八大直播作业-第七讲-Aringrhus

本次作业主要增加了对于变量的处理和分析，但是实际模型效果还是不太理想。

应该是某个环节还存在问题，需要进一步跟大家沟通交流学习。

# coding: utf-8



# In[1]:





import numpy as np

import pandas as pd

import sklearn.metrics as metrics

from sklearn.cross_validation import train_test_split

from sklearn.model_selection import GridSearchCV





# In[2]:





fraud_data = pd.read_excel(r'FRAUD_TRAIN_Samp.xlsx')

fraud_data.head()





# In[3]:





y = fraud_data['target']

X = fraud_data.iloc[:,2:]





# In[4]:





X.shape





# In[5]:





missing= (X.isnull().sum()/X.shape[0]).sort_values(ascending=False)

missing





# In[6]:





# 删除缺失率0.9以上的变量

X_nomissing = X.drop(missing[missing>0.9].index,axis=1)

missing = missing.drop(missing[missing>0.9].index)

X_nomissing.shape





# In[7]:





# 缺失率小于0.7的变量以均值填补

for idx in missing[(missing<0.7)].index:

#     print(idx)

    if idx in('F75', 'F81','F74'):

        vm = X_nomissing[idx].mode()[0]

    else:

        vm = X_nomissing[idx].mean()

    X_nomissing[idx].fillna(vm,inplace = True)





# In[8]:





# 缺失率大于0.7的变量生成是否缺失标识，并以均值填补

for idx in missing[(missing>0.7)].index:

    new_col = idx + 'is_na'

    X_nomissing[new_col] = X_nomissing[idx].map(bool).map(int)

    vmean = X_nomissing[idx].mean()

    X_nomissing[idx].fillna(vmean,inplace = True)





# In[9]:





map_dict = {}

for col in ('F75', 'F81','F74'):

    tmp_df = pd.DataFrame(X[col].value_counts())

    tmp_df['num'] = range(1,len(tmp_df)+1)

    tmp_df.columns = ['num',col]

    tmp_df.drop('num',axis=1,inplace = True)

    tmp_dict = tmp_df.to_dict(orient = 'dict')

    map_dict.update(tmp_dict)

    print(tmp_dict)



map_dict





# In[10]:





X_nomissing.replace(map_dict,inplace=True)





# In[11]:





missing2= (X_nomissing.isnull().sum()/X_nomissing.shape[0]).sort_values(ascending=False)

missing2





# In[12]:





X_train,X_test,y_train,y_test = train_test_split(X_nomissing,y,train_size = 0.7,random_state = 123)





# # 决策树



# In[13]:





import sklearn.tree as tree

param_grid = {

    'criterion':['entropy','gini'],

    'max_depth':[4,5,6,7,8,9],

    'min_samples_split':[6,8,10,14,18,20,22,24]

}

clf1 = tree.DecisionTreeClassifier()

clf1cv = GridSearchCV(estimator=clf1,param_grid=param_grid,scoring='roc_auc',cv = 4)

clf1cv.fit(X_train,y_train)





# In[14]:





test_pre = clf1cv.predict(X_test)





# In[15]:





print("decision tree accuracy:")

print(metrics.classification_report(y_test,test_pre))

print("decision tree AUC:")

fpr_test, tpr_test, th_test = metrics.roc_curve(y_test,test_pre)

print('AUC = %.4f' %metrics.auc(fpr_test, tpr_test))





# In[16]:





clf1cv.best_params_





# # ANN



# In[17]:





from sklearn.preprocessing import MinMaxScaler



scaler = MinMaxScaler()

scaler.fit(X_train)



X_train_sc = scaler.transform(X_train)

X_test_sc = scaler.transform(X_test)





# In[19]:





from sklearn.neural_network import MLPClassifier

param_grid = {

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

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

    'alpha':[0.001, 0.01, 0.1, 0.2, 0.3, 0.4,0.6,0.8,1,2,4,6,8,10]

}

mlp = MLPClassifier(max_iter=1000)

gcv = GridSearchCV(estimator=mlp, param_grid=param_grid, 

                   scoring='roc_auc', cv=4, n_jobs=-1)

gcv.fit(X_train_sc, y_train)





# In[20]:





test_pre = clf1cv.predict(X_test_sc)





# In[22]:





print("ANN accuracy:")

print(metrics.classification_report(y_test,test_pre))

print("ANN AUC:")

fpr_test, tpr_test, th_test = metrics.roc_curve(y_test,test_pre)

print('AUC = %.4f' %metrics.auc(fpr_test, tpr_test))





# In[23]:





gcv.best_params_





# # GBDT



# In[25]:





import sklearn.ensemble as ensemble





# In[26]:





param_grid = {

    'loss':['deviance','exponential'],

    'learning_rate':[0.1,0.3,0.5,0.7,1],

    'n_estimators':[10,20,30,40],

    'max_depth':[2,4,6,8,10,12,15],

    'min_samples_split':[8,14,20,22] 

    

}



gbc = ensemble.GradientBoostingClassifier()

gbccv = GridSearchCV(estimator=gbc, param_grid=param_grid, scoring='roc_auc', cv=4)

gbccv.fit(X_train, y_train)





# In[28]:





test_pre = gbccv.predict(X_test)





# In[29]:





print("gradient boosting accuracy:")

print(metrics.classification_report(y_test,test_pre))

print("gradient boosting AUC:")

fpr_test, tpr_test, th_test = metrics.roc_curve(y_test,test_pre)

print('AUC = %.4f' %metrics.auc(fpr_test, tpr_test))





# In[30]:





gbccv.best_params_





# In[ ]: