XGBRegressor 损失自定义函数

Question

我想在 XGBRegressor 中将损失函数自定义为分位数损失（pinball 损失）

我用这个代码

def xgb_quantile_eval(preds, dmatrix, quantile=0.2):
    labels = dmatrix.get_label()
    return ('q{}_loss'.format(quantile),
            np.nanmean((preds >= labels) * (1 - quantile) * (preds - labels) +
                       (preds < labels) * quantile * (labels - preds)))


def xgb_quantile_obj(preds, dmatrix, quantile=0.2):
    try:
        assert 0 <= quantile <= 1
    except AssertionError:
        raise ValueError("Quantile value must be float between 0 and 1.")

    labels = dmatrix.get_label()
    errors = preds - labels

    left_mask = errors < 0
    right_mask = errors > 0

    grad = -quantile * left_mask + (1 - quantile) * right_mask
    hess = np.ones_like(preds)

    return grad, hess

然后我像这样构建模型

def XGB(q, X_train, Y_train, X_valid, Y_valid, X_test):    
    # (a) Modeling  
    model = XGBRegressor(objective=xgb_quantile_obj, alpha=q,
                     n_estimators=10000, bagging_fraction=0.7, learning_rate=0.027, subsample=0.7)                 
                     
    model.fit(X_train, Y_train, eval_metric = xgb_quantile_eval, 
          eval_set=[(X_valid, Y_valid)], early_stopping_rounds=300, verbose=500)

    # (b) Predictions
    pred = pd.Series(model.predict(X_test).round(2))
    return model, pred

但是我得到一个错误

models_2, results_2 = XGB(0.5, X_train_1, Y_train_1, X_valid_1, Y_valid_1, X_test)
results_2

AttributeError: 'numpy.ndarray' 对象没有属性 'get_label'

我不确定自己过得好不好。请帮助我

Answer 1

哦，我发现错误了，我必须更改 xgb_quantile_obj 中 preds 和 dmatrix 之间的顺序，并将 dmatrix 更改为 labels

XGBRegressor 损失自定义函数

XGBRegressor loss functio custom

xgboost