Python catboost二进制分类：训练和测试日志损失计算不一致？

我正在使用Catboost解决一个二进制分类问题。我的数据集有5列（所有5列都是类别）。我将数据集分为两组（X1，y1）和（X2，y2）。这是我的密码：

import catboost
catboost_model = catboost.CatBoostClassifier(iterations=150, learning_rate=0.5, depth=4,
                        loss_function='Logloss', eval_metric='Logloss', thread_count=4) 
catboost_model.fit(X1, y1, cat_features=X_columns, eval_set=[(X1, y1)], verbose=True, use_best_model=True)

请注意，我的验证集（测试集）与训练集相同。我原以为列车和试验的对数损耗计算相同，但以下是输出（前10行）：

为什么“学习”logloss与“测试”logloss不同？我使用的是与列车和测试相同的（X1，y1）

0:  learn: 0.4785917    test: 0.4785909 best: 0.4785909 (0) total: 137ms    remaining: 20.4s
1:  learn: 0.4050160    test: 0.4049414 best: 0.4049414 (1) total: 184ms    remaining: 13.7s
2:  learn: 0.3747240    test: 0.3746158 best: 0.3746158 (2) total: 213ms    remaining: 10.5s
3:  learn: 0.3604948    test: 0.3603708 best: 0.3603708 (3) total: 243ms    remaining: 8.87s
4:  learn: 0.3536384    test: 0.3535104 best: 0.3535104 (4) total: 272ms    remaining: 7.88s
5:  learn: 0.3474158    test: 0.3450169 best: 0.3450169 (5) total: 310ms    remaining: 7.44s
6:  learn: 0.3448010    test: 0.3412129 best: 0.3412129 (6) total: 342ms    remaining: 6.99s
7:  learn: 0.3416549    test: 0.3383577 best: 0.3383577 (7) total: 376ms    remaining: 6.67s
8:  learn: 0.3398553    test: 0.3363817 best: 0.3363817 (8) total: 407ms    remaining: 6.38s
9:  learn: 0.3391258    test: 0.3357344 best: 0.3357344 (9) total: 442ms    remaining: 6.19s
...