lightgbm中如何使用自定义函数

 

sklearn  API

目前有自定义函数：

def self_metric1(y_true,y_predict):
    d = pd.DataFrame()
    d['prob'] = list(y_predict)
    d['y'] = list(y_true)
    d = d.sort_values(['prob'], ascending=[0])
    y = d.y
    PosAll = pd.Series(y).value_counts()[1]
    NegAll = pd.Series(y).value_counts()[0]
    pCumsum = d['y'].cumsum()
    nCumsum = np.arange(len(y)) - pCumsum + 1
    pCumsumPer = pCumsum / PosAll
    nCumsumPer = nCumsum / NegAll
    TR1 = pCumsumPer[abs(nCumsumPer-0.001).idxmin()]
    TR2 = pCumsumPer[abs(nCumsumPer-0.005).idxmin()]
    TR3 = pCumsumPer[abs(nCumsumPer-0.01).idxmin()]

    return "self_metric", 0.4 * TR1 + 0.3 * TR2 + 0.3 * TR3, True

使用方法：

clf = LGBMClassifier(
    learning_rate=0.05,
    n_estimators=10230,
    num_leaves=31,
    subsample=0.8,
    colsample_bytree=0.8,
    random_state=1023,
    metric=None
)

clf.fit(
            trn_x, trn_y,
            eval_set=[(val_x, val_y)],
    #         categorical_feature=cate_cols
            eval_metric=self_metric1,#eval_metric=lambda y_true, y_pred: [self_metric1(y_true, y_pred)],
            early_stopping_rounds=200,
            verbose=200
        )

 

 

 

xgboost的用法类似