lgb交叉验证

def Model_lgb(train,test,features,target,params,n_estimators):
    X_train = train[features].values
    y_train = train[target].values
    X_test = test[features].values

    folds = KFold(n_splits=5, shuffle=True, random_state=1)
    oof_lgb = np.zeros((X_train.shape[0]))
    predictions = np.zeros((len(test)))
    lgb_importance = []

    for fold_, (trn_idx, val_idx) in enumerate(folds.split(X_train, y_train)):
        print("fold n°{}".format(fold_+1))
        print(X_train[trn_idx].shape,type(X_train))
        x_tr,y_tr = X_train[trn_idx],y_train[trn_idx]
        x_va,y_va = X_train[val_idx],y_train[val_idx]

        trn_data = lgb.Dataset(x_tr, y_tr)
        val_data = lgb.Dataset(x_va, y_va)
        clf = lgb.train(params, trn_data, num_boost_round = n_estimators, valid_sets = [trn_data, val_data], verbose_eval=1, early_stopping_rounds = 50)#,
        oof_lgb[val_idx] = clf.predict(x_va, num_iteration=clf.best_iteration)
        lgb_importance.append(clf.feature_importance())
        predictions += clf.predict(X_test, num_iteration=clf.best_iteration) / folds.n_splits
    print('valid_ACC = {}'.format(abs(y_train-oof_lgb).mean()))
    gc.collect()
    return predictions,lgb_importance,oof_lgb