2\" >(28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row11_col3\" class=\"data row11 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row12_col0\" class=\"data row12 col0\" >GlobalMaxPooling1D</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row12_col1\" class=\"data row12 col1\" >(278, 28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row12_col2\" class=\"data row12 col2\" >(28)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row12_col3\" class=\"data row12 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row13_col0\" class=\"data row13 col0\" >Concatenate</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row13_col1\" class=\"data row13 col1\" >[(28), (28), (28), (28)]</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row13_col2\" class=\"data row13 col2\" >(112)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row13_col3\" class=\"data row13 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row14_col0\" class=\"data row14 col0\" >BatchNormalization</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row14_col1\" class=\"data row14 col1\" >(112)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row14_col2\" class=\"data row14 col2\" >(112)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row14_col3\" class=\"data row14 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row15_col0\" class=\"data row15 col0\" >Dense</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row15_col1\" class=\"data row15 col1\" >(112)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row15_col2\" class=\"data row15 col2\" >(46)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row15_col3\" class=\"data row15 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row16_col0\" class=\"data row16 col0\" >Dropout</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row16_col1\" class=\"data row16 col1\" >(46)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row16_col2\" class=\"data row16 col2\" >(46)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row16_col3\" class=\"data row16 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row17_col0\" class=\"data row17 col0\" >BatchNormalization</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row17_col1\" class=\"data row17 col1\" >(46)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row17_col2\" class=\"data row17 col2\" >(46)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row17_col3\" class=\"data row17 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row18_col0\" class=\"data row18 col0\" >Dense</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row18_col1\" class=\"data row18 col1\" >(46)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row18_col2\" class=\"data row18 col2\" >(32)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row18_col3\" class=\"data row18 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row19_col0\" class=\"data row19 col0\" >Dropout</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row19_col1\" class=\"data row19 col1\" >(32)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row19_col2\" class=\"data row19 col2\" >(32)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row19_col3\" class=\"data row19 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row20_col0\" class=\"data row20 col0\" >Dense</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row20_col1\" class=\"data row20 col1\" >(32)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row20_col2\" class=\"data row20 col2\" >(3)</td>\n",
       "                        <td id=\"T_e94b35a4_eeb8_11ea_9f2b_309c23ff89f2row20_col3\" class=\"data row20 col3\" >True</td>\n",
       "            </tr>\n",
       "    </tbody></table></div></body></html>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "for layer in tweets_model.layers[:-1] :\n",
    "    layer.trainable = False\n",
    "\n",
    "layers_df = pd.DataFrame()\n",
    "for layer in tweets_model.layers :\n",
    "    layers_df = layers_df.append([[str(layer.__class__).split('.')[-1][:-2], str(layer.input_shape).replace('None, ', '')\n",
    "                                   , str(layer.output_shape).replace('None, ', ''), layer.trainable]]\n",
    "                                 , ignore_index = True, verify_integrity=False\n",
    "    )\n",
    "layers_df.columns=['class', 'input_shape', 'output_shape', 'trainable']\n",
    "dataframe_pretty_print_center(layers_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "We perform our model warming-up using the same setup as the one we'll use next, during the fine-tuning of our model. That is that we use 'categorical_crossentropy' as our loss function, <small><b><em>keras.optimizers.Adam</em></b></small> as our optimizer, we do measure not only 'accuracy' (not that meaningful with imbalanced datasets) but also 'precision' and 'recall', both globally and per class&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 133,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "from tqdm.keras import TqdmCallback\n",
    "from keras.optimizers import Adam\n",
    "from keras.metrics import Precision, Recall, CategoricalCrossentropy, CategoricalAccuracy\n",
    "# f1_score = 2*(precision*recall)/(precision+recall+K.epsilon())\n",
    "\n",
    "warming_up_lr=0.001 # default: 0.01\n",
    "optimizer = Adam( learning_rate = warming_up_lr )\n",
    "metrics = [CategoricalCrossentropy(), CategoricalAccuracy()\n",
    "           , Precision(name='precision'), Recall(name='recall')\n",
    "           , Precision(name='precision_NEG', class_id=0), Recall(name='recall_NEG', class_id=0)\n",
    "           , Precision(name='precision_NEUT', class_id=1), Recall(name='recall_NEUT', class_id=1)\n",
    "           , Precision(name='precision_POS', class_id=2), Recall(name='recall_POS', class_id=2)\n",
    "          ]\n",
    "\n",
    "tweets_model.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=metrics)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a0f277584e234abba36babec2b32a6e1",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "verbose = 1\n",
    "warming_up_history = \\\n",
    "    tweets_model.fit(tweets_X_train_resampled, tweets_y_train_resampled_1_hot\n",
    "                     , batch_size = 1024, epochs=15\n",
    "                     , validation_data=(tweets_X_valid, tweets_y_valid_1_hot)\n",
    "                     , verbose = 0\n",
    "                     , callbacks = [TqdmCallback(verbose=verbose)]\n",
    "                    )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img align=\"left\" alt=\"fine-tuning widget\" src=\"./images/widget.png?uncache=464453868\" />\n",
    "\n",
    "<ul style=\"line-height:26px; list-style-type:none\">\n",
    "<li style=\"margin: 0px 10px 0px 190px;\">15/15 [16:55&lt;00:00, 67.70s/epoch, val_loss=1.54, val_categorical_crossentropy=1.13, val_categorical_accuracy=0.165, val_precision=0, val_recall=0, val_precision_NEG=0, val_recall_NEG=0, val_precision_NEUT=0, val_recall_NEUT=0, val_precision_POS=0, val_recall_POS=0, loss=1.51, categorical_crossentropy=1.09, categorical_accuracy=0.368, precision=0.485, recall=0.00517, precision_NEG=0, recall_NEG=0, precision_NEUT=0.487, recall_NEUT=0.0153, precision_POS=0.583, recall_POS=0.000263]</p>\n",
    "</li>\n",
    "</ul>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "The whole idea behind warming-up consists in bootstrapping the transfer learning. We do that by assigning weights to the new output layer that are more optimum than the initial random values. By freezing all the preceeding layers, the optimizer focuses and these last weights alone, which gives them a huge advantage when later actually operating the model fine-tuning in the upcoming section.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"tuning\"></a>\n",
    "<h3>5.3. Fine-Tuning <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Now that our model has learned from a generic large corpus of french texts and that our new output layer has been warmed-up on the target French Tweets dataset, the actual fine-tuning of our French Sentiment NLP model can start. Lets first \"un-freeze\" the model's inner layer&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 132,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <!DOCTYPE html><html><head>\n",
       "        <style>\n",
       "        div.centered_table_container {\n",
       "            text-align: center;\n",
       "        }\n",
       "        table {\n",
       "            display: inline-block;\n",
       "        }\n",
       "        </style></head>\n",
       "        <body>\n",
       "        <div class=\"centered_table_container\"><style  type=\"text/css\" >\n",
       "    #T_31e13f48_eebb_11ea_8396_309c23ff89f2 th {\n",
       "          text-align: center;\n",
       "    }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row0_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row0_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row0_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row0_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row1_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row1_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row1_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row1_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row2_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row2_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row2_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row2_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row3_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row3_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row3_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row3_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row4_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row4_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row4_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row4_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row5_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row5_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row5_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row5_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row6_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row6_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row6_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row6_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row7_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row7_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row7_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row7_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row8_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row8_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row8_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row8_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row9_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row9_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row9_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row9_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row10_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row10_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row10_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row10_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row11_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row11_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row11_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row11_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row12_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row12_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row12_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row12_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row13_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row13_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row13_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row13_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row14_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row14_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row14_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row14_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row15_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row15_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row15_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row15_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row16_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row16_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row16_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row16_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row17_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row17_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row17_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row17_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row18_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row18_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row18_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row18_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row19_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row19_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row19_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row19_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row20_col0 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row20_col1 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row20_col2 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }    #T_31e13f48_eebb_11ea_8396_309c23ff89f2row20_col3 {\n",
       "            width:  20em;\n",
       "            text-align:  center;\n",
       "        }</style><table id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" >class</th>        <th class=\"col_heading level0 col1\" >input_shape</th>        <th class=\"col_heading level0 col2\" >output_shape</th>        <th class=\"col_heading level0 col3\" >trainable</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row0_col0\" class=\"data row0 col0\" >InputLayer</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row0_col1\" class=\"data row0 col1\" >(280)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row0_col2\" class=\"data row0 col2\" >(280)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row0_col3\" class=\"data row0 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row1_col0\" class=\"data row1 col0\" >Embedding</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row1_col1\" class=\"data row1 col1\" >(280)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row1_col2\" class=\"data row1 col2\" >(280, 300)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row1_col3\" class=\"data row1 col3\" >False</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row2_col0\" class=\"data row2 col0\" >SpatialDropout1D</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row2_col1\" class=\"data row2 col1\" >(280, 300)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row2_col2\" class=\"data row2 col2\" >(280, 300)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row2_col3\" class=\"data row2 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row3_col0\" class=\"data row3 col0\" >Bidirectional</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row3_col1\" class=\"data row3 col1\" >(280, 300)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row3_col2\" class=\"data row3 col2\" >(280, 136)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row3_col3\" class=\"data row3 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row4_col0\" class=\"data row4 col0\" >Bidirectional</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row4_col1\" class=\"data row4 col1\" >(280, 300)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row4_col2\" class=\"data row4 col2\" >(280, 136)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row4_col3\" class=\"data row4 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row5_col0\" class=\"data row5 col0\" >Conv1D</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row5_col1\" class=\"data row5 col1\" >(280, 136)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row5_col2\" class=\"data row5 col2\" >(279, 28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row5_col3\" class=\"data row5 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row6_col0\" class=\"data row6 col0\" >Conv1D</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row6_col1\" class=\"data row6 col1\" >(280, 136)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row6_col2\" class=\"data row6 col2\" >(278, 28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row6_col3\" class=\"data row6 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row7_col0\" class=\"data row7 col0\" >Conv1D</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row7_col1\" class=\"data row7 col1\" >(280, 136)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row7_col2\" class=\"data row7 col2\" >(279, 28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row7_col3\" class=\"data row7 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row8_col0\" class=\"data row8 col0\" >Conv1D</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row8_col1\" class=\"data row8 col1\" >(280, 136)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row8_col2\" class=\"data row8 col2\" >(278, 28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row8_col3\" class=\"data row8 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row9_col0\" class=\"data row9 col0\" >GlobalMaxPooling1D</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row9_col1\" class=\"data row9 col1\" >(279, 28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row9_col2\" class=\"data row9 col2\" >(28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row9_col3\" class=\"data row9 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row10_col0\" class=\"data row10 col0\" >GlobalMaxPooling1D</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row10_col1\" class=\"data row10 col1\" >(278, 28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row10_col2\" class=\"data row10 col2\" >(28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row10_col3\" class=\"data row10 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row11_col0\" class=\"data row11 col0\" >GlobalMaxPooling1D</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row11_col1\" class=\"data row11 col1\" >(279, 28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row11_col2\" class=\"data row11 col2\" >(28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row11_col3\" class=\"data row11 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row12_col0\" class=\"data row12 col0\" >GlobalMaxPooling1D</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row12_col1\" class=\"data row12 col1\" >(278, 28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row12_col2\" class=\"data row12 col2\" >(28)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row12_col3\" class=\"data row12 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row13_col0\" class=\"data row13 col0\" >Concatenate</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row13_col1\" class=\"data row13 col1\" >[(28), (28), (28), (28)]</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row13_col2\" class=\"data row13 col2\" >(112)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row13_col3\" class=\"data row13 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row14_col0\" class=\"data row14 col0\" >BatchNormalization</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row14_col1\" class=\"data row14 col1\" >(112)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row14_col2\" class=\"data row14 col2\" >(112)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row14_col3\" class=\"data row14 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row15_col0\" class=\"data row15 col0\" >Dense</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row15_col1\" class=\"data row15 col1\" >(112)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row15_col2\" class=\"data row15 col2\" >(46)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row15_col3\" class=\"data row15 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row16_col0\" class=\"data row16 col0\" >Dropout</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row16_col1\" class=\"data row16 col1\" >(46)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row16_col2\" class=\"data row16 col2\" >(46)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row16_col3\" class=\"data row16 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row17_col0\" class=\"data row17 col0\" >BatchNormalization</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row17_col1\" class=\"data row17 col1\" >(46)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row17_col2\" class=\"data row17 col2\" >(46)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row17_col3\" class=\"data row17 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row18_col0\" class=\"data row18 col0\" >Dense</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row18_col1\" class=\"data row18 col1\" >(46)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row18_col2\" class=\"data row18 col2\" >(32)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row18_col3\" class=\"data row18 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row19_col0\" class=\"data row19 col0\" >Dropout</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row19_col1\" class=\"data row19 col1\" >(32)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row19_col2\" class=\"data row19 col2\" >(32)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row19_col3\" class=\"data row19 col3\" >True</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row20_col0\" class=\"data row20 col0\" >Dense</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row20_col1\" class=\"data row20 col1\" >(32)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row20_col2\" class=\"data row20 col2\" >(3)</td>\n",
       "                        <td id=\"T_31e13f48_eebb_11ea_8396_309c23ff89f2row20_col3\" class=\"data row20 col3\" >True</td>\n",
       "            </tr>\n",
       "    </tbody></table></div></body></html>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "for layer in tweets_model.layers[2:] :\n",
    "    layer.trainable = True\n",
    "\n",
    "layers_df = pd.DataFrame()\n",
    "for layer in tweets_model.layers :\n",
    "    layers_df = layers_df.append([[str(layer.__class__).split('.')[-1][:-2], str(layer.input_shape).replace('None, ', '')\n",
    "                                   , str(layer.output_shape).replace('None, ', ''), layer.trainable]]\n",
    "                                 , ignore_index = True, verify_integrity=False\n",
    "    )\n",
    "layers_df.columns=['class', 'input_shape', 'output_shape', 'trainable']\n",
    "dataframe_pretty_print_center(layers_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "We're now ready for \"<b>Step 3</b>\" of the summary figure seen in <a href=\"#page-top\"><b>&sect;0.</b></a>, i.e. the actual fine-tuning of our French NLP model&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "fine_tuning_lr = 0.001\n",
    "optimizer = Adam( learning_rate = fine_tuning_lr )\n",
    "tweets_model.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=metrics)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 137,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e971db6680bf409995bc543a71d04cd9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 0 ; lr: 0.001000\n",
      "epoch 30 ; lr: 0.000100\n",
      "epoch 40 ; lr: 0.000010\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from keras.callbacks import LearningRateScheduler\n",
    "from my_NLP_RNN_fr_lib.model.callbacks import LearningRatePrinter, lr_sheduler_func\n",
    "\n",
    "my_learning_rate_printer = LearningRatePrinter()\n",
    "\n",
    "learning_rates_decays = {30: .1, 40: .1}\n",
    "my_lr_scheduler = LearningRateScheduler(\n",
    "    lr_sheduler_func(learning_rates_decays = learning_rates_decays\n",
    "                     , learning_rate_printer = my_learning_rate_printer))\n",
    "\n",
    "verbose = 1\n",
    "callbacks = [my_lr_scheduler, my_learning_rate_printer, TqdmCallback(verbose=verbose)]\n",
    "\n",
    "fine_tuning_history = \\\n",
    "    tweets_model.fit(tweets_X_train_resampled, tweets_y_train_resampled_1_hot\n",
    "                     , batch_size = 512, epochs=45\n",
    "                     , validation_data=(tweets_X_valid, tweets_y_valid_1_hot)\n",
    "                     , verbose = 0, callbacks = callbacks)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img align=\"left\" alt=\"fine-tuning widget\" src=\"./images/widget.png?uncache=464453868\" />\n",
    "\n",
    "<ul style=\"line-height:26px; list-style-type:none\">\n",
    "<li style=\"margin: 0px 10px 0px 190px;\">\n",
    "<p style=\"font-family: Helvetica; font-size: 10pt\">45/45 [2:03:59&lt;00:00, 165.33s/epoch, val_loss=0.869, val_categorical_crossentropy=0.867, val_categorical_accuracy=0.63, val_precision=0.734, val_recall=0.508, val_precision_NEG=0.0952, val_recall_NEG=0.339, val_precision_NEUT=0.942, val_recall_NEUT=0.549, val_precision_POS=0.426, val_recall_POS=0.318, loss=0.942, categorical_crossentropy=0.94, categorical_accuracy=0.547, precision=0.632, recall=0.375, precision_NEG=0.575, recall_NEG=0.374, precision_NEUT=0.694, recall_NEUT=0.479, precision_POS=0.619, recall_POS=0.272, lr=1e-5]</p>\n",
    "</li>\n",
    "</ul>\n",
    "<p style=\"font-family: monospace\">\n",
    "epoch 0 ; lr: 0.001000<br />\n",
    "epoch 30 ; lr: 0.000100<br />\n",
    "epoch 40 ; lr: 0.000010</p>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Lets visualize the history of our 2-phases training. In the below figure, the grayed-out area represents the warming-up. The remaining covers the fine-tuning phase&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "warming_up_history_df = warming_up_history.history\n",
    "fine_tuning_history_df = fine_tuning_history.history"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 432,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "from my_NLP_RNN_fr_lib.model.plot import warming_up_fine_tuning_history\n",
    "\n",
    "fig1, fig2 = warming_up_fine_tuning_history(\n",
    "    warming_up_history_df, warming_up_lr, fine_tuning_history_df\n",
    "    , fig1_size=(6, 14), fig2_size=(6, 18)\n",
    ")# ; plt.show()\n",
    "\n",
    "fig_path = os.path.join('.', 'images')\n",
    "fig1.savefig(os.path.join(fig_path, 'tweets_model_history.png'), bbox_inches='tight')\n",
    "plt.close(fig1)\n",
    "fig2.savefig(os.path.join(fig_path, 'tweets_model_classes_history.png'), bbox_inches='tight')\n",
    "plt.close(fig2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "<img alt='French Tweets model history' style='border-right: 5px solid transparent;' width='446px' align='left' src='./images/tweets_model_history.png?uncache=6315568' />\n",
    "Thanks to the knowledge it had previously acquired during the Step 1 (pre-training), the French Tweets model is consistently better performing on the validation dataset than on the training dataset accross all the measured performance metrics. It is excellent at generalizing.<br />\n",
    "The only exception to that is if we look at 'class' level, for the 2 non-dominant classes, on the 'precision' front. But 'recall' is much stronger. It can be better observed on <a target='_blank' href='./images/tweets_model_classes_history.png'>that other arragement of the data <img href='.' src='./images/target_blank.png' style='vertical-align: baseline; display: inline;' /></a>.\n",
    "<br clear=\"left\" />\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<hr style=\"height:2px;border-width:0;color:gray;background-color:gray;width:80%\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "Now that the training of our French Sentiment NLP model has been completed, it is time to independently measure how good it is actuially performing and there's no better way to do that than to predict some French Tweets sentiment from a real-life data distribution that had never been presented to that model before ; the holdout dataset&nbsp;:\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "metadata": {},
   "outputs": [],
   "source": [
    "predictions_1_hot = tweets_model.predict(tweets_X_holdout)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 715,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "\n",
    "cm_df = pd.DataFrame(\n",
    "    confusion_matrix(tweets_y_holdout_1_hot.to_numpy().argmax(axis=1)\n",
    "                     , predictions_1_hot.argmax(axis=1))\n",
    ")\n",
    "cm_df.columns = cm_df.index = tweets_y_holdout_1_hot.columns\n",
    "cm_html = cm_df.to_html(formatters={'NEUTRAL': '{:,}'.format})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 731,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import classification_report\n",
    "from my_NLP_RNN_fr_lib.display_helper import classification_report_pretty_html\n",
    "\n",
    "report_dict = \\\n",
    "    classification_report(tweets_y_holdout_1_hot.to_numpy().argmax(axis=1)\n",
    "                          , predictions_1_hot.argmax(axis=1)\n",
    "                          , target_names = tweets_y_holdout_1_hot.columns\n",
    "                          , zero_division  = 1\n",
    "                          , output_dict = True)\n",
    "\n",
    "report_html = classification_report_pretty_html(report_dict)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 737,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Performance of the final model on the holdout dataset :\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "<!DOCTYPE html>\n",
       "<html>\n",
       "<head>\n",
       "<style>\n",
       "\n",
       "* {\n",
       "    box-sizing: border-box;\n",
       "}\n",
       "\n",
       ".left_column {\n",
       "    float: left;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "}\n",
       "\n",
       ".right_column {\n",
       "    float:right;\n",
       "    text-align; right;\n",
       "    padding: 10px;\n",
       "    #padding-left: 80px;\n",
       "    #background: yellow;\n",
       "}\n",
       "\n",
       "/* Clearfix (clear floats) */\n",
       ".row::after {\n",
       "    content: \"\";\n",
       "    clear: both;\n",
       "    display: table;\n",
       "}\n",
       "\n",
       "</style>\n",
       "</head>\n",
       "<body>\n",
       "\n",
       "<h2></h2>\n",
       "\n",
       "<div class=\"row\" style=\"width: 100%;\">\n",
       "  <div class=\"left_column\" style=\"width: 40.00%;\">\n",
       "      <center>\n",
       "          <br /><br /><div>Confusion Matrix</div><table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>NEGATIVE</th>\n",
       "      <th>NEUTRAL</th>\n",
       "      <th>POSITIVE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>NEGATIVE</th>\n",
       "      <td>61</td>\n",
       "      <td>18</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>NEUTRAL</th>\n",
       "      <td>500</td>\n",
       "      <td>1,705</td>\n",
       "      <td>282</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>POSITIVE</th>\n",
       "      <td>115</td>\n",
       "      <td>124</td>\n",
       "      <td>185</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "      </center>\n",
       "  </div>\n",
       "  <div class=\"right_column\" style=\"width: 60.00%;\">\n",
       "      <center>\n",
       "          <div>Classification Report</div><style  type=\"text/css\" >\n",
       "    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row0_col0 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row0_col1 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row0_col2 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row0_col3 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row0_col4 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row1_col0 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row1_col1 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row1_col2 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row1_col3 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row1_col4 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row2_col0 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row2_col1 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row2_col2 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row2_col3 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row2_col4 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row3_col0 {\n",
       "            border-top:  1px solid black;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row3_col1 {\n",
       "            border-top:  1px solid black;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row3_col2 {\n",
       "            border-top:  1px solid black;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row3_col3 {\n",
       "            border-top:  1px solid black;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row3_col4 {\n",
       "            border-top:  1px solid black;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row4_col0 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row4_col1 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row4_col2 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row4_col3 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row4_col4 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row5_col0 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row5_col1 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row5_col2 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row5_col3 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row5_col4 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row6_col0 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row6_col1 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row6_col2 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row6_col3 {\n",
       "            border:  0px;\n",
       "        }    #T_54e28f76_f060_11ea_8fb0_309c23ff89f2row6_col4 {\n",
       "            border:  0px;\n",
       "        }</style><table id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2\" ><thead>    <tr>        <th class=\"col_heading level0 col0\" ></th>        <th class=\"col_heading level0 col1\" >precision</th>        <th class=\"col_heading level0 col2\" >recall</th>        <th class=\"col_heading level0 col3\" >f1-score</th>        <th class=\"col_heading level0 col4\" >support</th>    </tr></thead><tbody>\n",
       "                <tr>\n",
       "                                <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row0_col0\" class=\"data row0 col0\" >NEGATIVE</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row0_col1\" class=\"data row0 col1\" >0.09</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row0_col2\" class=\"data row0 col2\" >0.69</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row0_col3\" class=\"data row0 col3\" >0.16</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row0_col4\" class=\"data row0 col4\" >89</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row1_col0\" class=\"data row1 col0\" >NEUTRAL</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row1_col1\" class=\"data row1 col1\" >0.92</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row1_col2\" class=\"data row1 col2\" >0.69</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row1_col3\" class=\"data row1 col3\" >0.79</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row1_col4\" class=\"data row1 col4\" >2,487</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row2_col0\" class=\"data row2 col0\" >POSITIVE</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row2_col1\" class=\"data row2 col1\" >0.39</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row2_col2\" class=\"data row2 col2\" >0.44</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row2_col3\" class=\"data row2 col3\" >0.41</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row2_col4\" class=\"data row2 col4\" >424</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row3_col0\" class=\"data row3 col0\" >&nbsp;</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row3_col1\" class=\"data row3 col1\" ></td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row3_col2\" class=\"data row3 col2\" ></td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row3_col3\" class=\"data row3 col3\" ></td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row3_col4\" class=\"data row3 col4\" ></td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row4_col0\" class=\"data row4 col0\" >accuracy</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row4_col1\" class=\"data row4 col1\" ></td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row4_col2\" class=\"data row4 col2\" ></td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row4_col3\" class=\"data row4 col3\" >0.65</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row4_col4\" class=\"data row4 col4\" >3,000</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row5_col0\" class=\"data row5 col0\" >macro avg</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row5_col1\" class=\"data row5 col1\" >0.47</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row5_col2\" class=\"data row5 col2\" >0.60</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row5_col3\" class=\"data row5 col3\" >0.45</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row5_col4\" class=\"data row5 col4\" >3,000</td>\n",
       "            </tr>\n",
       "            <tr>\n",
       "                                <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row6_col0\" class=\"data row6 col0\" >weighted avg</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row6_col1\" class=\"data row6 col1\" >0.82</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row6_col2\" class=\"data row6 col2\" >0.65</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row6_col3\" class=\"data row6 col3\" >0.72</td>\n",
       "                        <td id=\"T_54e28f76_f060_11ea_8fb0_309c23ff89f2row6_col4\" class=\"data row6 col4\" >3,000</td>\n",
       "            </tr>\n",
       "    </tbody></table>\n",
       "      </center>\n",
       "  </div>\n",
       "</div>\n",
       "</body>\n",
       "</html>\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('Performance of the final model on the holdout dataset :')\n",
    "two_columns_display('<br /><br /><div>Confusion Matrix</div>' + cm_html\n",
    "                    , '<div>Classification Report</div>' + report_html, left_ratio  = .4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "variables": {
     "'{:0.0f}%'.format(report['NEGATIVE']['precision'] * 100)": "9%",
     "'{:0.0f}%'.format(report['NEGATIVE']['recall'] * 100)": "69%",
     "'{:0.0f}%'.format(report['weighted avg']['precision'] * 100)": "82%",
     "'{:0.0f}%'.format(report['weighted avg']['recall'] * 100)": "65%"
    }
   },
   "source": [
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2);'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "As the Confusion Matrix and the Classification Report show it above &nbsp;:\n",
    "    <ul style=\"list-style-type: disc; margin: 0px 150px 0px 50px; /* top right bottom left*/\">\n",
    "        <li>the overall Precision of our model stands at the more than honorable level of {{'{:0.0f}%'.format(report['weighted avg']['precision'] * 100)}}. Despite all of our efforts, it does however still vary a lot accross classes. For example, <b>when our model predicts a French Tweet to be 'NEGATIVE', it is correct only {{'{:0.0f}%'.format(report['NEGATIVE']['precision'] * 100)}} of the time</b>. This is undesirable and shall increase when more training data becomes available.</li>\n",
    "        <li>if the overall Recall of our model is a little less impressive than its Precision with an however still respectable {{'{:0.0f}%'.format(report['weighted avg']['recall'] * 100)}}, it is more even accross classes than the latter. <b><span style=\"background-color: #00FFFF;\">&nbsp;Regarding NEGATIVE French Tweets for instance, our NLP model succesfully identifies them {{'{:0.0f}%'.format(report['NEGATIVE']['recall'] * 100)}} of the time. This is an impressive performance considering such an imbalanced real-life data distribution (and the tiny total occurences of NEGATIVE records in the entire population)&nbsp;</span></b>.</li>\n",
    "    </ul>\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<hr style=\"height: 4px; width: 70%; margin:0 auto;\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "backup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 413,
   "metadata": {},
   "outputs": [],
   "source": [
    "import dill as pickle\n",
    "root = os.path.realpath(os.path.join('.', 'model', 'tweets_model'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 402,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(os.path.join(root, \"model.json\"), \"w\") as json_file :\n",
    "    json_file.write(tweets_model.to_json())\n",
    "tweets_model.save_weights(os.path.join(root, 'model_weights.h5'))\n",
    "\n",
    "with open(os.path.join(root, 'warming_up_history_df.pickle'), 'wb') as file_pi:\n",
    "    pickle.dump(warming_up_history.history, file_pi)\n",
    "\n",
    "with open(os.path.join(root, 'fine_tuning_history_df.pickle'), 'wb') as file_pi:\n",
    "    pickle.dump(fine_tuning_history.history, file_pi)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "reload"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 409,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "from keras.models import model_from_json\n",
    "\n",
    "with open(os.path.join(root, \"model.json\"), 'r') as json_file :\n",
    "    loaded_model_json = json_file.read()\n",
    "tweets_model = model_from_json(loaded_model_json)\n",
    "tweets_model.load_weights(os.path.join(root, 'model_weights.h5'))\n",
    "\n",
    "with open(os.path.join(root, 'warming_up_history_df.pickle'), 'rb') as file_pi:\n",
    "    warming_up_history_df = pickle.load(file_pi)\n",
    "\n",
    "with open(os.path.join(root, 'fine_tuning_history_df.pickle'), 'rb') as file_pi:\n",
    "    fine_tuning_history_df = pickle.load(file_pi)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<hr style=\"height: 4px; width: 70%; margin:0 auto;\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"conclusion\"></a>\n",
    "<h2>6. Conclusion <small><em><a href=\"#page-top\">(go back to top &uarr;)]</a></em></small></h2>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<hr style=\"height:0pt; visibility:hidden; margin-bottom: 10px; margin-top:0px;\" />\n",
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2); padding-top:0px'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "NLP models on languages others than English are seldom. And it's mostly because labelled data in those languages is hard to come by.\n",
    "</div>\n",
    "</div>\n",
    "<hr style=\"height:0pt; visibility:hidden; margin-bottom: 10px; margin-top:0px;\" />\n",
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2); padding-bottom:0px'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "The project of the herein notebook being Sentiment Classification on French Texts, we've covered many aspects of techniques that can be handy when working on real-world data which can often be imperfect and/or costly to acquire&nbsp;:\n",
    "    <ul style=\"list-style-type: disc; margin: 0px 150px 0px 50px; /* top right bottom left*/\">\n",
    "        <li>we've seen the power of <u>transfer learning</u> from web-scrapped, rather generic, French Reviews to domain-specific French Tweets (specific to soccer in our use case).</li>\n",
    "        <li>we've also used <u>XGBoost</u> as a supercharger to our hyperparameters optimization (in <a href=\"https://github.com/aurelienmorgan/french_text_sentiment/blob/master/model/optimization/README.md\" target=\"_blank\">this separate dedicated notebook <img href='.' src='./images/target_blank.png' style='vertical-align: baseline; display: inline;' /></a>)</li>\n",
    "        <li>we've also covered means at our disposal when dealing with <u>highly imbalanced datasets</u></li>\n",
    "    </ul>\n",
    "</div>\n",
    "</div>\n",
    "<hr style=\"height:0pt; visibility:hidden; margin-bottom: 10px; margin-top:0px;\" />\n",
    "<div style='background-color: rgba(0, 255, 255, 0.04); border: 1px solid rgba(0, 255, 255, .2); padding-top:0px'>\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "The above-depicted technical \"tricks\" in our toolset allowed to train a dual bi-GRU model architecture in an optimum fashion for such a model to tackle efficiently some everyday-life problems.\n",
    "</div>\n",
    "<hr style=\"height:0pt; visibility:hidden; margin-bottom: 10px; margin-top:0px;\" />\n",
    "<div style='text-align: justify; margin-left: 5px; margin-right: 5px;'>\n",
    "    Ours here is <em>domain-specific to soccer</em>, but any public or private organization could \"tune\" our model, for instance on a <u>brand-specific dataset</u> (which it would posess internally), by simply following <a href=\"#warming\"><b>&sect;5.2.</b></a> &amp; <a href=\"#tuning\"><b>&sect;5.3.</b></a>, which is really quick.\n",
    "\n",
    "\n",
    "\n",
    "Such an NLP model would for instance be a well suited technical solution for anyone willing to implement an automatic Twitter monitoring system for a brand operating on French-speaking markets. This would allow to identify any potential \"bad buzz\" on the social media platform around the brand as-soon-as any ever occurs.\n",
    "On the other hand, being notified on POSITIVE French Tweets would for instance allow to implement a targetted promotional campaign involving enthousiastic/supportive customers and thus build a resilient community around the brand.\n",
    "</div>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<hr style=\"height:2px;border-width:0;color:gray;background-color:gray;width:80%\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# EXTRA"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<b>Export Notebook to HTML (with Markdown extension cells evaluated)</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "'main.ipynb' ; 'D:\\jupyter_notebooks\\NLP_Sentiment_Analysis_French\\main.html'\n",
      "done.\n"
     ]
    }
   ],
   "source": [
    "from my_NLP_RNN_fr_lib.jupyter_markdown_extension import md_extension_to_html\n",
    "\n",
    "md_extension_to_html(os.path.join(os.path.realpath('.'), 'main.ipynb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "TensorFlow-GPU-1.13.1",
   "language": "python",
   "name": "r-tensorflow"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.8"
  },
  "widgets": {
   "application/vnd.jupyter.widget-state+json": {
    "state": {
     "0053536bb734443d96fb28c5e1cf94f2": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "0439bf739c9b4d7abd188ec21b0326f1": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "04794d3820e14dd0ba74807cbd6b026b": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "description": "100%",
       "layout": "IPY_MODEL_99453bf82aea49a9b57d404657204db5",
       "max": 464208,
       "style": "IPY_MODEL_20b343d6d4e440c69d89f0916af02fda",
       "value": 464208
      }
     },
     "050e28e0cf9547c09c00d075ad7a4746": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "description": "100%",
       "layout": "IPY_MODEL_1d2783bf31a34d3f8dbed455cb4fb2c8",
       "max": 58676,
       "style": "IPY_MODEL_27f88416decb4ef0b27dae63c09fd056",
       "value": 58676
      }
     },
     "07111bf4cfb24e50b8ea9aa0f8b92d9d": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "07694115710d41d6b00cbb5b4a18162f": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "08b16f147d034bed82707572d57d312f": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "0b8fd8fb959742c59fcab4db81ef468e": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_050e28e0cf9547c09c00d075ad7a4746",
        "IPY_MODEL_17f24383a5564629ab1b664cab4d95a2"
       ],
       "layout": "IPY_MODEL_ddf3afb8bbc8447b805dc1eb96e02d0b"
      }
     },
     "1234c7f4bdb54421b715049f732fdd2e": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "135949fd02fc47b39cd9c2647f62bb3c": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_8b9a86b3d91748b9a634152b2b00daf5",
        "IPY_MODEL_b933a0490ae14e8bab56b30a4e25698f"
       ],
       "layout": "IPY_MODEL_c88f4a593f7140e98f1abc0bdb01d7c6"
      }
     },
     "15c8381417da46fd8732b88ab8dee595": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_3e43f825bdb54f54a3f9b353b16076c5",
       "style": "IPY_MODEL_391f0f4513a84a488971a006b046e782",
       "value": " 60/60 [3:53:14&lt;00:00, 233.23s/epoch, val_loss=0.876, val_categorical_crossentropy=0.875, val_categorical_accuracy=0.574, val_precision_3=0.691, val_recall_3=0.502, loss=0.943, categorical_crossentropy=0.942, categorical_accuracy=0.545, precision_3=0.63, recall_3=0.364, lr=1e-6]"
      }
     },
     "1637d459f7474829b3f32d3f90e077da": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "17f24383a5564629ab1b664cab4d95a2": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_ddace66a14f147bc8c511b2a2f2274ff",
       "style": "IPY_MODEL_0439bf739c9b4d7abd188ec21b0326f1",
       "value": " 114.6015625/114.6015625 [01:57&lt;00:00,  1.06s/batch, loss=0.848, categorical_crossentropy=0.798]"
      }
     },
     "1a4659818d2e4e2ba58363ea98511975": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "1a81fea9051d491b8d67793e7520b316": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "1a94daae40504d3b95d3d3c427e059b0": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "1d2783bf31a34d3f8dbed455cb4fb2c8": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "1ff9190918794c73b26c2eccbb77da39": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_6d45efb90a854523a8190fa4497c96da",
       "style": "IPY_MODEL_4d4c60c7e5974df2800118ff2b942630",
       "value": " 15/15 [25:15&lt;00:00, 101.06s/epoch, val_loss=1.74, val_categorical_crossentropy=1.32, val_categorical_accuracy=0.0342, val_precision_2=0.011, val_recall_2=0.001, loss=1.58, categorical_crossentropy=1.17, categorical_accuracy=0.331, precision_2=0.288, recall_2=0.0783]"
      }
     },
     "20b343d6d4e440c69d89f0916af02fda": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "22021a05d97547ceaec3c04d6bcf1af0": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "222a29c01214468a903621bd5356f4e3": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "236214fc14bb4d7896b17c4ff3ed7e8a": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_9fd938861837465b850bf524468e5ece",
        "IPY_MODEL_8d1db6c684f6429cb8b7e34c2a3112a7"
       ],
       "layout": "IPY_MODEL_a5b291b178a84e4cb60fbecafa468e06"
      }
     },
     "2428340a39bb444db9c03195e95f3527": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "26f8dda8e4b14d92ab51617a3e35c366": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "27f88416decb4ef0b27dae63c09fd056": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "2b787b4ed2204046924e52c1e4bd2af0": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "2c84d4cc8e1842eb920fd2e3add76dbe": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "2cba71c444da48b28967b65248a93756": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_1a81fea9051d491b8d67793e7520b316",
       "style": "IPY_MODEL_4cdaa843cc4c4b4c848da40e11c977d3",
       "value": " 45/45 [2:03:59&lt;00:00, 165.33s/epoch, val_loss=0.869, val_categorical_crossentropy=0.867, val_categorical_accuracy=0.63, val_precision=0.734, val_recall=0.508, val_precision_NEG=0.0952, val_recall_NEG=0.339, val_precision_NEUT=0.942, val_recall_NEUT=0.549, val_precision_POS=0.426, val_recall_POS=0.318, loss=0.942, categorical_crossentropy=0.94, categorical_accuracy=0.547, precision=0.632, recall=0.375, precision_NEG=0.575, recall_NEG=0.374, precision_NEUT=0.694, recall_NEUT=0.479, precision_POS=0.619, recall_POS=0.272, lr=1e-5]"
      }
     },
     "3338e3eb86804811989376acb411503c": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "36f7f0504f8d499f8a45152bee479e50": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "37144931edc1439f912076b38ea7346e": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "3739ef941b994995a0a6c468de5b0ea4": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "379d36afddeb42069d5a130a0e9673f1": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "383bf6356ffd4599a6f50dda1877df16": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "391f0f4513a84a488971a006b046e782": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "3a7ff25f4ec94b48be34130d5dc535ef": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "3c12c4ec17fe42cab2814917937265c4": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "3e43f825bdb54f54a3f9b353b16076c5": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "3f28c888a66343e7af8f4a5f94b9ecd8": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "3f7cd5f714404e4398f8af5c68e35b6f": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_da46e6eeb3e545ccbdd9fbad6db16c26",
       "style": "IPY_MODEL_e0833f1a0114461aa6f5b949f61365d4",
       "value": " 155.923828125/155.923828125 [02:44&lt;00:00,  1.10batch/s, loss=0.974, categorical_crossentropy=0.94, categorical_accuracy=0.547, precision=0.632, recall=0.375, precision_NEG=0.575, recall_NEG=0.374, precision_NEUT=0.694, recall_NEUT=0.479, precision_POS=0.619, recall_POS=0.272]"
      }
     },
     "3f952be93e8d42d6a48621e03042f4c7": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "414af0d0b5fe43d1a7ad33ad47693a78": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "417d9c2e71e942cfaed538f124c1eac0": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "4286d119d26d44d9862b90672410ca46": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "44ca446e5cea46e3bb1ffe4be027a823": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "4be9957348384a0e906669b33fee4e17": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "4cdaa843cc4c4b4c848da40e11c977d3": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "4d4c60c7e5974df2800118ff2b942630": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "4e26db0ff18545a5a5bb09c22b52b74f": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "4e5c6485f79c4b2bb8001459dd07d8f8": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_9257d6425dd14ba28e73416cfa9cbce6",
        "IPY_MODEL_1ff9190918794c73b26c2eccbb77da39"
       ],
       "layout": "IPY_MODEL_feb468cecb424b599a08e5ea0c59c6c0"
      }
     },
     "4f92a48e998c4614be6b781becd1d236": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "51ae4bd973df42c786c2fbcd3263f474": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_9b16de9986bf4f73bd5f895f1a423bad",
        "IPY_MODEL_87b6de3cc8ab4ba385d3ac4c055df8f6"
       ],
       "layout": "IPY_MODEL_7d5550b409c8427db3ffb0f05efece10"
      }
     },
     "59ea1616b62a4be78ef8c7e4127b59b1": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "5f44e005650146f1bde873b2d3c19ca7": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_37144931edc1439f912076b38ea7346e",
       "style": "IPY_MODEL_63535025113642b08478b7f71f6a74af",
       "value": " 0/? [53:25&lt;?, ?epoch/s]"
      }
     },
     "611ea8ec17704916ad85cb40fea73293": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "layout": "IPY_MODEL_414af0d0b5fe43d1a7ad33ad47693a78",
       "max": 1,
       "style": "IPY_MODEL_fa727a6fb2224d42ad0d073477eaaa37"
      }
     },
     "626fbb172c0b453cbccb4ecfc9d7e077": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_a10fd773a9c44a668b0a1e4506f18fb8",
       "style": "IPY_MODEL_75cd49b9b0b04f38a98e010f9ceb10ac",
       "value": " 114.6015625/114.6015625 [02:17&lt;00:00,  1.21s/batch, loss=0.794, custom_metric=0.798]"
      }
     },
     "63535025113642b08478b7f71f6a74af": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "638a2e49af1747a0b2a26f8b384d626e": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "639e0ad22a8e422486c0ddda7f763173": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "646fcb2606084e1597496761fc753311": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "success",
       "description": "100%",
       "layout": "IPY_MODEL_4f92a48e998c4614be6b781becd1d236",
       "style": "IPY_MODEL_8bd2c262a7e64fd89547fa29e3b3157f",
       "value": 100
      }
     },
     "64f5de60821e4ea5807e8b470530a11f": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_73874c12df1c42e9b462a2d09071e868",
       "style": "IPY_MODEL_26f8dda8e4b14d92ab51617a3e35c366",
       "value": " 100/100 [23:18:40&lt;00:00, 839.21s/epoch, val_loss=0.848, val_accuracy=0.605, loss=0.879, accuracy=0.583, lr=1e-6]"
      }
     },
     "6642473d0aa541729d4f49597bb60f8b": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_611ea8ec17704916ad85cb40fea73293",
        "IPY_MODEL_9e799a85d9f5429d822640cd9f6e517a"
       ],
       "layout": "IPY_MODEL_d1b4507fd1044b7f9ded72830a4b7ff0"
      }
     },
     "68f5dee6d8444ea1863ae3469604c214": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "69a5320d83834adc84b6e1fabcbd2cf4": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "6ca4a85a193440c3ac34b713b4ca8d24": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "6d128f06b4f54f2894802ab50a84c5a6": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "6d45efb90a854523a8190fa4497c96da": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "6ecc08838573424c9e79987aa0a14863": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "714286b0cad446b6a46a69080c6bc449": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_e50788e484274ac9a669b593577e0252",
        "IPY_MODEL_94f3023761ea4b059d151d6f7617dae3"
       ],
       "layout": "IPY_MODEL_638a2e49af1747a0b2a26f8b384d626e"
      }
     },
     "72b88f0911cf40d6b3be584c25bc0ae3": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "description": "100%",
       "layout": "IPY_MODEL_e9747afbad43453883ae05fcbbb36828",
       "max": 79833,
       "style": "IPY_MODEL_07111bf4cfb24e50b8ea9aa0f8b92d9d",
       "value": 79833
      }
     },
     "73874c12df1c42e9b462a2d09071e868": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "7593c10a773045dfbb6091e8fa2a93e8": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "75cd49b9b0b04f38a98e010f9ceb10ac": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "76f21179eb01441fa3c84a777164f25d": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "786295c26a6d48ed875b42cffae39b46": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_d580c2f7a8f942c18d913da7d52e4678",
       "style": "IPY_MODEL_379d36afddeb42069d5a130a0e9673f1",
       "value": " 0/? [00:00&lt;?, ?batch/s]"
      }
     },
     "78926dd737e0474c9f850233dc1c0d99": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_646fcb2606084e1597496761fc753311",
        "IPY_MODEL_64f5de60821e4ea5807e8b470530a11f"
       ],
       "layout": "IPY_MODEL_6ecc08838573424c9e79987aa0a14863"
      }
     },
     "79ea8c5f789146dca16832433c969002": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "7afda720af0948d981ff227786c307b5": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "7b36f2197da4405fa4a8ffc1799b293d": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "7be5157b5e104f6eb8b37f3b95d16827": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_da61b5d67a8b4eadbd017316034e7e2b",
       "style": "IPY_MODEL_68f5dee6d8444ea1863ae3469604c214",
       "value": " 15/15 [16:55&lt;00:00, 67.70s/epoch, val_loss=1.54, val_categorical_crossentropy=1.13, val_categorical_accuracy=0.165, val_precision=0, val_recall=0, val_precision_NEG=0, val_recall_NEG=0, val_precision_NEUT=0, val_recall_NEUT=0, val_precision_POS=0, val_recall_POS=0, loss=1.51, categorical_crossentropy=1.09, categorical_accuracy=0.368, precision=0.485, recall=0.00517, precision_NEG=0, recall_NEG=0, precision_NEUT=0.487, recall_NEUT=0.0153, precision_POS=0.583, recall_POS=0.000263]"
      }
     },
     "7d5550b409c8427db3ffb0f05efece10": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "7e540474f87345a2b20e747c41891666": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "7e86abc7eb6a420f9626125e80c59fa3": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_d9b80e887fb640f88785a1d6c5a2819d",
        "IPY_MODEL_15c8381417da46fd8732b88ab8dee595"
       ],
       "layout": "IPY_MODEL_e937b2bd918f4d60ab9d8b9dad0221a2"
      }
     },
     "7effe6532ffe473a867168410cbc5212": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "7fd73b38fc0846f1bd0673401da237c1": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "description": "100%",
       "layout": "IPY_MODEL_bbffa377a9f544c4b5b84636b27744ab",
       "max": 79833,
       "style": "IPY_MODEL_b8cdb68fa70e429d92f2d740cf2963c6",
       "value": 79833
      }
     },
     "8073b6305678461b8095ad979adfd1dc": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_93ec47bb08b74f598d7a18f52a77356b",
        "IPY_MODEL_786295c26a6d48ed875b42cffae39b46"
       ],
       "layout": "IPY_MODEL_b087a5ebdf07452a8758b6c940dac309"
      }
     },
     "851ea17ce32747969731ada8fdfb2a50": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "87b6de3cc8ab4ba385d3ac4c055df8f6": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_6ca4a85a193440c3ac34b713b4ca8d24",
       "style": "IPY_MODEL_59ea1616b62a4be78ef8c7e4127b59b1",
       "value": " 0/? [00:00&lt;?, ?epoch/s]"
      }
     },
     "8b9a86b3d91748b9a634152b2b00daf5": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "success",
       "description": "100%",
       "layout": "IPY_MODEL_d365970cea5b424895ea70d166b6b442",
       "style": "IPY_MODEL_7e540474f87345a2b20e747c41891666",
       "value": 100
      }
     },
     "8bd2c262a7e64fd89547fa29e3b3157f": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "8c1d3cba83334cd99cc8c8b40763373a": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "layout": "IPY_MODEL_8dc967b28d5547c69bc8ed43b5a85ca6",
       "max": 1,
       "style": "IPY_MODEL_a70e8d95101e4f0a9473ee9b68d337a1"
      }
     },
     "8d1db6c684f6429cb8b7e34c2a3112a7": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_bad4301aa395443ab7a964727b9fe087",
       "style": "IPY_MODEL_6d128f06b4f54f2894802ab50a84c5a6",
       "value": " 40/40 [1:32:34&lt;00:00, 138.87s/epoch, val_loss=0.808, val_custom_metric=0.805, loss=0.8, custom_metric=0.798, lr=1e-5]"
      }
     },
     "8dc967b28d5547c69bc8ed43b5a85ca6": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "8e4b9be26771452c85589fd14172bdc9": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_2428340a39bb444db9c03195e95f3527",
       "style": "IPY_MODEL_1a94daae40504d3b95d3d3c427e059b0",
       "value": " 0/? [53:25&lt;?, ?batch/s]"
      }
     },
     "8fd2cb122618480faa14c1deeba453a5": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "911fa36f95804654ac5bbbd7bbf0873f": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_36f7f0504f8d499f8a45152bee479e50",
       "style": "IPY_MODEL_851ea17ce32747969731ada8fdfb2a50",
       "value": " 906.65625/906.65625 [13:32&lt;00:00,  1.00s/batch, loss=0.889, accuracy=0.583]"
      }
     },
     "9257d6425dd14ba28e73416cfa9cbce6": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "success",
       "description": "100%",
       "layout": "IPY_MODEL_fc27d7b3b5fa4c7d971df4b437fcbaf7",
       "max": 15,
       "style": "IPY_MODEL_7afda720af0948d981ff227786c307b5",
       "value": 15
      }
     },
     "9271d0067d434be9b02a311dbc689aa9": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "930a018bcdef4f8f8a8a826b8a9633ce": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "93ec47bb08b74f598d7a18f52a77356b": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "layout": "IPY_MODEL_b30eaab32c414ed8ae71705365100c10",
       "max": 1,
       "style": "IPY_MODEL_8fd2cb122618480faa14c1deeba453a5"
      }
     },
     "94e2b8bc0a414f4cb97a9cf6cad58cc4": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "94f3023761ea4b059d151d6f7617dae3": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_da3d1cc4ec6648669936383f7dee0b3a",
       "style": "IPY_MODEL_3c12c4ec17fe42cab2814917937265c4",
       "value": " 0/? [00:00&lt;?, ?epoch/s]"
      }
     },
     "96bf9a015fb2451890a11147f8d4a7de": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "description": "100%",
       "layout": "IPY_MODEL_faa4b586414b42688cff7306bbb9a794",
       "max": 80304,
       "style": "IPY_MODEL_d2b26f2b0aa2466f98d6353fdb80a31a",
       "value": 80304
      }
     },
     "97b78d399d2f4b248894e5de6699947f": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_3f28c888a66343e7af8f4a5f94b9ecd8",
       "style": "IPY_MODEL_dbffff5bc55e4b3a9f8cbeaa111308da",
       "value": " 20/40 [37:21&lt;36:58, 110.92s/epoch, val_loss=0.831, val_categorical_crossentropy=0.806, loss=0.82, categorical_crossentropy=0.798, lr=1e-04]"
      }
     },
     "99453bf82aea49a9b57d404657204db5": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "9b16de9986bf4f73bd5f895f1a423bad": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "layout": "IPY_MODEL_1637d459f7474829b3f32d3f90e077da",
       "max": 1,
       "style": "IPY_MODEL_417d9c2e71e942cfaed538f124c1eac0"
      }
     },
     "9e3e0901b6a44e6db8f19379f87e78e3": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "description": "100%",
       "layout": "IPY_MODEL_7b36f2197da4405fa4a8ffc1799b293d",
       "max": 79833,
       "style": "IPY_MODEL_f1a39469717041cbb9ad3ddb8cb848b0",
       "value": 79833
      }
     },
     "9e799a85d9f5429d822640cd9f6e517a": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_1234c7f4bdb54421b715049f732fdd2e",
       "style": "IPY_MODEL_3739ef941b994995a0a6c468de5b0ea4",
       "value": " 0/? [00:00&lt;?, ?batch/s]"
      }
     },
     "9fd938861837465b850bf524468e5ece": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "success",
       "description": "100%",
       "layout": "IPY_MODEL_08b16f147d034bed82707572d57d312f",
       "max": 40,
       "style": "IPY_MODEL_3f952be93e8d42d6a48621e03042f4c7",
       "value": 40
      }
     },
     "a0f277584e234abba36babec2b32a6e1": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_b9fd8728cb674658b330d38ecdc1d22a",
        "IPY_MODEL_7be5157b5e104f6eb8b37f3b95d16827"
       ],
       "layout": "IPY_MODEL_0053536bb734443d96fb28c5e1cf94f2"
      }
     },
     "a10fd773a9c44a668b0a1e4506f18fb8": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "a2f7ce3a82e9423a8bc77fff1d9b6f2d": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "a4918f5017724417a94e12569aa39879": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_dfeb1273138a44628a5401a04a1fe844",
       "style": "IPY_MODEL_07694115710d41d6b00cbb5b4a18162f",
       "value": " 77.9619140625/77.9619140625 [00:56&lt;00:00,  1.68batch/s, loss=1.5, categorical_crossentropy=1.09, categorical_accuracy=0.368, precision=0.485, recall=0.00517, precision_NEG=0, recall_NEG=0, precision_NEUT=0.487, recall_NEUT=0.0153, precision_POS=0.583, recall_POS=0.000263]"
      }
     },
     "a57920a4581043d8afe4aad4edd0d203": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "a5b291b178a84e4cb60fbecafa468e06": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "a5b98e04af964cc5982c8871f7259886": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "a70e8d95101e4f0a9473ee9b68d337a1": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "a7109ce369fe4fdabf3c5221ea89cf33": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "success",
       "description": "100%",
       "layout": "IPY_MODEL_4be9957348384a0e906669b33fee4e17",
       "max": 45,
       "style": "IPY_MODEL_d7d069c6d15b4b6980f0a379fc5b3e72",
       "value": 45
      }
     },
     "a8ed6d2bd10b460690a2095103b658c8": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "ae566d478719492c94bf8ac7c66c7ff3": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_cdc7b85b467048578342a887c6c0bfbf",
       "style": "IPY_MODEL_bff80784bd9b45559b8d3d281848d061",
       "value": " 156.84375/156.84375 [02:45&lt;00:00,  1.12batch/s, loss=0.94, perf_metric=0.931]"
      }
     },
     "b087a5ebdf07452a8758b6c940dac309": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "b179d17b356d4428864c8d25faaf9ecc": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "description": " 50%",
       "layout": "IPY_MODEL_930a018bcdef4f8f8a8a826b8a9633ce",
       "max": 40,
       "style": "IPY_MODEL_22021a05d97547ceaec3c04d6bcf1af0",
       "value": 20
      }
     },
     "b30eaab32c414ed8ae71705365100c10": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "b766d80f7dc94e58b3f348ed74331487": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "b8cdb68fa70e429d92f2d740cf2963c6": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "b933a0490ae14e8bab56b30a4e25698f": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_f6e8228fda4e4ae4951b6b80c4ef5ec3",
       "style": "IPY_MODEL_a5b98e04af964cc5982c8871f7259886",
       "value": " 100/100 [4:33:55&lt;00:00, 164.36s/epoch, val_loss=0.879, val_perf_metric=0.879, loss=0.931, perf_metric=0.931, lr=1e-6]"
      }
     },
     "b9fd8728cb674658b330d38ecdc1d22a": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "success",
       "description": "100%",
       "layout": "IPY_MODEL_cabe6793caf84da893bf9b8e871efcd5",
       "max": 15,
       "style": "IPY_MODEL_1a4659818d2e4e2ba58363ea98511975",
       "value": 15
      }
     },
     "bad4301aa395443ab7a964727b9fe087": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "bbffa377a9f544c4b5b84636b27744ab": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "be7b069acfc941d1a5adb7ad0d04fcef": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "description": "100%",
       "layout": "IPY_MODEL_76f21179eb01441fa3c84a777164f25d",
       "max": 79833,
       "style": "IPY_MODEL_2b787b4ed2204046924e52c1e4bd2af0",
       "value": 79833
      }
     },
     "bff80784bd9b45559b8d3d281848d061": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "c88f4a593f7140e98f1abc0bdb01d7c6": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "ca100a72a9524aac9124d18fbb68b8f3": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "cabe6793caf84da893bf9b8e871efcd5": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "cdc7b85b467048578342a887c6c0bfbf": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "d1930ec03ede48e698f389aefcb9d623": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "d1b4507fd1044b7f9ded72830a4b7ff0": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "d2b26f2b0aa2466f98d6353fdb80a31a": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "d365970cea5b424895ea70d166b6b442": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "d580c2f7a8f942c18d913da7d52e4678": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "d5d6b5f1e9424f93aeca8414ed8018e5": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "d7d069c6d15b4b6980f0a379fc5b3e72": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "d9b80e887fb640f88785a1d6c5a2819d": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "success",
       "description": "100%",
       "layout": "IPY_MODEL_d5d6b5f1e9424f93aeca8414ed8018e5",
       "max": 60,
       "style": "IPY_MODEL_383bf6356ffd4599a6f50dda1877df16",
       "value": 60
      }
     },
     "da3d1cc4ec6648669936383f7dee0b3a": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "da46e6eeb3e545ccbdd9fbad6db16c26": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "da61b5d67a8b4eadbd017316034e7e2b": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "dbf0f4e908c945e5962ef7656bbd880c": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "description": "100%",
       "layout": "IPY_MODEL_3a7ff25f4ec94b48be34130d5dc535ef",
       "max": 58676,
       "style": "IPY_MODEL_94e2b8bc0a414f4cb97a9cf6cad58cc4",
       "value": 58676
      }
     },
     "dbffff5bc55e4b3a9f8cbeaa111308da": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "ddace66a14f147bc8c511b2a2f2274ff": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "ddf3afb8bbc8447b805dc1eb96e02d0b": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "dfeb1273138a44628a5401a04a1fe844": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "e0833f1a0114461aa6f5b949f61365d4": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "DescriptionStyleModel",
      "state": {
       "description_width": ""
      }
     },
     "e321ede022d641a5832189da180e540f": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_fb52218b623642eb9335584d24ed7fd4",
        "IPY_MODEL_5f44e005650146f1bde873b2d3c19ca7"
       ],
       "layout": "IPY_MODEL_639e0ad22a8e422486c0ddda7f763173"
      }
     },
     "e50788e484274ac9a669b593577e0252": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "info",
       "layout": "IPY_MODEL_d1930ec03ede48e698f389aefcb9d623",
       "max": 1,
       "style": "IPY_MODEL_44ca446e5cea46e3bb1ffe4be027a823"
      }
     },
     "e67f04a7196f43589a89eef3383d94d4": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_b179d17b356d4428864c8d25faaf9ecc",
        "IPY_MODEL_97b78d399d2f4b248894e5de6699947f"
       ],
       "layout": "IPY_MODEL_4e26db0ff18545a5a5bb09c22b52b74f"
      }
     },
     "e687c692f20d498bbcc2c79b1d49dbce": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "e937b2bd918f4d60ab9d8b9dad0221a2": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "e971db6680bf409995bc543a71d04cd9": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HBoxModel",
      "state": {
       "children": [
        "IPY_MODEL_a7109ce369fe4fdabf3c5221ea89cf33",
        "IPY_MODEL_2cba71c444da48b28967b65248a93756"
       ],
       "layout": "IPY_MODEL_3338e3eb86804811989376acb411503c"
      }
     },
     "e9747afbad43453883ae05fcbbb36828": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "f1635d1388444e37a1fd90f79bf23565": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_222a29c01214468a903621bd5356f4e3",
       "style": "IPY_MODEL_7effe6532ffe473a867168410cbc5212",
       "value": " 77.9619140625/77.9619140625 [00:57&lt;00:00,  1.85batch/s, loss=1.59, categorical_crossentropy=1.17, categorical_accuracy=0.331, precision_2=0.288, recall_2=0.0783]"
      }
     },
     "f1a39469717041cbb9ad3ddb8cb848b0": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "f6e8228fda4e4ae4951b6b80c4ef5ec3": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "f9be9fc4e7d94d17820a1f067335b54b": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "HTMLModel",
      "state": {
       "layout": "IPY_MODEL_ca100a72a9524aac9124d18fbb68b8f3",
       "style": "IPY_MODEL_9271d0067d434be9b02a311dbc689aa9",
       "value": " 155.923828125/155.923828125 [02:46&lt;00:00,  1.10batch/s, loss=0.908, categorical_crossentropy=0.942, categorical_accuracy=0.545, precision_3=0.63, recall_3=0.364]"
      }
     },
     "fa727a6fb2224d42ad0d073477eaaa37": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "description_width": "initial"
      }
     },
     "faa4b586414b42688cff7306bbb9a794": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "fb52218b623642eb9335584d24ed7fd4": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "1.5.0",
      "model_name": "FloatProgressModel",
      "state": {
       "bar_style": "success",
       "layout": "IPY_MODEL_69a5320d83834adc84b6e1fabcbd2cf4",
       "max": 1,
       "style": "IPY_MODEL_a8ed6d2bd10b460690a2095103b658c8"
      }
     },
     "fc27d7b3b5fa4c7d971df4b437fcbaf7": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     },
     "feb468cecb424b599a08e5ea0c59c6c0": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "1.2.0",
      "model_name": "LayoutModel",
      "state": {}
     }
    },
    "version_major": 2,
    "version_minor": 0
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}