XGBoost评分功能与培训和评估不符

Question

当我使用score()时，我的训练损失（损失，用于验证和培训）不匹配分数。它是一个二进制分类（Y是一个0或1的1d数组）。我想score()正在使用我提供的loss_function。如果没有，我想知道如何更改此指标。

                model = CatBoostClassifier(iterations=iterations,
                                           early_stopping_rounds=early_stopping_rounds, 
                                           learning_rate=learning_rate,
                                           loss_function='Logloss',
                                           depth=depth)
                model.fit(X_train_cat.iloc[batch*batch_size:(batch+1)*batch_size],
                              y_train.iloc[batch*batch_size:(batch+1)*batch_size],
                              eval_set=(X_val_cat, y_val),
                              cat_features=index_categ_feat,
                              use_best_model=True)

输出

...
99: learn: 0.3745852    test: 0.4720675 best: 0.4720675 (99)    total: 292ms    remaining: 0us

bestTest = 0.4720675457
bestIteration = 99

Out[45]: <catboost.core.CatBoostClassifier at 0x210aa5f25f8>

model.score(X_val_cat, y_val)
Out[46]: 0.8064516129032258

model.score(X_train_cat, y_train)
Out[47]: 0.9014014014014013