如何做GridSearchCV选择Skorch中的BEST_SCORE

Question

我有一个教练课，并且我定义了我的分数方法，

class Trainer(skorch.NeuralNet):
     """
     Other methods such as train_step_single,validation_step,infer,predict e.tc.
     """
        def score(self, X, Y):
           true_text_list,pred_text_list =[],[]
           for src,tgt in zip(X,Y):
               true_text, pred_text = self.predict(src,tgt)
               true_text_list.append(true_text),pred_text_list.append(pred_text)
           result = bleu.corpus_score(pred_text_list,[true_text_list])
           score = result.score
           print(score)
           return score

初始化了培训师类，

epoch_bleu = EpochScoring(scoring=None, lower_is_better=False)

trainer = Trainer(
            module = Seq2SeqTransformer,
            module__num_encoder_layers = 1,
            module__num_decoder_layers = 1,
            module__emb_size = 8,
            module__nhead = 2,
            module__src_vocab_size = SRC_VOCAB_SIZE,
            module__tgt_vocab_size =TGT_VOCAB_SIZE,
            module__dim_feedforward = 8,
            module__dropout = 0.2,
            criterion = criterion,
            optimizer=torch.optim.Adam,
            lr=0.001,
            batch_size = 5, 
            dataset= translate_dataset,
            optimizer__betas=(0.9, 0.98),
            optimizer__eps=1e-9,
            train_split= split,
            max_epochs=3,
            callbacks=[epoch_bleu],
            device=DEVICE,
           )

我使用fit方法

trainer.fit(translate_dataset)
gs = GridSearchCV(trainer,params,scoring= None)
gs.fit(train_X ,train_y)

结果是我的结果

 0.026192733573541495
  epoch    score    train_loss    valid_loss     dur
-------  -------  ------------  ------------  ------
      1   0.0262       10.7645       10.7903  1.1093
0.026192733573541495
      2   0.0262       10.7644       10.7906  1.0927
0.026192733573541495
      3   0.0262       10.7645       10.7903  1.0986
0.026192733573541495

print(gs.best_score_, gs.best_params_)
0.010477093429416598 {'lr': 0.003, 'module__dropout': 0.2}

是我的gs.cv_results _

{'mean_fit_time': array([11.94453502, 13.77196174, 14.05028057, 14.40144386, 14.16046715,
        12.6896481 , 11.84360743, 15.2292047 , 12.22255855]),
 'std_fit_time': array([0.2618251 , 1.07753732, 2.36808454, 1.23219222, 1.86023803,
        1.01203765, 0.28616016, 0.74719985, 0.61968229]),
 'mean_score_time': array([2.27710428, 2.91913524, 2.32345171, 2.53260927, 2.43208547,
        2.31732111, 2.22918859, 2.57366171, 2.31969309]),
 'std_score_time': array([0.1341744 , 0.34804386, 0.1126    , 0.28583804, 0.15907173,
        0.25955979, 0.0675016 , 0.30772863, 0.16703222]),
 'param_lr': masked_array(data=[0.001, 0.001, 0.001, 0.002, 0.002, 0.002, 0.003, 0.003,
                    0.003],
              mask=[False, False, False, False, False, False, False, False,
                    False],
        fill_value='?',
             dtype=object),
 'param_module__dropout': masked_array(data=[0.1, 0.2, 0.3, 0.1, 0.2, 0.3, 0.1, 0.2, 0.3],
              mask=[False, False, False, False, False, False, False, False,
                    False],
        fill_value='?',
             dtype=object),
 'params': [{'lr': 0.001, 'module__dropout': 0.1},
  {'lr': 0.001, 'module__dropout': 0.2},
  {'lr': 0.001, 'module__dropout': 0.3},
  {'lr': 0.002, 'module__dropout': 0.1},
  {'lr': 0.002, 'module__dropout': 0.2},
  {'lr': 0.002, 'module__dropout': 0.3},
  {'lr': 0.003, 'module__dropout': 0.1},
  {'lr': 0.003, 'module__dropout': 0.2},
  {'lr': 0.003, 'module__dropout': 0.3}],
 'split0_test_score': array([0., 0., 0., 0., 0., 0., 0., 0., 0.]),
 'split1_test_score': array([0., 0., 0., 0., 0., 0., 0., 0., 0.]),
 'split2_test_score': array([0.        , 0.        , 0.        , 0.        , 0.        ,
        0.        , 0.        , 0.03447158, 0.        ]),
 'split3_test_score': array([0.02619273, 0.        , 0.02619273, 0.        , 0.        ,
        0.        , 0.        , 0.        , 0.        ]),
 'split4_test_score': array([0.02619273, 0.        , 0.        , 0.        , 0.        ,
        0.        , 0.        , 0.        , 0.        ]),
 'mean_test_score': array([0.01047709, 0.        , 0.00523855, 0.        , 0.        ,
        0.        , 0.        , 0.00689432, 0.        ]),
 'std_test_score': array([0.01283177, 0.        , 0.01047709, 0.        , 0.        ,
        0.        , 0.        , 0.01378863, 0.        ]),
 'rank_test_score': array([1, 4, 3, 4, 4, 4, 4, 2, 4], dtype=int32)}

best_score _不使用我的分数方法选择最佳分数。我对我做错了什么有些困惑。

我的问题是

1. gridSearchCV使用哪些函数来选择最佳分数，如果我使用评分= none，它使用我的score（）函数在培训师中定义班级。
2. 有什么方法可以使GridSearchCV使用我定义的方法选择最佳分数。