八大直播，八大案例_第六讲

作业要求：

1、背景介绍：

一家婚恋网站公司希望根据已注册用户的历史相亲数据,建立新用户相亲成功可能性的预测模型，数据存放在“date_data2.csv”中。

2、主要变量说明如下：

#income-月均收入（元）

#attractive-由婚恋网站评定出的个人魅力值,分值从0-100。

#assets-资产(万元)

#edueduclass-教育等级:1=小学,2=初中;3=高中,4=本科,5=硕士及以上

#Dated-是否相亲成功:1代表成功

3、作业安排：

3.1 基础知识：

      1）比较逻辑回归、决策树、神经网络的算法差异性比较。

3.2 案例解答步骤如下：

     1）使用决策树、神经网络建立相亲成功预测模型并通过调节超参数进行模型调优，比较两个模型的优劣。

      2)对income,attractive,assets进行分箱(5分箱)处理，用分箱后的数据建模，并比较与1）步骤中模型的表现是否有差异。

# coding: utf-8



# In[68]:



import pandas as pd

import numpy as np

import os

import matplotlib.pyplot as plt

import seaborn as sns

import sklearn.feature_selection as feature_selection

import sklearn.model_selection as cross_validation

import sklearn.metrics as metrics





# In[16]:



os.chdir(r'E:\数据分析\天善-PYTHON\八场直播，八大案例（金融风控）\第6讲\提交-第六讲：使用决策树和神经网络预测客户流失倾向\作业')

data = pd.read_csv(r'E:\数据分析\天善-PYTHON\八场直播，八大案例（金融风控）\第6讲\提交-第六讲：使用决策树和神经网络预测客户流失倾向\作业\date_data2.csv',encoding = 'gbk')

data['income_new'] = data['income']/10000

data.head()





# In[11]:









# In[18]:



#决策树

#将连续变量分箱

data['income_bin'] = pd.qcut(data.income_new,5)

data['attractive_bin'] = pd.qcut(data.attractive,5)

data['assets_bin'] = pd.qcut(data.assets,5)

data['income_new'].astype('int64').groupby(data['income_bin']).agg(['count', 'mean'])





# In[48]:



#卡方检验

feature_selection.chi2(data[[ 'income_new','attractive','assets','edueduclass']], data['Dated'])





# In[58]:



#划分训练集

target = data['Dated']

model_data = data[['income_new','attractive','assets','edueduclass']]

train_data,test_data,train_target,test_target = cross_validation.train_test_split(model_data,target,test_size = 0.4,train_size = 0.6,random_state =12345)





# In[60]:



# 选择决策树进行建模

import sklearn.tree as tree

clf = tree.DecisionTreeClassifier(criterion='entropy', max_depth=5, min_samples_split=10) 

clf.fit(train_data, train_target)





# In[69]:



# 查看模型预测结果

train_set = clf.predict(train_data)

train_set_p = clf.predict_proba(train_data)[:,1]

test_set = clf.predict(test_data) 

test_set_p = clf.predict_proba(test_data)[:,1]

pd.DataFrame({'test_target':test_target,'test_set':test_set,'test_set_p':test_set_p})





# In[70]:



print(metrics.confusion_matrix(test_target, test_set,labels=[0,1]))  # 混淆矩阵

print(metrics.classification_report(test_target, test_set))  # 计算评估指标

print(pd.DataFrame(list(zip(data.columns, clf.feature_importances_))))  # 变量重要性指标





# In[72]:



red, blue = sns.color_palette("Set1", 2)

sns.distplot(test_set_p[test_target == 1], kde=False, bins=15, color=red)

sns.distplot(test_set_p[test_target == 0], kde=False, bins=15,color=blue)

plt.show()





# In[73]:



fpr_test, tpr_test, th_test = metrics.roc_curve(test_target, test_set_p)

fpr_train, tpr_train, th_train = metrics.roc_curve(train_target, train_set_p)

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

plt.plot(fpr_test, tpr_test, color=blue)

plt.plot(fpr_train, tpr_train, color=red)

plt.title('ROC curve')

plt.show()





# In[74]:



#参数调优

from sklearn.model_selection import GridSearchCV

from sklearn import metrics





# In[80]:



param_grid = {

    'max_depth':[2,3,4,5,7,9,10],

    'min_samples_split':[4,8,12,16,20,24,28,34] 

}

clf = tree.DecisionTreeClassifier(criterion='entropy')

clfcv = GridSearchCV(estimator=clf, param_grid=param_grid, 

                   scoring='roc_auc', cv=4)

clfcv.fit(train_data, train_target)





# In[81]:



train_set = clfcv.predict(train_data)  

train_set_p=clfcv.predict_proba(train_data)[:,1]  

test_set=clfcv.predict(test_data)  

test_set_p=clfcv.predict_proba(test_data)[:,1] 

fpr_test, tpr_test, th_test = metrics.roc_curve(test_target, test_set_p)

fpr_train, tpr_train, th_train = metrics.roc_curve(train_target, train_set_p)

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

plt.plot(fpr_test, tpr_test, color=blue)

plt.plot(fpr_train, tpr_train, color=red)

plt.title('ROC curve')

plt.show()





# In[91]:



clfcv.best_params_

clf = tree.DecisionTreeClassifier(criterion='entropy', max_depth=3, min_samples_split=4) # 当前支持计算信息增益和GINI

clf.fit(train_data, train_target)  #  使用训练数据建模





# In[92]:



import pydotplus

from IPython.display import Image

import sklearn.tree as tree





# In[93]:



dot_data = tree.export_graphviz(

    clf, 

    out_file=None, 

    feature_names=train_data.columns,

    max_depth=3,

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

    filled=True

) 

            

graph = pydotplus.graph_from_dot_data(dot_data)  

Image(graph.create_png()) 





# In[94]:



#神经网络

from sklearn.model_selection import train_test_split





# In[101]:



data_new = data[['income','attractive','assets','edueduclass']]

target = data['Dated']

train_data,test_data,train_target,test_target = train_test_split(data_new,target,test_size = 0.4,train_size = 0.6,random_state = 123)





# In[126]:



from sklearn.preprocessing import MinMaxScaler



scaler = MinMaxScaler()

scaler.fit(train_data)



scaled_train_data = scaler.transform(train_data)

scaled_test_data = scaler.transform(test_data)

train_data.head()





# In[141]:



from sklearn.neural_network import MLPClassifier

mlp = MLPClassifier(hidden_layer_sizes=(10,), 

                    activation='logistic', alpha=0.1, max_iter=1000)



mlp.fit(scaled_train_data, train_target)

mlp





# In[147]:



train_predict = mlp.predict(scaled_train_data)

test_predict = mlp.predict(scaled_test_data)

train_proba = mlp.predict_proba(scaled_train_data)[:, 1]  

test_proba = mlp.predict_proba(scaled_test_data)[:, 1]





# In[150]:



from sklearn import metrics



print(metrics.confusion_matrix(test_target, test_predict, labels=[0, 1]))

print(metrics.classification_report(test_target, test_predict))





# In[151]:





fpr_test, tpr_test, th_test = metrics.roc_curve(test_target, test_proba)

fpr_train, tpr_train, th_train = metrics.roc_curve(train_target, train_proba)



plt.figure(figsize=[4, 4])

plt.plot(fpr_test, tpr_test, '')

plt.plot(fpr_train, tpr_train, '')

plt.title('ROC curve')

plt.show()



print('AUC = %6.4f' %metrics.auc(fpr_test, tpr_test))