回归LightGBM Cross Validation

使用LightGBM，Cross Validation进行回归，RMSE loss。

import lightgbm as lgb
import matplotlib.pyplot as plt
import numpy as np
import random
import os
from lightgbm import LGBMClassifier
from sklearn import metrics
from sklearn.metrics import mean_squared_error
from sklearn.metrics import roc_auc_score, roc_curve
from sklearn.metrics import log_loss
from sklearn.model_selection import KFold, StratifiedKFold
from sklearn.preprocessing import LabelEncoder
import seaborn as sns
import gc
def kfold_lightgbm(train_df, test_df, y, num_folds, stratified = False, debug= False):
    if stratified:
        folds = StratifiedKFold(n_splits= num_folds, shuffle=True, random_state=17)
    else:
        folds = KFold(n_splits= num_folds, shuffle=True, random_state=17)
    # Create arrays and dataframes to store results
    oof_preds = np.zeros(train_df.shape[0])
    sub_preds = np.zeros(test_df.shape[0])
    feature_importance_df = pd.DataFrame()
    feats = train.columns.tolist()
    test_df = test_df[feats]
    #test_df = csr_matrix(test_df)
    for n_fold, (train_idx, valid_idx) in enumerate(folds.split(train_df, y)):
        print('FOLD {}'.format(n_fold))
        train_x, train_y = train_df.iloc[train_idx], y.iloc[train_idx]
        valid_x, valid_y = train_df.iloc[valid_idx], y.iloc[valid_idx]

        # LightGBM parameters found by Bayesian optimization
        lgb_params =  {
            'task': 'train',
            'boosting_type': 'gbdt',
            'objective': 'regression',
            'metric': 'rmse',
            #"n_estimators":10000,
            "learning_rate": 0.01,
            
            'num_leaves': 60,
            'subsample': 0.6143,
            'colsample_bytree': 0.6453,
            'min_split_gain': np.power(10, -2.5988),
            'reg_alpha': np.power(10, -2.2887),
            'reg_lambda': np.power(10, 1.7570),
            'min_child_weight': np.power(10, -0.1477),
            'max_depth': -1,
            #'zero_as_missing':True
        }
        '''
        lgb_params = {
            'objective': 'regression',
            'metric': 'rmse',
            'learning_rate': 0.01,
            'num_leaves': 16,
            'max_depth': -1,
            'min_child_samples': 1,
            'max_bin': 300,
            'subsample': 1.0,
            'subsample_freq': 1,
            'colsample_bytree': 0.5,
            'min_child_weight': 10,
            'reg_lambda': 0.1,
            'reg_alpha': 0.0,
            'scale_pos_weight': 1,
            'zero_as_missing': True,
            'num_threads': -1,
        }
        '''
        #train_x = csr_matrix(train_x)
        #valid_x = csr_matrix(valid_x)
        lgtrain = lgb.Dataset(train_x, train_y,
                        feature_name=feats,
                        categorical_feature = 'auto')
        lgvalid = lgb.Dataset(valid_x, valid_y,
                        feature_name=feats,
                        categorical_feature = 'auto')
        clf = lgb.train(
            lgb_params,
            lgtrain,
            num_boost_round=3000,
            valid_sets=[lgtrain, lgvalid],
            valid_names=['train','valid'],
            early_stopping_rounds=200,
            verbose_eval=100
        )
        
       # clf.fit(train_x, train_y, eval_set=[(train_x, train_y), (valid_x, valid_y)], 
            #eval_metric= 'auc', verbose= 100, early_stopping_rounds= 200)

        #oof_preds[valid_idx] = clf.predict_proba(valid_x, num_iteration=clf.best_iteration_)[:, 1]
        #sub_preds += clf.predict_proba(test_df[feats], num_iteration=clf.best_iteration_)[:, 1] / folds.n_splits
        
        oof_preds[valid_idx] = clf.predict(valid_x, num_iteration=clf.best_iteration)
        sub_preds += clf.predict(test_df, num_iteration=clf.best_iteration)/ folds.n_splits
        
        fold_importance_df = pd.DataFrame()
        fold_importance_df["feature"] = feats
        fold_importance_df["importance"] = clf.feature_importance()
        #fold_importance_df["importance"] = clf.feature_importances_
        fold_importance_df["fold"] = n_fold + 1
        feature_importance_df = pd.concat([feature_importance_df, fold_importance_df], axis=0)
        #print('Fold %2d AUC : %.6f' % (n_fold + 1, roc_auc_score(valid_y, oof_preds[valid_idx])))
        print('Fold %2d RMSE : %.6f' % (n_fold, np.sqrt(metrics.mean_squared_error(valid_y, oof_preds[valid_idx]))))      
        del clf, train_x, train_y, valid_x, valid_y
        gc.collect()

    rmse = np.sqrt(metrics.mean_squared_error(y, oof_preds))
    #print('Full AUC score %.6f' % roc_auc_score(y, oof_preds))
    print('Full RMSE score %.6f' % rmse)
    
    display_importances(feature_importance_df)
    return feature_importance_df, sub_preds
# Display/plot feature importance
def display_importances(feature_importance_df_):
    cols = feature_importance_df_[["feature", "importance"]].groupby("feature").mean().sort_values(by="importance", ascending=False)[:40].index
    best_features = feature_importance_df_.loc[feature_importance_df_.feature.isin(cols)]
    plt.figure(figsize=(8, 10))
    sns.barplot(x="importance", y="feature", data=best_features.sort_values(by="importance", ascending=False))
    plt.title('LightGBM Features (avg over folds)')
    plt.tight_layout()
    plt.savefig('lgb_importances.png')