8_2_回归模型评估.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Coef: [0.27247571 0.268934  ]\n",
      "MSE: 0.059538014493206\n",
      "MAE: 0.20435511892083338\n",
      "R2: 0.16759924962620698\n"
     ]
    }
   ],
   "source": [
    "##学习神经网络之前需要安装三个包：\n",
    "#pip install tensorflow\n",
    "# conda install pip\n",
    "# pip install keras\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.preprocessing import MinMaxScaler,StandardScaler\n",
    "from sklearn.preprocessing import LabelEncoder,OneHotEncoder\n",
    "from sklearn.preprocessing import Normalizer\n",
    "from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
    "from sklearn.decomposition import PCA\n",
    "\n",
    "\n",
    "#画出决策树-graphviz\n",
    "import os\n",
    "os.environ[\"PATH\"]+=os.pathsep+\"D:/CHEN/biancheng/JupyterTest/graphviz-2.38/release/bin/\"\n",
    "import pydotplus\n",
    "\n",
    "def hr_preprocessing(sl=False,le=False,npr=False,amh=False,tsc=False,wa=False,pl5=False,dp=False,slr=False,lower_d=False,ld_n=1):\n",
    "    df=pd.read_csv('./data/HR.csv')\n",
    "   \n",
    "    #1.清洗数据(去除异常值、抽样)\n",
    "    df=df.dropna(subset=['satisfaction_level','last_evaluation'])\n",
    "    df=df[df['satisfaction_level']<=1][df['salary']!='nme']\n",
    "     #2。得到标注\n",
    "    label = df['left']\n",
    "    df=df.drop('left',axis=1)#axis=1指定以列删除\n",
    "    #3.特征选择\n",
    "    #4.特征处理  \n",
    "    scaler_lst=[sl,le,npr,amh,tsc,wa,pl5]\n",
    "    column_lst=['satisfaction_level','last_evaluation','number_project','average_monthly_hours','time_spend_company','Work_accident','promotion_last_5years']\n",
    "    for i in range(len(scaler_lst)):\n",
    "        if not scaler_lst[i]:\n",
    "            df[column_lst[i]]=\\\n",
    "                MinMaxScaler().fit_transform(df[column_lst[i]].values.reshape(-1,1)).reshape(1,-1)[0]#二维格式\n",
    "        else:\n",
    "            df[column_lst[i]]=\\\n",
    "                StandardScaler().fit_transform(df[column_lst[i]].values.reshape(-1,1)).reshape(1,-1)[0]\n",
    "      \n",
    "    scaler_lst=[dp,slr]\n",
    "    column_lst=['department','salary']\n",
    "    for i in range(len(scaler_lst)):\n",
    "        if not scaler_lst[i]:\n",
    "            if column_lst[i]=='salary':\n",
    "                df[column_lst[i]]=[map_salary(s) for s in df['salary'].values]\n",
    "            else:\n",
    "                df[column_lst[i]]=LabelEncoder().fit_transform(df[column_lst[i]])\n",
    "            df[column_lst[i]]=MinMaxScaler().fit_transform(df[column_lst[i]].values.reshape(-1,1)).reshape(1,-1)[0]\n",
    "        else:\n",
    "            df=pd.get_dummies(df,columns=[column_lst[i]])\n",
    "    if lower_d:\n",
    "        return PCA(n_components=ld_n).fit_transform(df,values).label#PVA降维，可不使用标注\n",
    "    return df,label\n",
    "#使salary返回对应的等级而不是012\n",
    "d=dict([('low',0),('medium',1),('high',2)])\n",
    "def map_salary(s):\n",
    "    return d.get(s,0)\n",
    "#建模\n",
    "def hr_modeling(features,label):\n",
    "    from sklearn.model_selection import train_test_split  #切分训练集测试集\n",
    "    f_v=features.values\n",
    "    f_names=features.columns.values #得到特征名称\n",
    "    l_v=label.values\n",
    "    X_tt,X_validation,Y_tt,Y_validation=train_test_split(f_v,l_v,test_size=0.2)   #获得验证集占20%\n",
    "    X_train,X_test,Y_train,Y_test=train_test_split(X_tt,Y_tt,test_size=0.25)  #训练集、测试集\n",
    "    #print(len(X_train),len(X_validation),len(X_test))#得到6：2：2\n",
    "    #KNN\n",
    "    from sklearn.neighbors import NearestNeighbors,KNeighborsClassifier\n",
    "    from sklearn.metrics import accuracy_score,recall_score,f1_score#F一\n",
    "    from sklearn.naive_bayes import GaussianNB,BernoulliNB#离散型条件下表现更好，对特征有严格要求\n",
    "    #决策树包\n",
    "    from sklearn.tree import DecisionTreeClassifier,export_graphviz\n",
    "    from sklearn.externals.six import StringIO\n",
    "    from sklearn.ensemble import RandomForestClassifier\n",
    "    from sklearn.ensemble import AdaBoostClassifier\n",
    "    from sklearn.linear_model import LogisticRegression  #引入逻辑回归\n",
    "    #深度神经网络\n",
    "    from keras.models import Sequential #序列化的容器\n",
    "    from keras.layers import Dense,Activation  #稠密层\n",
    "    from keras.optimizers import SGD  #随机梯度下降算法\n",
    "    mdl=Sequential()\n",
    "    #建一个稠密层表示输入，传入参数表示输出就可以了，只有在输入层表示输入，输出代表隐含层神经元个数\n",
    "    #inpout_dim:输入维度与特征个数密切相关\n",
    "    mdl.add(Dense(50,input_dim=len(f_v[0])))\n",
    "    mdl.add(Activation('sigmoid'))\n",
    "    mdl.add(Dense(2))#输出层  标注只有两个维度\n",
    "    mdl.add(Activation('softmax'))#保证归一化\n",
    "    sgd=SGD(lr=0.01)#构造优化器，lr:学习率--同α#lr=0.03/0.05有了召回，准确率增加些\n",
    "    #编译 ，loss最优化函数：对每个输出减去对应输出值平方再开方，得到的值让它最小\n",
    "    #mdl.compile(loss='mean_squared_error',optimizer=sgd) #优化器adam(亚当优化器)\n",
    "    mdl.compile(loss='mean_squared_error',optimizer='adam')\n",
    "    #nb_epoch迭代代数，batch_size下降梯度#np_epoch=10000,batch_size=8999\n",
    "    #mdl.fit(X_train,np.array([[0,1] if i==1 else [1,0] for i in Y_train]),nb_epoch=1000,batch_size=2048)\n",
    "    mdl.fit(X_train,np.array([[0,1] if i==1 else [1,0] for i in Y_train]),nb_epoch=10000,batch_size=8999)\n",
    "    #预测\n",
    "    xy_lst=[(X_train,Y_train),(X_validation,Y_validation),(X_test,Y_test)]#依次用0，1，2表示\n",
    "    ##分类评估ROC、AUC\n",
    "    import matplotlib.pyplot as plt\n",
    "    from sklearn.metrics import roc_curve,auc,roc_auc_score\n",
    "    f=plt.figure()\n",
    "    for i in range(len(xy_lst)):\n",
    "        X_part=xy_lst[i][0]\n",
    "        Y_part=xy_lst[i][1]\n",
    "        Y_pred=mdl.predict(X_part)#输出分类标注\n",
    "        print(Y_pred)\n",
    "        #onehot类型，只需要取出值为1数据，（-1，1）指的是一行，只需要在第一列调配\n",
    "        Y_pred=np.array(Y_pred[:,-1]).reshape((-1,1))[0]\n",
    "        \n",
    "#         print(i)\n",
    "#         print(NN,'-ACC',accuracy_score(Y_part,Y_pred))\n",
    "#         print(NN,'-REC',recall_score(Y_part,Y_pred))\n",
    "#         print(NN,'-F1',f1_score(Y_part,Y_pred))\n",
    "        f.add_subplot(1,3,i+1)\n",
    "        fpr,tpr,threshold=roc_curve(Y_part,Y_pred)\n",
    "        plt.plot(fpr,tpr)\n",
    "        print(\"NN\",\"AUC\",auc(fpr,tpr))\n",
    "        print('NN','AUC_Score',roc_auc_score(Y_part,Y_pred))\n",
    "    plt.show()\n",
    "    return\n",
    "    models=[]\n",
    "    #models.append(('KNN',KNeighborsClassifier(n_neighbors=3)))\n",
    "    #models.append(('GaussianNB',GaussianNB()))\n",
    "    #models.append(('BernoulliNB',BernoulliNB()))\n",
    "    \n",
    "    models.append(('DecisionTreeGini',DecisionTreeClassifier()))\n",
    "    #models.append(('DecisionTreeGini',DecisionTreeClassifier(min_impurity_split=0.1)))#效果不好舍弃\n",
    "    models.append(('DecisionTreeEntropy',DecisionTreeClassifier(criterion='entropy')))\n",
    "    models.append(('OriginalRandomForest',RandomForestClassifier()))\n",
    "    models.append(('RandomForest',RandomForestClassifier(n_estimators=81,max_features=None))) #随机森林 max_features=None,bootstrap=False\n",
    "    models.append(('AdaBoost',AdaBoostClassifier(n_estimators=1000)))#base_estimators=SVC()   algorithm='SAMME'\n",
    "    models.append(('LogisticRegression',LogisticRegression(C=1000,tol=le-10,solver='saga')))#经运行，效果不如前面方法，通过调参方式\n",
    "    #C越大 ，正则化越小；tol:精度；solver：方法---显然这样是不行的，罗吉斯特映射线性模型，而我们的模型不是线性可分的，所以达到这种效果\n",
    "    for clf_name,clf in models:\n",
    "        clf.fit(X_train,Y_train)\n",
    "        xy_lst=[(X_train,Y_train),(X_validation,Y_validation),(X_test,Y_test)]#依次用0，1，2表示\n",
    "        for i in range(len(xy_lst)):\n",
    "            X_part=xy_lst[i][0]\n",
    "            Y_part=xy_lst[i][1]\n",
    "            Y_pred=clf.predict(X_part)\n",
    "            print(i)\n",
    "            print(clf_name,'-ACC',accuracy_score(Y_part,Y_pred))\n",
    "            print(clf_name,'-REC',recall_score(Y_part,Y_pred))\n",
    "            print(clf_name,'-F1',f1_score(Y_part,Y_pred))\n",
    "            #dot_data=export_graphviz(clf,out_file=None,  #此时直接输出dot_data,outfile不为None,通过句柄输出\n",
    "             #                       feature_names=f_names,\n",
    "              #                      class_names=['NL','L'],  #标注名称\n",
    "               #                     filled=True,\n",
    "                #                    rounded=True,\n",
    "                 #                   special_characters=True)\n",
    "            #绘制树\n",
    "            #graph=pydotplus.graph_from_dot_data(dot_data)\n",
    "            #graph.write_pdf('dt_tree_2.pdf')\n",
    " #线性回归 \n",
    "def regr_test(features,label):\n",
    "    #print('X',features)\n",
    "    #print('Y',label)\n",
    "    from sklearn.linear_model import LinearRegression,Ridge,Lasso#分别引入线性回归，岭回归和Lasso\n",
    "    #regr=LinearRegression()\n",
    "    regr=Ridge(alpha=1)#alpha=0,0.2,0.4,0.8,1  效果不明显\n",
    "    #regr=Lasso(alpha=0)#alpha=0,0.02  #不断调参直至参数趋于稳定\n",
    "    regr.fit(features.values,label.values)\n",
    "    Y_pred=regr.predict(features.values)\n",
    "    print('Coef:',regr.coef_)\n",
    "    from sklearn.metrics import mean_squared_error,mean_absolute_error,r2_score\n",
    "    print('MSE:',mean_squared_error(Y_pred,label.values))#平均平方误差\n",
    "    print('MAE:',mean_absolute_error(label.values,Y_pred))\n",
    "    print('R2:',r2_score(label.values,Y_pred))\n",
    "def main():\n",
    "    features,label=hr_preprocessing()\n",
    "    #print(hr_preprocessing())\n",
    "    #print(hr_processing(sl=True,le=True,npr=True,dp=True,lower_d=True,ld_n=3))\n",
    "    #hr_modeling(features,label)\n",
    "    regr_test(features[['number_project','average_monthly_hours']],features['last_evaluation'])\n",
    "if __name__=='__main__':\n",
    "    main()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#打印出来依次是特征、标注、参数、评估"
   ]
  }
 ],
 "metadata": {
  "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}