run.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "== Cuda: True ===\n"
     ]
    }
   ],
   "source": [
    "import sys\n",
    "sys.path.insert(1, './src')\n",
    "\n",
    "from tagger import CrfModel\n",
    "from data_loader import BuildDataLoader\n",
    "from environment import LabelEnv\n",
    "from agent import ParamRNN, AgentParamRNN\n",
    "\n",
    "from gensim.models import KeyedVectors\n",
    "import numpy as np\n",
    "import random\n",
    "import argparse\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "import pickle\n",
    "\n",
    "import warnings; warnings.simplefilter('ignore')\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "import torch.nn.functional as F\n",
    "import torchvision.transforms as T\n",
    "from torch.autograd import Variable  \n",
    "\n",
    "# if gpu is to be used\n",
    "device = torch.device('cuda' if torch.cuda.is_available() else \"cpu\")\n",
    "print (\"== Cuda: {} ===\".format(torch.cuda.is_available()))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "=== data loading ===\n",
      "word dict size: 25\n",
      "label dict size: 126\n",
      "Replace digit with NUM: True\n",
      "use embedding: False\n",
      "embedding size: 25\n",
      "number of instances: 1335\n",
      "=== Tagger initialized ===\n",
      "-- label dict size: 126\n",
      "-- word dict size: 25\n"
     ]
    }
   ],
   "source": [
    "SOURCE = 'sod'\n",
    "DATA_PATH = \"./datasets/\" + SOURCE\n",
    "NUM_FLAG = True\n",
    "EMBED_FLAG = False\n",
    "FEAT = 'all'\n",
    "\n",
    "dataloader = BuildDataLoader(SOURCE, DATA_PATH, NUM_FLAG, EMBED_FLAG)\n",
    "crf = CrfModel(dataloader, FEAT)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "max_len is: 14\n",
      "crf para size: (201, 126)\n",
      "=== Env: data ===\n",
      "train     : 600\n",
      "validation: 200\n",
      "test      : 200\n",
      "=== Env: setup ===\n",
      "reward reweight: valid2Tx\n",
      "similarity for valid to test done -- size=(200,)\n",
      "budget: 2\n",
      "=== Agent: created\n",
      "Q-net parameter size: 46769\n",
      ">>>> Start play\n",
      "cost 2: queried 382 with greedy 0:-0.3412691354751587, acc=(0.6507142857142857, 0.6660714285714284)\n"
     ]
    }
   ],
   "source": [
    "samples = [4]\n",
    "VALID_N = 200\n",
    "TEST_N = 200\n",
    "TRAIN_N = 600\n",
    "REWEIGHT = 'valid2Tx'\n",
    "GEEDY = 'te'\n",
    "BUDGET = 2\n",
    "max_len = dataloader.get_max_len()\n",
    "embedding_size = dataloader.get_embed_size()\n",
    "parameter_shape = crf.get_para_shape()\n",
    "print (\"max_len is: {}\".format(max_len))\n",
    "print (\"crf para size: {}\".format(parameter_shape))\n",
    "\n",
    "qvalue_list = []\n",
    "action_mark_list = []\n",
    "prob_list = []\n",
    "for seed in samples:\n",
    "    env = LabelEnv(dataloader, crf, seed, VALID_N, TEST_N, TRAIN_N, REWEIGHT, BUDGET)\n",
    "    agent = AgentParamRNN(GEEDY, max_len, embedding_size, parameter_shape)\n",
    "    \n",
    "    print (\">>>> Start play\")\n",
    "    step = 0\n",
    "    while env.cost < BUDGET:\n",
    "        env.resume()\n",
    "        observation = env.get_state()\n",
    "        observ = [observation[0], observation[1], observation[3], observation[4], observation[5]] \n",
    "        greedy_flg, action, q_value = agent.get_action(observ)\n",
    "        reward, observation2, terminal = env.feedback(action)\n",
    "        (acc_test, acc_valid) = env.eval_tagger()\n",
    "        print (\"cost {}: queried {} with greedy {}:{}, acc=({}, {})\".format(env.cost, action, greedy_flg, \n",
    "                                                                            q_value.item(), acc_test, acc_valid))\n",
    "        \n",
    "        qvalue_list.append(q_value.item())\n",
    "        action_mark_list.append(greedy_flg)\n",
    "        prob_list.append(observation[1][action])\n",
    "        \n",
    "        if env.cost % 10 == 0:\n",
    "            step += 10\n",
    "        \n",
    "        if env.cost < 3:\n",
    "            continue\n",
    "            \n",
    "        for n in range(20 + step):\n",
    "            env.reboot()\n",
    "            while env.terminal == False:\n",
    "                observation = env.get_state()\n",
    "                observ = [observation[0], observation[1], observation[3], observation[4], observation[5]] \n",
    "                greedy_flg, action, q_value = agent.get_action(observ)\n",
    "                reward, observation2, terminal = env.feedback(action)\n",
    "                observ2 = [observation2[0], observation2[1], observation2[3], observation2[4], observation2[5]]\n",
    "                agent.update(observ, action, reward, observ2, terminal)\n",
    "            "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "cost_list = sorted(env.acc_trace.keys())\n",
    "acc_list = [env.acc_trace[i][0] for i in cost_list]\n",
    "acc_valid_list = [env.acc_trace[i][1] for i in cost_list]\n",
    "\n",
    "num = \"num\" if NUM_FLAG else \"\"\n",
    "emb = \"embed\" if EMBED_FLAG else \"\"\n",
    "filename = \"./results_run/\" + SOURCE + num + emb + \"_\" + str(BUDGET) + \"bgt_\" + GEEDY + \"_\" + REWEIGHT + \"_\" + FEAT\n",
    "\n",
    "with open(filename + \".bin\", \"wb\") as result:\n",
    "    pickle.dump((cost_list, acc_list, acc_valid_list), result)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[-0.3412691354751587] [0] [0.8046558188143749]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print (qvalue_list, action_mark_list, prob_list)\n",
    "# ======================================== sanity check =====================================================\n",
    "x = range(len(qvalue_list))\n",
    "\n",
    "fig, axes = plt.subplots(ncols=2, nrows=1)\n",
    "ax = axes.flatten()\n",
    "\n",
    "ax[0].plot(x, qvalue_list, color='0.5')\n",
    "x2 = []\n",
    "y2 = []\n",
    "for i in x:\n",
    "    if action_mark_list[i] == 1:\n",
    "        x2.append(i)\n",
    "        y2.append(qvalue_list[i])\n",
    "l1 = ax[0].scatter(x2, y2, color='r', marker='o')\n",
    "x2 = []\n",
    "y2 = []\n",
    "for i in x:\n",
    "    if action_mark_list[i] == 0:\n",
    "        x2.append(i)\n",
    "        y2.append(qvalue_list[i])\n",
    "l2 = ax[0].scatter(x2, y2, color='g', marker='x')\n",
    "ax[0].legend((l1,l2),\n",
    "           ('$action > \\\\epsilon$', '$action < \\\\epsilon$'))\n",
    "\n",
    "ax[0].set_title('{} dataset'.format(SOURCE))\n",
    "# plt.xlim(0, 20)\n",
    "ax[0].set_ylabel('Q value')\n",
    "ax[0].set_xlabel('Sequence number')\n",
    "\n",
    "# print (qvalue_list)\n",
    "ax[1].scatter(qvalue_list, prob_list)\n",
    "for i in range(len(qvalue_list)):\n",
    "    ax[1].annotate(i, (qvalue_list[i], prob_list[i]))\n",
    "ax[1].set_xlabel('Q value')\n",
    "ax[1].set_ylabel('Likelihood')\n",
    "plt.subplots_adjust(wspace=0.1)\n",
    "fig.set_size_inches(15,5)\n",
    "# plt.show()\n",
    "plt.savefig(filename + '_check.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# sim beta\n",
    "with open(\"./results/conll15_200_75budget_fully_none.bin\", \"rb\") as in_file:\n",
    "    (cost_1, _, _, all_acc_1, acc_valid_1) = pickle.load(in_file)\n",
    "#     (cost_1, all_acc_1, acc_valid_1) = pickle.load(in_file)\n",
    "       \n",
    "with open(\"./results/conllnumembed_75bgt_te_valid2V_all.bin\", \"rb\") as in_file:\n",
    "    (cost_2, all_acc_2, acc_valid_2) = pickle.load(in_file)\n",
    "\n",
    "with open(\"./results/conll30numembed_200_100budget_valid2VRL_all_te_acc_20step_5batch20decay_100loop-add10each10step-3_64rnn_16filter_4size_2stride.bin\", \"rb\") as in_file:\n",
    "    (cost_3, all_acc_3, acc_valid_3) = pickle.load(in_file)\n",
    "\n",
    "with open(\"./results/conll30numembed_200_100budget_valid2VRLtrellis_all_te_acc_20step_5batch20decay_100loop-add10each10step-3_64rnn_16filter_4size_2stride.bin\", \"rb\") as in_file:\n",
    "    (cost_4, all_acc_4, acc_valid_4) = pickle.load(in_file)\n",
    "    \n",
    "with open(\"./results/CNNDQN-conll-20.bin\", \"rb\") as in_file:\n",
    "    (cost_5, all_acc_5, acc_valid_5) = pickle.load(in_file)\n",
    "\n",
    "# with open(\"./results/sod15num_200_testVRL_acc2_all_rand_20step_5batch20decay_90loopdecay_64rnn_16filter_4size_2stride.bin\", \"rb\") as in_file:\n",
    "#     (cost_6, in_acc_6, out_acc_6, all_acc_6, acc_valid_6) = pickle.load(in_file)\n",
    "\n",
    "# with open(\"./results/sod15num_200_testVRL_acc2_all_te_20step_5batch20decay_90loopdecay_64rnn_16filter_4size_2stride.bin\", \"rb\") as in_file:\n",
    "#     (cost_7, in_acc_7, out_acc_7, all_acc_7, acc_valid_7) = pickle.load(in_file)\n",
    "\n",
    "plt.plot(cost_1, all_acc_1,\n",
    "         cost_2, all_acc_2,\n",
    "         cost_3, all_acc_3,\n",
    "         cost_4, all_acc_4,\n",
    "         cost_5, all_acc_5,)\n",
    "#          cost_6, all_acc_6,)\n",
    "#          cost_7, acc_valid_7)\n",
    "plt.legend(['TE', 'curRL', 'RL_pretrain30', 'RL_pretrain30_trellis', 'PAL' ,'RL-test', 'RL-valid'], loc='lower right', fancybox=True, fontsize = 9)\n",
    "# plt.xlim(200, 600)\n",
    "\n",
    "plt.title('CoNLL dataset')\n",
    "plt.xlabel('Number of training sequences')\n",
    "plt.ylabel('Prediction accuracy')\n",
    "plt.savefig('./results/sod.png', bbox_inches='tight')\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python3.6",
   "language": "python",
   "name": "ll5fy36"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}