第四次作业-婚恋预测

#%%

import pandas as pd

import numpy as np

import matplotlib

import matplotlib.pyplot as plt

import seaborn as sns



import os 

os.chdir(r'D:\give_me_five\githome\DataAnalysis\Case5')



matplotlib.rcParams['axes.unicode_minus']=False#解决保存图像时负号'-'显示为方块的问题

plt.rcParams['font.sans-serif'] = ['SimHei']#指定默认字体 

#%%

dat1 = pd.read_csv('./dataset/date_data2.csv')# 读取数据

dat1.head()

#%%

# 比例正好1:1

dat1.Dated.value_counts()

#%%

from sklearn.tree import DecisionTreeClassifier

from sklearn.model_selection import train_test_split

from sklearn.metrics import roc_curve



# 拆分数据集

dat1_x = dat1.drop(['Dated'],axis=1)

dat1_y = dat1.Dated

X_train,X_test,y_train,y_test = train_test_split(dat1_x,dat1_y,test_size=0.4,random_state=1024)

#%%



from sklearn.model_selection import GridSearchCV

# 使用网格搜索，交叉验证，寻找最优超参数

model_dt = DecisionTreeClassifier()

params={'criterion':['gini','entropy'],'max_depth':[2,3,4,5,6,7,8],

        'min_samples_split':[2,3,4,5,10,20,30]}

gridsearch = GridSearchCV(model_dt,params,scoring='roc_auc',cv=3)

gridsearch.fit(X_train,y_train)

# 最优超参数是{'criterion': 'gini', 'max_depth': 2, 'min_samples_split': 2}

gridsearch.best_params_

#%%

# 使用最优超参数建模

model_dt = DecisionTreeClassifier(criterion='entropy',max_depth=3,min_samples_split=2)

model_dt.fit(X_train,y_train)

#%%

# 使用模型预测训练集和测试集数据

y_train_predict = model_dt.predict(X_train)# 用模型预测训练集的结果

y_train_predict_p = model_dt.predict_proba(X_train)[:,1]#用模型预测训练集的概率

y_test_predict = model_dt.predict(X_test)#用模型预测测试集的结果

y_test_predict_p = model_dt.predict_proba(X_test)[:,1]#用模型预测测试集的概率

#%%

# 查看训练集和测试集的roc 曲线

# 训练集的roc曲线，比测试集roc曲线面积大一点，且完全包住，有点过拟合，数量太少

fpr_test, tpr_test, th_test = roc_curve(y_test, y_test_predict_p)

fpr_train, tpr_train, th_train = roc_curve(y_train, y_train_predict_p)

plt.figure(figsize=[6,6])

plt.plot(fpr_test, tpr_test, color='blue',label='test')

plt.plot(fpr_train, tpr_train, color='red',label='train')

plt.legend()

plt.xlabel('False Positive Rate')

plt.ylabel('True Positive Rate')

plt.title(' Roc Curve')

plt.show()

#%%

import pydotplus

from IPython.display import Image

import sklearn.tree as tree



# 画出决策树的每一层  看出assets重要性较高

dot_data = tree.export_graphviz(

    model_dt, 

    out_file=None, 

    feature_names=dat1_x.columns,

    max_depth=3,

    class_names=['0','1'],

    filled=True

) 

            

graph = pydotplus.graph_from_dot_data(dot_data)  

Image(graph.create_png()) 

#%%



# 神经网络

from sklearn.neural_network import MLPClassifier

from sklearn.preprocessing import MinMaxScaler



# 极差标准化 将数值型特征列的取值范围 变化成统一数量级

mmScaler = MinMaxScaler()

scaled_X_train = mmScaler.fit_transform(X_train)

scaled_X_test = mmScaler.fit_transform(X_test)

#%%

# 使用网格搜索，交叉验证，寻找最优参数

param_grid = {

    'hidden_layer_sizes':[(10, ), (15, ), (20, ), (5, 5)],

    'activation':['logistic', 'tanh', 'relu'], 

    'alpha':[0.001, 0.01, 0.1, 0.2, 0.4, 1, 10],

    'max_iter':[200,500,600,1000]

}

model_mlp = MLPClassifier()

gridsearch = GridSearchCV(estimator=model_mlp, param_grid=param_grid, 

                   scoring='roc_auc', cv=3, n_jobs=-1)

gridsearch.fit(scaled_X_train, y_train)

#%%

# 最优参数是{'activation': 'relu','alpha': 0.4,'hidden_layer_sizes': (5,5),'max_iter': 1000}

gridsearch.best_params_

#%%

# 使用最优参数建模

model_mlp = MLPClassifier(activation='relu',

                          alpha=0.4,hidden_layer_sizes=(5,5),max_iter=1000)

model_mlp.fit(scaled_X_train,y_train)

#%%

# 使用模型预测训练集和测试集

y_train_predict = model_mlp.predict(scaled_X_train) # 测试集结果

y_train_predict_p = model_mlp.predict_proba(scaled_X_train)[:,1]# 训练集概率

y_test_predict = model_mlp.predict(scaled_X_test) # 测试集结果

y_test_predict_p = model_mlp.predict_proba(scaled_X_test)[:,1]# 测试集概率

#%%

# 查看训练集和测试集的roc 曲线

# 训练集的roc曲线，和测试集 roc曲线相差不大

fpr_test, tpr_test, th_test = roc_curve(y_test, y_test_predict_p)

fpr_train, tpr_train, th_train = roc_curve(y_train, y_train_predict_p)

plt.figure(figsize=[6,6])

plt.plot(fpr_test, tpr_test, color='blue',label='test')

plt.plot(fpr_train, tpr_train, color='red',label='train')

plt.legend()

plt.xlabel('False Positive Rate')

plt.ylabel('True Positive Rate')

plt.title(' Roc Curve')

plt.show()

#%%



# 使用分箱

# income [0,4500,6500,9000,11800]

dat1['income_bins'] = pd.qcut(dat1.income,5)

# attrative [0,21,39,65,79]

dat1['attractive_bins'] = pd.qcut(dat1.attractive,5)

# assets [0,25,52,94,145]

dat1['assets_bins'] = pd.qcut(dat1.assets,5)

#%%

income_bins = [0,4500,6500,9000,11800,34000]

dat1['income_bins'] = pd.cut(dat1.income,bins=income_bins)

attractive_bins = [0,21,39,65,79,100]

dat1['attractive_bins'] = pd.cut(dat1.attractive,bins=attractive_bins)

assets_bins = [0,25,52,94,145,486]

dat1['assets_bins'] = pd.cut(dat1.assets,bins=assets_bins)

#%%

# 计算column中每一类的WOE值和column的IV值

# 返回字典，其中WOE_dict表示WOE编码表

# IV表示column的IV值



from pandas import DataFrame



def Func_CalcWOE(df,col,target):

    total = df.groupby([col])[target].count()

    total = DataFrame({'total':total})

    bad = df.groupby([col])[target].sum()

    bad = DataFrame({'bad':bad})

    regroup = total.merge(bad,left_index=True,right_index=True,how='left')

    regroup['good'] = regroup.total - regroup.bad

    N = regroup.total.sum()

    B = regroup.bad.sum()

    G = N - B

    regroup['bad_pcnt'] = regroup.bad.map(lambda x:x/B)

    regroup['good_pcnt'] = regroup.good.map(lambda x:x/G)

    regroup['woe'] = regroup.apply(lambda x:np.log(x.good_pcnt/x.bad_pcnt),axis=1)

    WOE_dict = regroup['woe'].to_dict()

    IV = regroup.apply(lambda x:(x.good_pcnt - x.bad_pcnt)*x.woe,axis=1)

    IV = sum(IV)

    return {'WOE':WOE_dict,'IV':IV}

#%%

# income_bins IV 值1.33

woe = Func_CalcWOE(dat1,'income_bins','Dated')['WOE']

dat1.income_bins = dat1.income_bins.map(lambda x:woe.get(x)).astype('float64')

#%%

# attractive IV 值 0.359

woe = Func_CalcWOE(dat1,'attractive_bins','Dated')['WOE']

dat1.attractive_bins = dat1.attractive_bins.map(lambda x:woe.get(x)).astype('float64')

#%%

# assets 特征列比较特殊，0和1的集中度较高，使用婚恋成功率来表示

dat2 = pd.crosstab(dat1.assets_bins,dat1.Dated)

dat2 = dat2.div(dat2.sum(1),axis=0)

assets_dict = dat2[1].to_dict()

dat1.assets_bins = dat1.assets_bins.map(lambda x:assets_dict.get(x)).astype('float64')

#%%

# 拆分数据集

datx_bins = dat1.iloc[:,-3:]

daty_bins = dat1.Dated

X_train,X_test,y_train,y_test = train_test_split(datx_bins,daty_bins,test_size=0.4,random_state=1024)



#%%

# 使用决策树建模

from sklearn.model_selection import GridSearchCV

# 使用网格搜索，交叉验证，寻找最优超参数

model_dt = DecisionTreeClassifier()

params={'criterion':['gini','entropy'],'max_depth':[2,3,4,5,6,7,8],

        'min_samples_split':[2,3,4,5,10,20,30]}

gridsearch = GridSearchCV(model_dt,params,scoring='roc_auc',cv=3)

gridsearch.fit(X_train,y_train)

# 最优超参数是{'criterion': 'gini', 'max_depth': 2, 'min_samples_split': 2}

gridsearch.best_params_

#%%

# 使用最优超参数建模

model_dt = DecisionTreeClassifier(criterion='entropy',max_depth=3,min_samples_split=2)

model_dt.fit(X_train,y_train)

#%%

# 使用模型预测训练集和测试集数据

y_train_predict = model_dt.predict(X_train)# 用模型预测训练集的结果

y_train_predict_p = model_dt.predict_proba(X_train)[:,1]#用模型预测训练集的概率

y_test_predict = model_dt.predict(X_test)#用模型预测测试集的结果

y_test_predict_p = model_dt.predict_proba(X_test)[:,1]#用模型预测测试集的概率

#%%

# 查看训练集和测试集的roc 曲线

# 训练集的roc曲线 和测试集的roc曲线相比  效果好很多，存在过拟合

fpr_test, tpr_test, th_test = roc_curve(y_test, y_test_predict_p)

fpr_train, tpr_train, th_train = roc_curve(y_train, y_train_predict_p)

plt.figure(figsize=[6,6])

plt.plot(fpr_test, tpr_test, color='blue',label='test')

plt.plot(fpr_train, tpr_train, color='red',label='train')

plt.legend()

plt.xlabel('False Positive Rate')

plt.ylabel('True Positive Rate')

plt.title(' Roc Curve')

plt.show()

#%%

# 神经网络

from sklearn.neural_network import MLPClassifier

from sklearn.preprocessing import MinMaxScaler



# 极差标准化 将数值型特征列的取值范围 变化成统一数量级

mmScaler = MinMaxScaler()

scaled_X_train = mmScaler.fit_transform(X_train)

scaled_X_test = mmScaler.fit_transform(X_test)

#%%

# 使用网格搜索，交叉验证，寻找最优参数

param_grid = {

    'hidden_layer_sizes':[(10, ), (15, ), (20, ), (5, 5)],

    'activation':['logistic', 'tanh', 'relu'], 

    'alpha':[0.001, 0.01, 0.1, 0.2, 0.4, 1, 10],

    'max_iter':[200,500,600,1000]

}

model_mlp = MLPClassifier()

gridsearch = GridSearchCV(estimator=model_mlp, param_grid=param_grid, 

                   scoring='roc_auc', cv=3, n_jobs=-1)

gridsearch.fit(scaled_X_train, y_train)

#%%

# 最优参数是{'activation': 'tanh','alpha': 0.001,'hidden_layer_sizes': (5,5),'max_iter': 600}

gridsearch.best_params_

#%%

# 使用最优参数建模

model_mlp = MLPClassifier(activation='tanh',

                          alpha=0.001,hidden_layer_sizes=(5,5),max_iter=600)

model_mlp.fit(scaled_X_train,y_train)

#%%

# 使用模型预测训练集和测试集

y_train_predict = model_mlp.predict(scaled_X_train) # 测试集结果

y_train_predict_p = model_mlp.predict_proba(scaled_X_train)[:,1]# 训练集概率

y_test_predict = model_mlp.predict(scaled_X_test) # 测试集结果

y_test_predict_p = model_mlp.predict_proba(scaled_X_test)[:,1]# 测试集概率

#%%

# 查看训练集和测试集的roc 曲线

# 训练集的roc曲线 和 测试集roc曲线 有差异

fpr_test, tpr_test, th_test = roc_curve(y_test, y_test_predict_p)

fpr_train, tpr_train, th_train = roc_curve(y_train, y_train_predict_p)

plt.figure(figsize=[6,6])

plt.plot(fpr_test, tpr_test, color='blue',label='test')

plt.plot(fpr_train, tpr_train, color='red',label='train')

plt.legend()

plt.xlabel('False Positive Rate')

plt.ylabel('True Positive Rate')

plt.title(' Roc Curve')

plt.show()