《Python金融大資料風控模組化實戰》 第17章 內建學習
- 本章引言
- Python代碼實作及注釋
本章引言
內建學習旨在通過訓練多個模型，擴充假設空間，進而逐漸接近真實資料集中蘊含的規則。同時，多個訓練模型同時陷入局部最小值的機率較低，保證了測試集可以得到相對較優的結果。
目前，內建學習大緻可分為兩種：并行的內建方法Bagging和串行的內建方法Boosting。并行的內建方法中，基學習器的建構是互相獨立的，沒有先後順序，可以同時進行模組化。而串行的內建方法中，各個基學習器之間有強烈的依賴關系，即後一個模型是在前一個模型的基礎上建立的。內建學習的核心是優勢互補，是以如何增加基學習器的獨立性和多樣性是內建學習的關鍵，不同的算法有不同的政策。
Python代碼實作及注釋
# 第17章：內建學習
import os
import sys
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
import variable_encode as var_encode
from sklearn.metrics import confusion_matrix,recall_score, auc, roc_curve,precision_score,accuracy_score
from sklearn.model_selection import GridSearchCV
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import AdaBoostClassifier
from sklearn.ensemble import GradientBoostingClassifier
from xgboost import XGBClassifier
import matplotlib.pyplot as plt
import matplotlib
matplotlib.rcParams['font.sans-serif']=['SimHei']
matplotlib.rcParams['axes.unicode_minus']=False
import warnings
warnings.filterwarnings("ignore") ##忽略警告
##資料讀取
def data_read(data_path,file_name):
df = pd.read_csv( os.path.join(data_path, file_name), delim_whitespace = True, header = None )
##變量重命名
columns = ['status_account','duration','credit_history','purpose', 'amount',
'svaing_account', 'present_emp', 'income_rate', 'personal_status',
'other_debtors', 'residence_info', 'property', 'age',
'inst_plans', 'housing', 'num_credits',
'job', 'dependents', 'telephone', 'foreign_worker', 'target']
df.columns = columns
##将标簽變量由狀态1,2轉為0,1;0表示好使用者，1表示壞使用者
df.target = df.target - 1
##資料分為data_train和 data_test兩部分，訓練集用于得到編碼函數，驗證集用已知的編碼規則對驗證集編碼
data_train, data_test = train_test_split(df, test_size=0.2, random_state=0,stratify=df.target)
return data_train, data_test
##離散變量與連續變量區分
def category_continue_separation(df,feature_names):
categorical_var = []
numerical_var = []
if 'target' in feature_names:
feature_names.remove('target')
##先判斷類型，如果是int或float就直接作為連續變量
numerical_var = list(df[feature_names].select_dtypes(include=['int','float','int32','float32','int64','float64']).columns.values)
categorical_var = [x for x in feature_names if x not in numerical_var]
return categorical_var,numerical_var
if __name__ == '__main__':
path = 'D:\\code\\chapter17'
data_path = os.path.join(path ,'data')
file_name = 'german.csv'
##讀取資料
data_train, data_test = data_read(data_path,file_name)
sum(data_train.target ==0)
data_train.target.sum()
##區分離散變量與連續變量
feature_names = list(data_train.columns)
feature_names.remove('target')
categorical_var,numerical_var = category_continue_separation(data_train,feature_names)
###離散變量直接WOE編碼
var_all_bin = list(data_train.columns)
var_all_bin.remove('target')
##訓練集WOE編碼
df_train_woe, dict_woe_map, dict_iv_values ,var_woe_name = var_encode.woe_encode(data_train,data_path,categorical_var, data_train.target,'dict_woe_map', flag='train')
##測試集WOE編碼
df_test_woe, var_woe_name = var_encode.woe_encode(data_test,data_path,categorical_var, data_test.target, 'dict_woe_map',flag='test')
#####連續變量缺失值做填補
for i in numerical_var:
if sum(data_train[i].isnull()) >0:
data_train[i].fillna(data_train[i].mean(),inplace=True)
###組成分箱後的訓練集與測試集
data_train.reset_index(drop=True,inplace=True)
data_test.reset_index(drop=True,inplace=True)
var_1 = numerical_var
var_1.append('target')
data_train_1 = pd.concat([df_train_woe[var_woe_name],data_train[var_1]],axis=1)
data_test_1 = pd.concat([df_test_woe[var_woe_name],data_test[var_1]],axis=1)
####取出訓練資料與測試資料
var_all = list(data_train_1.columns)
var_all.remove('target')
####變量歸一化
scaler = StandardScaler().fit(data_train_1[var_all])
data_train_1[var_all] = scaler.transform(data_train_1[var_all])
data_test_1[var_all] = scaler.transform(data_test_1[var_all])
x_train = np.array(data_train_1[var_all])
y_train = np.array(data_train_1.target)
x_test = np.array(data_test_1[var_all])
y_test = np.array(data_test_1.target)
########随機森林模型
##設定待優化的超參數
rf_param = {'n_estimators': list(range(50, 400, 50)),
'max_depth': list(range(2, 10, 1)),
'class_weight': [{1: 1, 0: 1}, {1: 2, 0: 1}, {1: 3, 0: 1}]}
##初始化網格搜尋
rf_gsearch = GridSearchCV(estimator=RandomForestClassifier(random_state=0, criterion='entropy',
max_features=0.8, bootstrap=True),
param_grid=rf_param, cv=3, scoring='f1', n_jobs=-1, verbose=2)
##執行超參數優化
rf_gsearch.fit(x_train, y_train)
print('RandomForest model best_score_ is {0},and best_params_ is {1}'.format(rf_gsearch.best_score_,
rf_gsearch.best_params_))
##模型訓練
##用最優參數，初始化随機森林模型
RF_model_2 = RandomForestClassifier(random_state=0, n_jobs=-1, criterion='entropy',
n_estimators=rf_gsearch.best_params_['n_estimators'],
max_depth=rf_gsearch.best_params_['max_depth'],
max_features=0.8,
min_samples_split=50,
class_weight=rf_gsearch.best_params_['class_weight'],
bootstrap=True)
##訓練随機森林模型
RF_model_2_fit = RF_model_2.fit(x_train, y_train)
##屬性
# ss = RF_model_2_fit.estimators_
# RF_model_2_fit.classes_
# RF_model_2_fit.n_features_
# RF_model_2_fit.feature_importances_
##模型預測
y_pred = RF_model_2_fit.predict(x_test)
##計算混淆矩陣與recall、precision
cnf_matrix = confusion_matrix(y_test, y_pred)
recall_value = recall_score(y_test, y_pred)
precision_value = precision_score(y_test, y_pred)
acc = accuracy_score(y_test, y_pred)
print(cnf_matrix)
print('Validation set: model recall is {0},and percision is {1}'.format(recall_value,
precision_value))
##變量重要性排序
##取出變量重要性結果
len_1 = len(RF_model_2.feature_importances_)
var_name = [str(x).split('_woe')[0] for x in var_all]
##重要性繪圖
plt.figure(figsize=(10,6))
fontsize_1 = 14
plt.barh(np.arange(0,20),RF_model_2.feature_importances_,color =
'c',tick_label=var_name)
plt.xticks( fontsize=fontsize_1)
plt.yticks( fontsize=fontsize_1)
plt.show()
###Adaboost模型
##設定待優化的超參數
ada_param = {'n_estimators': list(range(50, 500, 50)),
'learning_rate': list(np.arange(0.1, 1, 0.2))}
##初始化網格搜尋
ada_gsearch = GridSearchCV(estimator=AdaBoostClassifier(algorithm='SAMME.R',random_state=0),
param_grid=ada_param, cv=3, n_jobs=-1, verbose=2)
##執行超參數優化
ada_gsearch.fit(x_train, y_train)
print('AdaBoostClassifier model best_score_ is {0},and best_params_ is {1}'.format(ada_gsearch.best_score_,
ada_gsearch.best_params_))
##模型訓練
##用最優參數，初始化Adaboost模型
ada_model_2 = AdaBoostClassifier( n_estimators=ada_gsearch.best_params_['n_estimators'],
learning_rate=ada_gsearch.best_params_['learning_rate'],
algorithm='SAMME.R',random_state=0)
##訓練adaboost模型
ada_model_2_fit = ada_model_2.fit(x_train, y_train)
##模型預測
y_pred = ada_model_2_fit.predict(x_test)
##計算混淆矩陣與recall、precision
cnf_matrix = confusion_matrix(y_test, y_pred)
recall_value = recall_score(y_test, y_pred)
precision_value = precision_score(y_test, y_pred)
acc = accuracy_score(y_test, y_pred)
print(cnf_matrix)
print('Validation set: model recall is {0},and percision is {1}'.format(recall_value,
precision_value))
###檢視模型訓練過程準确率的變化
##取出訓練過程中模型得分
n_estimators = ada_gsearch.best_params_['n_estimators']
per_train = list(ada_model_2_fit.staged_score(x_train, y_train))
per_test= list(ada_model_2_fit.staged_score(x_test, y_test))
##預測準确率繪圖
plt.figure(figsize=(10,6))
fontsize_1 = 14
plt.plot(np.arange(0,n_estimators),per_train,'--',color ='c',label='訓練集')
plt.plot(np.arange(0,n_estimators),per_test,':',color ='b',label='測試集')
plt.xticks( fontsize=fontsize_1)
plt.yticks( fontsize=fontsize_1)
plt.xlabel('n_estimators',fontsize=fontsize_1)
plt.ylabel('score',fontsize=fontsize_1)
plt.legend(fontsize=fontsize_1)
plt.show()
####GBDT模型
##設定待優化的超參數
gbdt_param = {'n_estimators': list(range(50, 400, 50)),
'max_depth': list(range(2, 5, 1)),
'learning_rate': list(np.arange(0.01, 0.5, 0.05)) }
##初始化網格搜尋
gbdt_gsearch = GridSearchCV(estimator=GradientBoostingClassifier( subsample=0.8,max_features=0.8, validation_fraction=0.1,
n_iter_no_change =3,random_state=0),param_grid=gbdt_param,
cv=3, scoring='f1', n_jobs=-1, verbose=2)
##執行超參數優化
gbdt_gsearch.fit(x_train, y_train)
print('gbdt model best_score_ is {0},and best_params_ is {1}'.format(gbdt_gsearch.best_score_,
gbdt_gsearch.best_params_))
##模型訓練
##用最優參數，初始化GBDT模型
GBDT_model= GradientBoostingClassifier(subsample=0.8,max_features=0.8, validation_fraction=0.1,
n_iter_no_change =3,random_state=0 ,
n_estimators=gbdt_gsearch.best_params_['n_estimators'],
max_depth=gbdt_gsearch.best_params_['max_depth'],
learning_rate=gbdt_gsearch.best_params_['learning_rate'])
##訓練GBDT模型
GBDT_model_fit = GBDT_model.fit(x_train, y_train)
###看一下混沌矩陣
y_pred = GBDT_model_fit.predict(x_test)
cnf_matrix = confusion_matrix(y_test, y_pred)
recall_value = recall_score(y_test, y_pred)
precision_value = precision_score(y_test, y_pred)
acc = accuracy_score(y_test, y_pred)
print(cnf_matrix)
print('Validation set: model recall is {0},and percision is {1}'.format(recall_value,
precision_value))
###xgboost模型
##設定待優化的超參數
xgb_param = {'max_depth': list(range(2, 6, 1)), 'min_child_weight': list(range(1, 4, 1)),
'learning_rate': list(np.arange(0.01, 0.3, 0.05)), 'scale_pos_weight': list(range(1, 5, 1))}
##初始化網格搜尋
xgb_gsearch = GridSearchCV(
estimator=XGBClassifier(random_state=0, n_estimators=500, subsample=0.8, colsample_bytree=0.8),
param_grid=xgb_param, cv=3, scoring='f1', n_jobs=-1, verbose=2)
##執行超參數優化
xgb_gsearch.fit(x_train, y_train)
print('xgboost model best_score_ is {0},and best_params_ is {1}'.format(xgb_gsearch.best_score_,
xgb_gsearch.best_params_))
##用最優參數，初始化xgboost模型
xgboost_model = XGBClassifier(random_state=0, n_jobs=-1,
n_estimators=500,
max_depth=xgb_gsearch.best_params_['max_depth'],
subsample=0.8, colsample_bytree=0.8,
learning_rate=xgb_gsearch.best_params_['learning_rate'],
scale_pos_weight=xgb_gsearch.best_params_['scale_pos_weight'])
##訓練xgboost模型
xgboost_model_fit = xgboost_model.fit(x_train, y_train)
##模型預測
y_pred = xgboost_model_fit.predict(x_test)
##計算混淆矩陣與recall、precision
cnf_matrix = confusion_matrix(y_test, y_pred)
recall_value = recall_score(y_test, y_pred)
precision_value = precision_score(y_test, y_pred)
acc = accuracy_score(y_test, y_pred)
print(cnf_matrix)
print('Validation set: model recall is {0},and percision is {1}'.format(recall_value,
precision_value))
##給出機率預測結果
y_score_test = xgboost_model_fit.predict_proba(x_test)[:, 1]
##計算AR。gini等
fpr, tpr, thresholds = roc_curve(y_test, y_score_test)
roc_auc = auc(fpr, tpr)
ks = max(tpr - fpr)
ar = 2*roc_auc-1
print('test set: model AR is {0},and ks is {1},auc={2}'.format(ar,
ks,roc_auc))
####ks曲線
plt.figure(figsize=(10,6))
fontsize_1 = 12
plt.plot(np.linspace(0,1,len(tpr)),tpr,'--',color='black')
plt.plot(np.linspace(0,1,len(tpr)),fpr,':',color='black')
plt.plot(np.linspace(0,1,len(tpr)),tpr - fpr,'-',color='grey')
plt.grid()
plt.xticks( fontsize=fontsize_1)
plt.yticks( fontsize=fontsize_1)
plt.xlabel('機率分組',fontsize=fontsize_1)
plt.ylabel('累積占比%',fontsize=fontsize_1)
![XGBoost Plotting API以及GBDT組合特征實踐 XGBoost Plotting API以及GBDT組合特征實踐[圖]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)