- 主要思想:通过GridSearchCV算法进行特征的自动化筛选
import numpy as np
import pandas as pd
from time import strftime
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import train_test_split
from sklearn.externals import joblib
from sklearn.metrics import classification_report
from sklearn.metrics import roc_curve
from sklearn.metrics import auc
def selectParam(clf,param,features=[]):
"""
param:希望对模型进行调优的参数;
feature:RF模型调参时默认对所有特征进行处理;LR模型调参时选择RF筛选留下的特征
"""
grid_search = GridSearchCV(estimator = clf, param_grid=param, n_jobs=4,scoring='roc_auc',cv=5)
x1 = np.asarray(self.x_train)
if len(features) > 0:
x1 = np.asarray(self.x_train[features])
y1 = np.asarray(self.y_train)
grid_search.fit(x1, y1)
best_parameters = grid_search.best_estimator_.get_params()
return best_parameters
def top_features(self):
"""
step1:调整参数获得随机森林最优参数;
step2: 通过最优参数建模并获得rf_model
step3: 通过rf_model获得特征重要性
返回值:特征及其重要性并其按重要性排序的输出
"""
threshold =0.005
clf_RF = RandomForestClassifier(random_state=10)
param_grid =[ {"max_depth": range(10,50,3),
"n_estimators": range(10,50,3),
}
]
best_parameters = self.selectParam(clf_RF,param_grid)
best_parameters.pop('random_state')
clf_RF_prior = RandomForestClassifier(random_state=10,**best_parameters)
clf_RF_prior.fit(self.x_train,self.y_train)
print("the best paramers of RF to choose features resulted in score : ", clf_RF_prior.score(self.x_test, self.y_test))
feature_imp = pd.DataFrame([self.x_train.columns,clf_RF_prior.feature_importances_]).T
feature_imp.columns = ['features','features_importance']
feature_imp = feature_imp.sort_values(['features_importance'],ascending=False)
feature_imp = feature_imp[feature_imp['features_importance']>threshold]
return feature_imp