finding_donors.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 机器学习纳米学位\n",
    "## 监督学习\n",
    "## 项目2: 为*CharityML*寻找捐献者"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "欢迎来到机器学习工程师纳米学位的第二个项目！在此文件中，有些示例代码已经提供给你，但你还需要实现更多的功能让项目成功运行。除非有明确要求，你无须修改任何已给出的代码。以**'练习'**开始的标题表示接下来的代码部分中有你必须要实现的功能。每一部分都会有详细的指导，需要实现的部分也会在注释中以'TODO'标出。请仔细阅读所有的提示！\n",
    "\n",
    "除了实现代码外，你还必须回答一些与项目和你的实现有关的问题。每一个需要你回答的问题都会以**'问题 X'**为标题。请仔细阅读每个问题，并且在问题后的**'回答'**文字框中写出完整的答案。我们将根据你对问题的回答和撰写代码所实现的功能来对你提交的项目进行评分。\n",
    ">**提示：**Code 和 Markdown 区域可通过**Shift + Enter**快捷键运行。此外，Markdown可以通过双击进入编辑模式。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 开始\n",
    "\n",
    "在这个项目中，你将使用1994年美国人口普查收集的数据，选用几个监督学习算法以准确地建模被调查者的收入。然后，你将根据初步结果从中选择出最佳的候选算法，并进一步优化该算法以最好地建模这些数据。你的目标是建立一个能够准确地预测被调查者年收入是否超过50000美元的模型。这种类型的任务会出现在那些依赖于捐款而存在的非营利性组织。了解人群的收入情况可以帮助一个非营利性的机构更好地了解他们要多大的捐赠，或是否他们应该接触这些人。虽然我们很难直接从公开的资源中推断出一个人的一般收入阶层，但是我们可以（也正是我们将要做的）从其他的一些公开的可获得的资源中获得一些特征从而推断出该值。\n",
    "\n",
    "这个项目的数据集来自[UCI机器学习知识库](https://archive.ics.uci.edu/ml/datasets/Census+Income)。这个数据集是由Ron Kohavi和Barry Becker在发表文章_\"Scaling Up the Accuracy of Naive-Bayes Classifiers: A Decision-Tree Hybrid\"_之后捐赠的，你可以在Ron Kohavi提供的[在线版本](https://www.aaai.org/Papers/KDD/1996/KDD96-033.pdf)中找到这个文章。我们在这里探索的数据集相比于原有的数据集有一些小小的改变，比如说移除了特征`'fnlwgt'` 以及一些遗失的或者是格式不正确的记录。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "## 探索数据\n",
    "运行下面的代码单元以载入需要的Python库并导入人口普查数据。注意数据集的最后一列`'income'`将是我们需要预测的列（表示被调查者的年收入会大于或者是最多50,000美元），人口普查数据中的每一列都将是关于被调查者的特征。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>age</th>\n",
       "      <th>workclass</th>\n",
       "      <th>education_level</th>\n",
       "      <th>education-num</th>\n",
       "      <th>marital-status</th>\n",
       "      <th>occupation</th>\n",
       "      <th>relationship</th>\n",
       "      <th>race</th>\n",
       "      <th>sex</th>\n",
       "      <th>capital-gain</th>\n",
       "      <th>capital-loss</th>\n",
       "      <th>hours-per-week</th>\n",
       "      <th>native-country</th>\n",
       "      <th>income</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>39</td>\n",
       "      <td>State-gov</td>\n",
       "      <td>Bachelors</td>\n",
       "      <td>13.0</td>\n",
       "      <td>Never-married</td>\n",
       "      <td>Adm-clerical</td>\n",
       "      <td>Not-in-family</td>\n",
       "      <td>White</td>\n",
       "      <td>Male</td>\n",
       "      <td>2174.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>40.0</td>\n",
       "      <td>United-States</td>\n",
       "      <td>&lt;=50K</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   age   workclass education_level  education-num  marital-status  \\\n",
       "0   39   State-gov       Bachelors           13.0   Never-married   \n",
       "\n",
       "      occupation    relationship    race    sex  capital-gain  capital-loss  \\\n",
       "0   Adm-clerical   Not-in-family   White   Male        2174.0           0.0   \n",
       "\n",
       "   hours-per-week  native-country income  \n",
       "0            40.0   United-States  <=50K  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 为这个项目导入需要的库\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from time import time\n",
    "from IPython.display import display # 允许为DataFrame使用display()\n",
    "\n",
    "# 导入附加的可视化代码visuals.py\n",
    "import visuals as vs\n",
    "\n",
    "# 为notebook提供更加漂亮的可视化\n",
    "%matplotlib inline\n",
    "\n",
    "# 导入人口普查数据\n",
    "data = pd.read_csv(\"census.csv\")\n",
    "\n",
    "# 成功 - 显示第一条记录\n",
    "display(data.head(n=1))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 练习：数据探索\n",
    "首先我们对数据集进行一个粗略的探索，我们将看看每一个类别里会有多少被调查者？并且告诉我们这些里面多大比例是年收入大于50,000美元的。在下面的代码单元中，你将需要计算以下量：\n",
    "\n",
    "- 总的记录数量，`'n_records'`\n",
    "- 年收入大于50,000美元的人数，`'n_greater_50k'`.\n",
    "- 年收入最多为50,000美元的人数 `'n_at_most_50k'`.\n",
    "- 年收入大于50,000美元的人所占的比例， `'greater_percent'`.\n",
    "\n",
    "**提示：** 您可能需要查看上面的生成的表，以了解`'income'`条目的格式是什么样的。 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total number of records: 45222\n",
      "Individuals making more than $50,000: 11208\n",
      "Individuals making at most $50,000: 34014\n",
      "Percentage of individuals making more than $50,000: 24.78%\n"
     ]
    }
   ],
   "source": [
    "# TODO：总的记录数\n",
    "n_records = len(data)\n",
    "\n",
    "# TODO：被调查者的收入大于$50,000的人数\n",
    "n_greater_50k = len(data[data.income == '>50K'])\n",
    "\n",
    "# TODO：被调查者的收入最多为$50,000的人数\n",
    "n_at_most_50k = len(data[data.income == '<=50K'])\n",
    "\n",
    "# TODO：被调查者收入大于$50,000所占的比例\n",
    "greater_percent = float(n_greater_50k) / n_records * 100\n",
    "\n",
    "# 打印结果\n",
    "print (\"Total number of records: {}\".format(n_records))\n",
    "print (\"Individuals making more than $50,000: {}\".format(n_greater_50k))\n",
    "print (\"Individuals making at most $50,000: {}\".format(n_at_most_50k))\n",
    "print (\"Percentage of individuals making more than $50,000: {:.2f}%\".format(greater_percent))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "## 准备数据\n",
    "在数据能够被作为输入提供给机器学习算法之前，它经常需要被清洗，格式化，和重新组织 - 这通常被叫做**预处理**。幸运的是，对于这个数据集，没有我们必须处理的无效或丢失的条目，然而，由于某一些特征存在的特性我们必须进行一定的调整。这个预处理都可以极大地帮助我们提升几乎所有的学习算法的结果和预测能力。\n",
    "\n",
    "### 获得特征和标签\n",
    "`income` 列是我们需要的标签，记录一个人的年收入是否高于50K。 因此我们应该把他从数据中剥离出来，单独存放。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 将数据切分成特征和对应的标签\n",
    "income_raw = data['income']\n",
    "features_raw = data.drop('income', axis = 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 转换倾斜的连续特征\n",
    "\n",
    "一个数据集有时可能包含至少一个靠近某个数字的特征，但有时也会有一些相对来说存在极大值或者极小值的不平凡分布的的特征。算法对这种分布的数据会十分敏感，并且如果这种数据没有能够很好地规一化处理会使得算法表现不佳。在人口普查数据集的两个特征符合这个描述：'`capital-gain'`和`'capital-loss'`。\n",
    "\n",
    "运行下面的代码单元以创建一个关于这两个特征的条形图。请注意当前的值的范围和它们是如何分布的。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbbb0e6f0b8>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 可视化 'capital-gain'和'capital-loss' 两个特征\n",
    "vs.distribution(features_raw)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "对于高度倾斜分布的特征如`'capital-gain'`和`'capital-loss'`，常见的做法是对数据施加一个<a href=\"https://en.wikipedia.org/wiki/Data_transformation_(statistics)\">对数转换</a>，将数据转换成对数，这样非常大和非常小的值不会对学习算法产生负面的影响。并且使用对数变换显著降低了由于异常值所造成的数据范围异常。但是在应用这个变换时必须小心：因为0的对数是没有定义的，所以我们必须先将数据处理成一个比0稍微大一点的数以成功完成对数转换。\n",
    "\n",
    "运行下面的代码单元来执行数据的转换和可视化结果。再次，注意值的范围和它们是如何分布的。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbbae97b358>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 对于倾斜的数据使用Log转换\n",
    "skewed = ['capital-gain', 'capital-loss']\n",
    "features_raw[skewed] = data[skewed].apply(lambda x: np.log(x + 1))\n",
    "\n",
    "# 可视化对数转换后 'capital-gain'和'capital-loss' 两个特征\n",
    "vs.distribution(features_raw, transformed = True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 规一化数字特征\n",
    "除了对于高度倾斜的特征施加转换，对数值特征施加一些形式的缩放通常会是一个好的习惯。在数据上面施加一个缩放并不会改变数据分布的形式（比如上面说的'capital-gain' or 'capital-loss'）；但是，规一化保证了每一个特征在使用监督学习器的时候能够被平等的对待。注意一旦使用了缩放，观察数据的原始形式不再具有它本来的意义了，就像下面的例子展示的。\n",
    "\n",
    "运行下面的代码单元来规一化每一个数字特征。我们将使用[`sklearn.preprocessing.MinMaxScaler`](http://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.MinMaxScaler.html)来完成这个任务。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>age</th>\n",
       "      <th>workclass</th>\n",
       "      <th>education_level</th>\n",
       "      <th>education-num</th>\n",
       "      <th>marital-status</th>\n",
       "      <th>occupation</th>\n",
       "      <th>relationship</th>\n",
       "      <th>race</th>\n",
       "      <th>sex</th>\n",
       "      <th>capital-gain</th>\n",
       "      <th>capital-loss</th>\n",
       "      <th>hours-per-week</th>\n",
       "      <th>native-country</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.30137</td>\n",
       "      <td>State-gov</td>\n",
       "      <td>Bachelors</td>\n",
       "      <td>0.8</td>\n",
       "      <td>Never-married</td>\n",
       "      <td>Adm-clerical</td>\n",
       "      <td>Not-in-family</td>\n",
       "      <td>White</td>\n",
       "      <td>Male</td>\n",
       "      <td>0.02174</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.397959</td>\n",
       "      <td>United-States</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       age   workclass education_level  education-num  marital-status  \\\n",
       "0  0.30137   State-gov       Bachelors            0.8   Never-married   \n",
       "\n",
       "      occupation    relationship    race    sex  capital-gain  capital-loss  \\\n",
       "0   Adm-clerical   Not-in-family   White   Male       0.02174           0.0   \n",
       "\n",
       "   hours-per-week  native-country  \n",
       "0        0.397959   United-States  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.preprocessing import MinMaxScaler\n",
    "\n",
    "# 初始化一个 scaler，并将它施加到特征上\n",
    "scaler = MinMaxScaler()\n",
    "numerical = ['age', 'education-num', 'capital-gain', 'capital-loss', 'hours-per-week']\n",
    "features_raw[numerical] = scaler.fit_transform(data[numerical])\n",
    "\n",
    "# 显示一个经过缩放的样例记录\n",
    "display(features_raw.head(n = 1))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 练习：数据预处理\n",
    "\n",
    "从上面的**数据探索**中的表中，我们可以看到有几个属性的每一条记录都是非数字的。通常情况下，学习算法期望输入是数字的，这要求非数字的特征（称为类别变量）被转换。转换类别变量的一种流行的方法是使用**独热编码**方案。独热编码为每一个非数字特征的每一个可能的类别创建一个_“虚拟”_变量。例如，假设`someFeature`有三个可能的取值`A`，`B`或者`C`，。我们将把这个特征编码成`someFeature_A`, `someFeature_B`和`someFeature_C`.\n",
    "\n",
    "| 特征X |                    | 特征X_A | 特征X_B | 特征X_C |\n",
    "| :-: |                            | :-: | :-: | :-: |\n",
    "|  B  |  | 0 | 1 | 0 |\n",
    "|  C  | ----> 独热编码 ----> | 0 | 0 | 1 |\n",
    "|  A  |  | 1 | 0 | 0 |\n",
    "\n",
    "此外，对于非数字的特征，我们需要将非数字的标签`'income'`转换成数值以保证学习算法能够正常工作。因为这个标签只有两种可能的类别（\"<=50K\"和\">50K\"），我们不必要使用独热编码，可以直接将他们编码分别成两个类`0`和`1`，在下面的代码单元中你将实现以下功能：\n",
    " - 使用[`pandas.get_dummies()`](http://pandas.pydata.org/pandas-docs/stable/generated/pandas.get_dummies.html?highlight=get_dummies#pandas.get_dummies)对`'features_raw'`数据来施加一个独热编码。\n",
    " - 将目标标签`'income_raw'`转换成数字项。\n",
    "   - 将\"<=50K\"转换成`0`；将\">50K\"转换成`1`。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "103 total features after one-hot encoding.\n"
     ]
    }
   ],
   "source": [
    "# TODO：使用pandas.get_dummies()对'features_raw'数据进行独热编码\n",
    "features = pd.get_dummies(features_raw)\n",
    "\n",
    "# TODO：将'income_raw'编码成数字值\n",
    "income = income_raw.apply(lambda x: 1 if x == '>50K' else 0)\n",
    "\n",
    "# 打印经过独热编码之后的特征数量\n",
    "encoded = list(features.columns)\n",
    "print (\"{} total features after one-hot encoding.\".format(len(encoded)))\n",
    "\n",
    "# 移除下面一行的注释以观察编码的特征名字\n",
    "#print encoded"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 混洗和切分数据\n",
    "现在所有的 _类别变量_ 已被转换成数值特征，而且所有的数值特征已被规一化。和我们一般情况下做的一样，我们现在将数据（包括特征和它们的标签）切分成训练和测试集。其中80%的数据将用于训练和20%的数据用于测试。然后再进一步把训练数据分为训练集和验证集，用来选择和优化模型。\n",
    "\n",
    "运行下面的代码单元来完成切分。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training set has 28941 samples.\n",
      "Validation set has 7236 samples.\n",
      "Testing set has 9045 samples.\n"
     ]
    }
   ],
   "source": [
    "# 导入 train_test_split\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "# 将'features'和'income'数据切分成训练集和测试集\n",
    "X_train, X_test, y_train, y_test = train_test_split(features, income, test_size = 0.2, random_state = 0,\n",
    "                                                    stratify = income)\n",
    "# 将'X_train'和'y_train'进一步切分为训练集和验证集\n",
    "X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size=0.2, random_state=0,\n",
    "                                                    stratify = y_train)\n",
    "\n",
    "# 显示切分的结果\n",
    "print (\"Training set has {} samples.\".format(X_train.shape[0]))\n",
    "print (\"Validation set has {} samples.\".format(X_val.shape[0]))\n",
    "print (\"Testing set has {} samples.\".format(X_test.shape[0]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "## 评价模型性能\n",
    "在这一部分中，我们将尝试四种不同的算法，并确定哪一个能够最好地建模数据。四种算法包含一个*天真的预测器* 和三个你选择的监督学习器。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 评价方法和朴素的预测器\n",
    "*CharityML*通过他们的研究人员知道被调查者的年收入大于\\$50,000最有可能向他们捐款。因为这个原因*CharityML*对于准确预测谁能够获得\\$50,000以上收入尤其有兴趣。这样看起来使用**准确率**作为评价模型的标准是合适的。另外，把*没有*收入大于\\$50,000的人识别成年收入大于\\$50,000对于*CharityML*来说是有害的，因为他想要找到的是有意愿捐款的用户。这样，我们期望的模型具有准确预测那些能够年收入大于\\$50,000的能力比模型去**查全**这些被调查者*更重要*。我们能够使用**F-beta score**作为评价指标，这样能够同时考虑查准率和查全率：\n",
    "\n",
    "$$ F_{\\beta} = (1 + \\beta^2) \\cdot \\frac{precision \\cdot recall}{\\left( \\beta^2 \\cdot precision \\right) + recall} $$\n",
    "\n",
    "\n",
    "尤其是，当 $\\beta = 0.5$ 的时候更多的强调查准率，这叫做**F$_{0.5}$ score** （或者为了简单叫做F-score）。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 问题 1 - 天真的预测器的性能\n",
    "\n",
    "通过查看收入超过和不超过 \\$50,000 的人数，我们能发现多数被调查者年收入没有超过 \\$50,000。如果我们简单地预测说*“这个人的收入没有超过 \\$50,000”*，我们就可以得到一个 准确率超过 50% 的预测。这样我们甚至不用看数据就能做到一个准确率超过 50%。这样一个预测被称作是天真的。通常对数据使用一个*天真的预测器*是十分重要的，这样能够帮助建立一个模型表现是否好的基准。 使用下面的代码单元计算天真的预测器的相关性能。将你的计算结果赋值给`'accuracy'`, `‘precision’`, `‘recall’` 和 `'fscore'`，这些值会在后面被使用，请注意这里不能使用scikit-learn，你需要根据公式自己实现相关计算。\n",
    "\n",
    "*如果我们选择一个无论什么情况都预测被调查者年收入大于 \\$50,000 的模型，那么这个模型在**验证集上**的准确率，查准率，查全率和 F-score是多少？*  \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Naive Predictor on validation data: \n",
      "     Accuracy score: 0.2478 \n",
      "     Precision: 0.2478 \n",
      "     Recall: 1.0000 \n",
      "     F-score: 0.2917\n"
     ]
    }
   ],
   "source": [
    "#不能使用scikit-learn，你需要根据公式自己实现相关计算。\n",
    "TP = y_val[y_val == 1].size\n",
    "FP = y_val[y_val == 0].size\n",
    "FN = 0\n",
    "\n",
    "#TODO： 计算准确率\n",
    "accuracy = float(TP) / float(y_val.size)\n",
    "\n",
    "# TODO： 计算查准率 Precision\n",
    "precision = float(TP) / (float(TP) + float(FP))\n",
    "\n",
    "# TODO： 计算查全率 Recall\n",
    "recall = float(TP) / (float(TP) + float(FN))\n",
    "\n",
    "# TODO： 使用上面的公式，设置beta=0.5，计算F-score\n",
    "beta = 0.5\n",
    "fscore = (1 + np.square(beta)) * (precision * recall) / ((np.square(beta) * precision) + recall)\n",
    "\n",
    "# 打印结果\n",
    "print (\"Naive Predictor on validation data: \\n \\\n",
    "    Accuracy score: {:.4f} \\n \\\n",
    "    Precision: {:.4f} \\n \\\n",
    "    Recall: {:.4f} \\n \\\n",
    "    F-score: {:.4f}\".format(accuracy, precision, recall, fscore))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 监督学习模型\n",
    "### 问题 2 - 模型应用\n",
    "\n",
    "你能够在 [`scikit-learn`](http://scikit-learn.org/stable/supervised_learning.html) 中选择以下监督学习模型\n",
    "- 高斯朴素贝叶斯 (GaussianNB)\n",
    "- 决策树 (DecisionTree)\n",
    "- 集成方法 (Bagging, AdaBoost, Random Forest, Gradient Boosting)\n",
    "- K近邻 (K Nearest Neighbors)\n",
    "- 随机梯度下降分类器 (SGDC)\n",
    "- 支撑向量机 (SVM)\n",
    "- Logistic回归（LogisticRegression）\n",
    "\n",
    "从上面的监督学习模型中选择三个适合我们这个问题的模型，并回答相应问题。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 模型1\n",
    "\n",
    "**模型名称**\n",
    "\n",
    "回答：支撑向量机 (SVM)\n",
    "\n",
    "\n",
    "**描述一个该模型在真实世界的一个应用场景。（你需要为此做点研究，并给出你的引用出处）**\n",
    "\n",
    "回答：\n",
    "- 人的行为认知判断\n",
    "- 文本分类\n",
    "\n",
    "**这个模型的优势是什么？他什么情况下表现最好？**\n",
    "\n",
    "回答：\n",
    "- SVM是小样本学习方法，对小规模的数据表现很好；\n",
    "- 避开了从归纳到演绎的传统过程，大大简化了通常的分类和回归等问题。\n",
    "\n",
    "**这个模型的缺点是什么？什么条件下它表现很差？**\n",
    "\n",
    "回答：\n",
    "- 难以解决多分类问题；\n",
    "- 不适合大数据集样本；\n",
    "- 对输入数据的表达形式很敏感；\n",
    "- 若各属性有一定的关联性，分类效果会受到较大影响。\n",
    "\n",
    "**根据我们当前数据集的特点，为什么这个模型适合这个问题。**\n",
    "\n",
    "回答：数据规模较小，SVM适合处理二分类问题。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 模型2\n",
    "\n",
    "**模型名称**\n",
    "\n",
    "回答：K近邻 (K Nearest Neighbors)\n",
    "\n",
    "\n",
    "**描述一个该模型在真实世界的一个应用场景。（你需要为此做点研究，并给出你的引用出处）**\n",
    "\n",
    "回答：手写体数字的识别问题（包括验证码识别、二维码识别等）\n",
    "\n",
    "**这个模型的优势是什么？他什么情况下表现最好？**\n",
    "\n",
    "回答：\n",
    "- 简单、易实现、易理解，无需参数估计及训练；\n",
    "- 适用于多分类问题、可用于非线性问题；\n",
    "- 算法复杂度低O(n)；\n",
    "- 准确度较高，对异常值不敏感。\n",
    "\n",
    "**这个模型的缺点是什么？什么条件下它表现很差？**\n",
    "\n",
    "回答：\n",
    "\n",
    "- 对于大数据集的计算耗时长；\n",
    "- 内存开销较大；\n",
    "- 样本分布不平衡时表现不佳；\n",
    "- 每次分类都是一次全局计算。\n",
    "- 对异常值敏感\n",
    "- 需要提前确定k值\n",
    "\n",
    "**根据我们当前数据集的特点，为什么这个模型适合这个问题。**\n",
    "\n",
    "回答：这是一个分类问题，数据量不大，计算时间不会很长，k值也较容易确定。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 模型3\n",
    "\n",
    "**模型名称**\n",
    "\n",
    "回答：集成方法 (Random Forest)\n",
    "\n",
    "**描述一个该模型在真实世界的一个应用场景。（你需要为此做点研究，并给出你的引用出处）**\n",
    "\n",
    "回答：随机森林（Random Forest），应用场景：用户流失分析、风险评估\n",
    "\n",
    "**这个模型的优势是什么？他什么情况下表现最好？**\n",
    "\n",
    "回答：\n",
    "- 随机森林方法被证明对大规模数据集和存在大量且有时不相关特征的项（item）来说很有用\n",
    "- 准确率很高\n",
    "- 不易发生过拟合\n",
    "\n",
    "**这个模型的缺点是什么？什么条件下它表现很差？**\n",
    "\n",
    "回答：\n",
    "- 在某些噪音较大的分类或回归问题上会过拟合\n",
    "- 对异常值敏感\n",
    "\n",
    "**根据我们当前数据集的特点，为什么这个模型适合这个问题。**\n",
    "\n",
    "回答：\n",
    "- 小数据范围噪声不会过拟合，准确度高\n",
    "- 本数据集数据分布较均匀，异常值较少"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 练习 - 创建一个训练和预测的流水线\n",
    "为了正确评估你选择的每一个模型的性能，创建一个能够帮助你快速有效地使用不同大小的训练集并在验证集上做预测的训练和验证的流水线是十分重要的。\n",
    "你在这里实现的功能将会在接下来的部分中被用到。在下面的代码单元中，你将实现以下功能：\n",
    "\n",
    " - 从[`sklearn.metrics`](http://scikit-learn.org/stable/modules/classes.html#sklearn-metrics-metrics)中导入`fbeta_score`和`accuracy_score`。\n",
    " - 用训练集拟合学习器，并记录训练时间。\n",
    " - 对训练集的前300个数据点和验证集进行预测并记录预测时间。\n",
    " - 计算预测训练集的前300个数据点的准确率和F-score。\n",
    " - 计算预测验证集的准确率和F-score。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "# TODO：从sklearn中导入两个评价指标 - fbeta_score和accuracy_score\n",
    "from sklearn.metrics import fbeta_score, accuracy_score\n",
    "\n",
    "def train_predict(learner, sample_size, X_train, y_train, X_val, y_val): \n",
    "    '''\n",
    "    inputs:\n",
    "       - learner: the learning algorithm to be trained and predicted on\n",
    "       - sample_size: the size of samples (number) to be drawn from training set\n",
    "       - X_train: features training set\n",
    "       - y_train: income training set\n",
    "       - X_val: features validation set\n",
    "       - y_val: income validation set\n",
    "    '''\n",
    "    \n",
    "    results = {}\n",
    "    \n",
    "    # TODO：使用sample_size大小的训练数据来拟合学习器\n",
    "    # TODO: Fit the learner to the training data using slicing with 'sample_size'\n",
    "    start = time() # 获得程序开始时间\n",
    "    learner = learner.fit(X_train[:sample_size], y_train[:sample_size]) # 训练算法\n",
    "    end = time() # 获得程序结束时间\n",
    "    \n",
    "    # TODO：计算训练时间\n",
    "    results['train_time'] = end - start\n",
    "    \n",
    "    # TODO: 得到在验证集上的预测值\n",
    "    #       然后得到对前300个训练数据的预测结果\n",
    "    start = time() # 获得程序开始时间\n",
    "    predictions_val = learner.predict(X_val)\n",
    "    predictions_train = learner.predict(X_train[:300])\n",
    "    end = time() # 获得程序结束时间\n",
    "    \n",
    "    # TODO：计算预测用时\n",
    "    results['pred_time'] = end - start\n",
    "            \n",
    "    # TODO：计算在最前面的300个训练数据的准确率\n",
    "    results['acc_train'] = accuracy_score(y_train[:300], predictions_train)\n",
    "        \n",
    "    # TODO：计算在验证上的准确率\n",
    "    results['acc_val'] = accuracy_score(y_val, predictions_val)\n",
    "    \n",
    "    # TODO：计算在最前面300个训练数据上的F-score\n",
    "    results['f_train'] = fbeta_score(y_train[:300], predictions_train, beta=0.5)\n",
    "        \n",
    "    # TODO：计算验证集上的F-score\n",
    "    results['f_val'] = fbeta_score(y_val, predictions_val, beta=0.5)\n",
    "       \n",
    "    # 成功\n",
    "    print (\"{} trained on {} samples.\".format(learner.__class__.__name__, sample_size))\n",
    "        \n",
    "    # 返回结果\n",
    "    return results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 练习：初始模型的评估\n",
    "在下面的代码单元中，您将需要实现以下功能：             \n",
    "- 导入你在前面讨论的三个监督学习模型。             \n",
    "- 初始化三个模型并存储在`'clf_A'`，`'clf_B'`和`'clf_C'`中。\n",
    "  - 使用模型的默认参数值，在接下来的部分中你将需要对某一个模型的参数进行调整。             \n",
    "  - 设置`random_state`  (如果有这个参数)。       \n",
    "- 计算1%， 10%， 100%的训练数据分别对应多少个数据点，并将这些值存储在`'samples_1'`, `'samples_10'`, `'samples_100'`中\n",
    "\n",
    "**注意：**取决于你选择的算法，下面实现的代码可能需要一些时间来运行！"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "SVC trained on 289 samples.\n",
      "SVC trained on 2894 samples.\n",
      "SVC trained on 28941 samples.\n",
      "KNeighborsClassifier trained on 289 samples.\n",
      "KNeighborsClassifier trained on 2894 samples.\n",
      "KNeighborsClassifier trained on 28941 samples.\n",
      "RandomForestClassifier trained on 289 samples.\n",
      "RandomForestClassifier trained on 2894 samples.\n",
      "RandomForestClassifier trained on 28941 samples.\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbbab212c18>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# TODO：从sklearn中导入三个监督学习模型\n",
    "from sklearn.svm import SVC\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "\n",
    "# TODO：初始化三个模型\n",
    "clf_A = SVC(kernel = 'linear')\n",
    "clf_B = KNeighborsClassifier(n_neighbors = 2)\n",
    "clf_C = RandomForestClassifier(random_state = 0)\n",
    "\n",
    "# TODO：计算1%， 10%， 100%的训练数据分别对应多少点\n",
    "samples_1 = int(len(X_train) * 0.01)\n",
    "samples_10 = int(len(X_train) * 0.10)\n",
    "samples_100 = int(len(X_train))\n",
    "\n",
    "# 收集学习器的结果\n",
    "results = {}\n",
    "for clf in [clf_A, clf_B, clf_C]:\n",
    "    clf_name = clf.__class__.__name__\n",
    "    results[clf_name] = {}\n",
    "    for i, samples in enumerate([samples_1, samples_10, samples_100]):\n",
    "        results[clf_name][i] = train_predict(clf, samples, X_train, y_train, X_val, y_val)\n",
    "\n",
    "# 对选择的三个模型得到的评价结果进行可视化\n",
    "vs.evaluate(results, accuracy, fscore)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "## 提高效果\n",
    "\n",
    "在这最后一节中，您将从三个有监督的学习模型中选择 *最好的* 模型来使用学生数据。你将在整个训练集（`X_train`和`y_train`）上使用网格搜索优化至少调节一个参数以获得一个比没有调节之前更好的 F-score。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 问题 3 - 选择最佳的模型\n",
    "\n",
    "*基于你前面做的评价，用一到两段话向 *CharityML* 解释这三个模型中哪一个对于判断被调查者的年收入大于 \\$50,000 是最合适的。*             \n",
    "**提示：**你的答案应该包括评价指标，预测/训练时间，以及该算法是否适合这里的数据。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**回答：**\n",
    "- 训练时间（Model Training）图表显示，SVM耗时大幅超过其它两种算法\n",
    "- 预测时间（Model Predicting）图表显示，KNeighbors耗时最多，RandomForest最少\n",
    "- 准确度（Accuracy Score）图表显示，RandomForest在所有训练集上准确度都是最高，KNeighbors的准确度表现也非常不错，在验证集上三者的准确度表现得比较接近\n",
    "- F-score图表显示，训练集分数排名依次为：RandomForest > KNeighbors > SVM，验证集上三者分数差距不大，RandomForest表现很好，KNeighbors表现稳定\n",
    "- 总体表现上RandomForest最佳，训练时间时间短，分类准确度高，是比较理想的算法，不过仍然有调优的空间。不过SVM的在准确度及F-Score的表现上也是可圈可点，若不评估训练时间，SVM算法也是个不错的选择。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 问题 4 - 用通俗的话解释模型\n",
    "\n",
    "*用一到两段话，向 *CharityML* 用外行也听得懂的话来解释最终模型是如何工作的。你需要解释所选模型的主要特点。例如，这个模型是怎样被训练的，它又是如何做出预测的。避免使用高级的数学或技术术语，不要使用公式或特定的算法名词。*"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**回答： ** \n",
    "\n",
    "- 随机森林采用多个决策树的投票机制来改善决策树，我们假设随机森林使用了m棵决策树，那么就需要产生m个一定数量的样本集来训练每一棵树，如果用全样本去训练m棵决策树显然是不可取的，全样本训练忽视了局部样本的规律，对于模型的泛化能力是有害的;\n",
    "- 采用有放回的抽样法产生n个样本，最终结果采用多数投票机制策略来获得\n",
    "\n",
    "- 随机森林的生成方法：\n",
    "\n",
    "  - 从样本集中通过重采样的方式产生n个样本\n",
    "\n",
    "  - 假设样本特征数目为a，对n个样本选择a中的k个特征，用建立决策树的方式获得最佳分割点\n",
    "\n",
    "  - 重复m次，产生m棵决策树\n",
    "\n",
    "  - 多数投票机制来进行预测\n",
    "\n",
    "（需要注意的一点是，这里m是指循环的次数，n是指样本的数目，n个样本构成训练的样本集，而m次循环中又会产生m个这样的样本集）"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 练习：模型调优\n",
    "调节选择的模型的参数。使用网格搜索（GridSearchCV）来至少调整模型的重要参数（至少调整一个），这个参数至少需尝试3个不同的值。你要使用整个训练集来完成这个过程。在接下来的代码单元中，你需要实现以下功能：\n",
    "\n",
    "- 导入[`sklearn.model_selection.GridSearchCV`](http://scikit-learn.org/stable/modules/generated/sklearn.model_selection.GridSearchCV.html) 和 [`sklearn.metrics.make_scorer`](http://scikit-learn.org/stable/modules/generated/sklearn.metrics.make_scorer.html).\n",
    "- 初始化你选择的分类器，并将其存储在`clf`中。\n",
    " - 设置`random_state` (如果有这个参数)。\n",
    "- 创建一个对于这个模型你希望调整参数的字典。\n",
    " - 例如: parameters = {'parameter' : [list of values]}。\n",
    " - **注意：** 如果你的学习器有 `max_features` 参数，请不要调节它！\n",
    "- 使用`make_scorer`来创建一个`fbeta_score`评分对象（设置$\\beta = 0.5$）。\n",
    "- 在分类器clf上用'scorer'作为评价函数运行网格搜索，并将结果存储在grid_obj中。\n",
    "- 用训练集（X_train, y_train）训练grid search object,并将结果存储在`grid_fit`中。\n",
    "\n",
    "**注意：** 取决于你选择的参数列表，下面实现的代码可能需要花一些时间运行！"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_clf\n",
      "------\n",
      "RandomForestClassifier(bootstrap=True, class_weight=None, criterion='gini',\n",
      "            max_depth=None, max_features='auto', max_leaf_nodes=None,\n",
      "            min_impurity_decrease=0.0, min_impurity_split=None,\n",
      "            min_samples_leaf=1, min_samples_split=2,\n",
      "            min_weight_fraction_leaf=0.0, n_estimators=100, n_jobs=1,\n",
      "            oob_score=False, random_state=0, verbose=0, warm_start=False)\n",
      "\n",
      "Unoptimized model\n",
      "------\n",
      "Accuracy score on validation data: 0.8389\n",
      "F-score on validation data: 0.6812\n",
      "\n",
      "Optimized Model\n",
      "------\n",
      "Final accuracy score on the validation data: 0.8456\n",
      "Final F-score on the validation data: 0.6940\n"
     ]
    }
   ],
   "source": [
    "# TODO：导入'GridSearchCV', 'make_scorer'和其他一些需要的库\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "from sklearn.metrics import fbeta_score,make_scorer\n",
    "\n",
    "# TODO：初始化分类器\n",
    "clf = RandomForestClassifier(random_state=0)\n",
    "\n",
    "# TODO：创建你希望调节的参数列表\n",
    "parameters = { 'n_estimators' : [25, 50, 100,150] }\n",
    "\n",
    "# TODO：创建一个fbeta_score打分对象\n",
    "scorer = make_scorer(fbeta_score, beta=0.5)\n",
    "\n",
    "# TODO：在分类器上使用网格搜索，使用'scorer'作为评价函数\n",
    "grid_obj = GridSearchCV(clf, parameters, scorer)\n",
    "\n",
    "# TODO：用训练数据拟合网格搜索对象并找到最佳参数\n",
    "grid_obj = grid_obj.fit(X_train, y_train)\n",
    "\n",
    "# 得到estimator\n",
    "best_clf = grid_obj.best_estimator_\n",
    "\n",
    "# 使用没有调优的模型做预测\n",
    "predictions = (clf.fit(X_train, y_train)).predict(X_val)\n",
    "best_predictions = best_clf.predict(X_val)\n",
    "\n",
    "# 汇报调优后的模型\n",
    "print (\"best_clf\\n------\")\n",
    "print (best_clf)\n",
    "\n",
    "# 汇报调参前和调参后的分数\n",
    "print (\"\\nUnoptimized model\\n------\")\n",
    "print (\"Accuracy score on validation data: {:.4f}\".format(accuracy_score(y_val, predictions)))\n",
    "print (\"F-score on validation data: {:.4f}\".format(fbeta_score(y_val, predictions, beta = 0.5)))\n",
    "print (\"\\nOptimized Model\\n------\")\n",
    "print (\"Final accuracy score on the validation data: {:.4f}\".format(accuracy_score(y_val, best_predictions)))\n",
    "print (\"Final F-score on the validation data: {:.4f}\".format(fbeta_score(y_val, best_predictions, beta = 0.5)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 问题 5 - 最终模型评估\n",
    "\n",
    "_你的最优模型在测试数据上的准确率和 F-score 是多少？这些分数比没有优化的模型好还是差？_\n",
    "**注意：**请在下面的表格中填写你的结果，然后在答案框中提供讨论。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 结果:\n",
    " \n",
    "| 评价指标         |  天真预测器 |未优化的模型        | 优化的模型        |\n",
    "| :------------: |  :---------------: |  :---------------: | :-------------: | \n",
    "| 准确率          |   0.2478    |    0.8389          |      0.8456          |\n",
    "| F-score        |   0.2917    |    0.6812         |      0.6940          |"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**回答：**\n",
    "- 比没有优化的模型要好"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "## 特征的重要性\n",
    "\n",
    "在数据上（比如我们这里使用的人口普查的数据）使用监督学习算法的一个重要的任务是决定哪些特征能够提供最强的预测能力。专注于少量的有效特征和标签之间的关系，我们能够更加简单地理解这些现象，这在很多情况下都是十分有用的。在这个项目的情境下这表示我们希望选择一小部分特征，这些特征能够在预测被调查者是否年收入大于\\$50,000这个问题上有很强的预测能力。\n",
    "\n",
    "选择一个有 `'feature_importance_'` 属性的scikit学习分类器（例如 AdaBoost，随机森林）。`'feature_importance_'` 属性是对特征的重要性排序的函数。在下一个代码单元中用这个分类器拟合训练集数据并使用这个属性来决定人口普查数据中最重要的5个特征。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 问题 6 - 观察特征相关性\n",
    "\n",
    "当**探索数据**的时候，它显示在这个人口普查数据集中每一条记录我们有十三个可用的特征。             \n",
    "_在这十三个记录中，你认为哪五个特征对于预测是最重要的，选择每个特征的理由是什么？你会怎样对他们排序？_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**回答：**\n",
    "- 特征1：age，年龄越大，工作年限越长，社会收入及赚钱渠道会更广，财富积累会更多\n",
    "- 特征2：capital-gain，有其他收益的，或者是理财的，一般对收入贡献度很大\n",
    "- 特征3：marital-status，结婚的人一般经济和社会条件比较成熟，收入一般不会很低\n",
    "- 特征4：workclass，劳动类型不同，收入相差也会较大\n",
    "- 特征5：education_level，受教育水平越高，收入一般较高\n",
    "- age>capital-gain>marital-status>workclass>education_level"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 练习 - 提取特征重要性\n",
    "\n",
    "选择一个`scikit-learn`中有`feature_importance_`属性的监督学习分类器，这个属性是一个在做预测的时候根据所选择的算法来对特征重要性进行排序的功能。\n",
    "\n",
    "在下面的代码单元中，你将要实现以下功能：\n",
    " - 如果这个模型和你前面使用的三个模型不一样的话从sklearn中导入一个监督学习模型。\n",
    " - 在整个训练集上训练一个监督学习模型。\n",
    " - 使用模型中的 `'feature_importances_'`提取特征的重要性。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbbab14c438>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# TODO：导入一个有'feature_importances_'的监督学习模型\n",
    "from sklearn.ensemble import AdaBoostClassifier\n",
    "\n",
    "# TODO：在训练集上训练一个监督学习模型\n",
    "model = AdaBoostClassifier(random_state =5)\n",
    "model.fit(X_train,y_train)\n",
    "\n",
    "# TODO： 提取特征重要性\n",
    "importances = model.feature_importances_\n",
    "\n",
    "# 绘图\n",
    "vs.feature_plot(importances, X_train, y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 问题 7 - 提取特征重要性\n",
    "观察上面创建的展示五个用于预测被调查者年收入是否大于\\$50,000最相关的特征的可视化图像。\n",
    "\n",
    "_这五个特征的权重加起来是否超过了0.5?_<br>\n",
    "_这五个特征和你在**问题 6**中讨论的特征比较怎么样？_<br>\n",
    "_如果说你的答案和这里的相近，那么这个可视化怎样佐证了你的想法？_<br>\n",
    "_如果你的选择不相近，那么为什么你觉得这些特征更加相关？_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**回答：**\n",
    "- 五个特征的权重加起来 0.56 ，超过了 0.5\n",
    "- 和我在问题 6中讨论的特征比较差距不是特别大，capital-gain与age对收入影响较大\n",
    "- capital loss 资本损失这个概念理解不是特别透彻，文中也没特别说明"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 特征选择\n",
    "\n",
    "如果我们只是用可用特征的一个子集的话模型表现会怎么样？通过使用更少的特征来训练，在评价指标的角度来看我们的期望是训练和预测的时间会更少。从上面的可视化来看，我们可以看到前五个最重要的特征贡献了数据中**所有**特征中超过一半的重要性。这提示我们可以尝试去**减小特征空间**，简化模型需要学习的信息。下面代码单元将使用你前面发现的优化模型，并**只使用五个最重要的特征**在相同的训练集上训练模型。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Final Model trained on full data\n",
      "------\n",
      "Accuracy on validation data: 0.8456\n",
      "F-score on validation data: 0.6940\n",
      "\n",
      "Final Model trained on reduced data\n",
      "------\n",
      "Accuracy on validation data: 0.8230\n",
      "F-score on validation data: 0.6482\n"
     ]
    }
   ],
   "source": [
    "# 导入克隆模型的功能\n",
    "from sklearn.base import clone\n",
    "\n",
    "# 减小特征空间\n",
    "X_train_reduced = X_train[X_train.columns.values[(np.argsort(importances)[::-1])[:5]]]\n",
    "X_val_reduced = X_val[X_val.columns.values[(np.argsort(importances)[::-1])[:5]]]\n",
    "\n",
    "# 在前面的网格搜索的基础上训练一个“最好的”模型\n",
    "clf_on_reduced = (clone(best_clf)).fit(X_train_reduced, y_train)\n",
    "\n",
    "# 做一个新的预测\n",
    "reduced_predictions = clf_on_reduced.predict(X_val_reduced)\n",
    "\n",
    "# 对于每一个版本的数据汇报最终模型的分数\n",
    "print (\"Final Model trained on full data\\n------\")\n",
    "print (\"Accuracy on validation data: {:.4f}\".format(accuracy_score(y_val, best_predictions)))\n",
    "print (\"F-score on validation data: {:.4f}\".format(fbeta_score(y_val, best_predictions, beta = 0.5)))\n",
    "print (\"\\nFinal Model trained on reduced data\\n------\")\n",
    "print (\"Accuracy on validation data: {:.4f}\".format(accuracy_score(y_val, reduced_predictions)))\n",
    "print (\"F-score on validation data: {:.4f}\".format(fbeta_score(y_val, reduced_predictions, beta = 0.5)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 问题 8 - 特征选择的影响\n",
    "\n",
    "*最终模型在只是用五个特征的数据上和使用所有的特征数据上的 F-score 和准确率相比怎么样？*  \n",
    "*如果训练时间是一个要考虑的因素，你会考虑使用部分特征的数据作为你的训练集吗？*"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**回答：**\n",
    "- 使用所有的特征数据的准确率、F-Score 高出 5 个特征数据的值，但是差距不是很大 \n",
    "- 会考虑使用部分特征的数据作为训练集，节省了训练时间"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 问题 9 - 在测试集上测试你的模型\n",
    "\n",
    "终于到了测试的时候，记住，测试集只能用一次。\n",
    "\n",
    "*使用你最有信心的模型，在测试集上测试，计算出准确率和 F-score。*\n",
    "*简述你选择这个模型的原因，并分析测试结果*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.865118850193\n",
      "0.745963514364\n"
     ]
    }
   ],
   "source": [
    "#TODO test your model on testing data and report accuracy and F score\n",
    "from sklearn.ensemble import AdaBoostClassifier\n",
    "from sklearn.metrics import fbeta_score, accuracy_score\n",
    "\n",
    "bdt = AdaBoostClassifier(n_estimators=150, random_state=1)\n",
    "bdt = bdt.fit(X_train, y_train)\n",
    "predict_value = bdt.predict(X_test)\n",
    "\n",
    "_accuracy_score = accuracy_score(y_test, predict_value)\n",
    "_fbeta_score = fbeta_score(y_test, predict_value, beta=0.5)\n",
    "\n",
    "print (_accuracy_score)\n",
    "print (_fbeta_score)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> **注意：** 当你写完了所有的代码，并且回答了所有的问题。你就可以把你的 iPython Notebook 导出成 HTML 文件。你可以在菜单栏，这样导出**File -> Download as -> HTML (.html)**把这个 HTML 和这个 iPython notebook 一起做为你的作业提交。"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}