lgbm输出特征重要性

lgb_clf = lgb.LGBMClassifier(objective='binary',num_leaves=35,max_depth=6,learning_rate=0.05,seed=2018,
        colsample_bytree=0.8,subsample=0.9,n_estimators=20000)
lgb_model = lgb_clf.fit(train_x[features], train_x[target], eval_set=[(test_x[features], test_x[target])], early_stopping_rounds=200)
lgb_predictors = [i for i in train_x[features].columns]
lgb_feat_imp = pd.Series(lgb_model.feature_importances_, lgb_predictors).sort_values(ascending=False)
lgb_feat_imp.to_csv('lgb_feat_imp.csv')

然而特征重要性的结果并不是很可靠，也不能反应特征相互组合对logloss的影响。故我们使用warpper的方式来进行特征选择。将前向搜索、后向搜索和随机搜索进行组合筛选出最终特征。

Ref：https://zhuanlan.zhihu.com/p/32749489