kaggle google Quest比赛代码阅读笔记

 

关于抽取bert里面第几层的代码:

(

＃我们取零，因为据我了解，这就是[CLS]令牌...
＃想法是也要合并最后4层而不是最后一层，因为它太接近输出了
＃层，它可能没有那么有效，因为它受到o / p的更多控制。

)

https://www.kaggle.com/c/google-quest-challenge/discussion/123770

class BertForSequenceClassification_v2(BertPreTrainedModel):
    r"""
        **labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size,)``:
    Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
        **loss**: (`optional`, returned when ``labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
            Classification (or regression if config.num_labels==1) loss.
        **logits**: ``torch.FloatTensor`` of shape ``(batch_size, config.num_labels)``
            Classification (or regression if config.num_labels==1) scores (before SoftMax).
        **hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
            list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
            of shape ``(batch_size, sequence_length, hidden_size)``:
            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        outputs = model(input_ids, labels=labels)
        loss, logits = outputs[:2]
    """
    def __init__(self, config):

        super(BertForSequenceClassification_v2, self).__init__(config)

       # config.output_hidden_states=True (make sure)
        self.num_labels = config.num_labels
        self.bert = BertModel(config)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        self.classifier = nn.Linear(config.hidden_size, self.config.num_labels)
        self.init_weights()

        def forward(self, input_ids=None, attention_mask=None, token_type_ids=None,
                position_ids=None, head_mask=None, inputs_embeds=None, labels=None, 
                extra_feats=None):

        outputs = self.bert(input_ids,
                            attention_mask=attention_mask,
                            token_type_ids=token_type_ids,
                            position_ids=position_ids,
                            head_mask=head_mask,
                            inputs_embeds=inputs_embeds)

        # sequence_output = outputs[0]
        # pooled_output = outputs[1]

        hidden_states = outputs[2] #hidden_states: 12 layers tuples each is of (batch_size, sequence_length, hidden_size) + embedding``
        # print(seq[-1].shape, seq[-1][:, 0].shape)

        # we are taking zero because in my understanding that's the [CLS] token...
        # idea is to pool last 4 layers as well instead of just the last one, since it's too close to the output
        # layers, it might not be that efficient as it's more regulated by the o/p's..

        h12 = hidden_states[-1][:, 0].reshape((-1, 1, 768))
        h11 = hidden_states[-2][:, 0].reshape((-1, 1, 768))
        h10 = hidden_states[-3][:, 0].reshape((-1, 1, 768))
        h9  = hidden_states[-4][:, 0].reshape((-1, 1, 768))

        all_h = torch.cat([h9, h10, h11, h12], 1) #Also don't forget to add the last CLS token seq_op/pooled_op as you wish..
        mean_pool = torch.mean(all_h, 1)

        pooled_output = self.dropout(mean_pool)
        logits = self.classifier(pooled_output)
        outputs = (logits,) + outputs[2:]  # add hidden states and attention if they are here

        return outputs  # (loss), logits, (hidden_states), (attentions)