from typing import Optional, Tuple, Union import torch from torch import nn from torch.nn import CrossEntropyLoss, BCEWithLogitsLoss, MSELoss from transformers import BertPreTrainedModel, BertModel, BertForSequenceClassification from transformers.modeling_outputs import SequenceClassifierOutput from trc_model.temporal_relation_classification_config import TemporalRelationClassificationConfig class TokenPooler(nn.Module): def __init__(self, config): super().__init__() self.dense = nn.Linear(config.hidden_size, config.hidden_size) self.activation = nn.Tanh() def forward(self, token_tensor: torch.Tensor) -> torch.Tensor: # We "pool" the model by simply taking the hidden state corresponding # to the first token. pooled_output = self.dense(token_tensor) pooled_output = self.activation(pooled_output) return pooled_output class TemporalRelationClassification(BertForSequenceClassification): config_class = TemporalRelationClassificationConfig def __init__(self, config): super().__init__(config) self.num_labels = config.num_labels self.special_markers = config.special_markers self.pool_tokens = config.pool_tokens self.ES_ID = config.ES_ID self.EMS1 = config.EMS1 self.EMS2 = config.EMS2 self.architecture = config.architecture self.config = config self.bert = BertModel.from_pretrained(config.base_lm) if self.bert.config.vocab_size != config.vocab_size: self.bert.resize_token_embeddings(config.vocab_size) classifier_dropout = ( config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob ) if config.pool_tokens: self.ems_1_pooler = TokenPooler(config) self.ems_2_pooler = TokenPooler(config) self.e_1_pooler = TokenPooler(config) self.e_2_pooler = TokenPooler(config) self.dropout = nn.Dropout(classifier_dropout) self.classification_layers = None if self.architecture == 'SEQ_CLS': self.classification_layers = nn.Sequential( nn.Linear(config.hidden_size, config.num_labels) ) if self.architecture == 'EMP': self.e_1_linear = nn.Linear(config.hidden_size * 2, config.hidden_size) self.e_2_linear = nn.Linear(config.hidden_size * 2, config.hidden_size) if self.architecture in ['ESS', 'EF', 'EMP']: self.classification_layers = nn.Sequential( nn.Linear(config.hidden_size * 2, config.hidden_size), nn.Linear(config.hidden_size, config.num_labels) ) # Initialize weights and apply final processing # self.post_init() def _get_entities_and_start_markers_indices(self, input_ids): if not self.special_markers: event_1_start, event_2_start = torch.tensor( [(ids == self.ES_ID).nonzero().squeeze().tolist() for ids in input_ids]).T return event_1_start, event_1_start + 1, event_2_start, event_2_start + 1 em1_s = torch.tensor([(ids == self.EMS1).nonzero().item() for ids in input_ids], device=self.device) entity_1 = em1_s + 1 em2_s = torch.tensor([(ids == self.EMS2).nonzero().item() for ids in input_ids], device=self.device) entity_2 = em2_s + 1 return em1_s, entity_1, em2_s, entity_2 def forward( self, input_ids: Optional[torch.Tensor] = None, attention_mask: Optional[torch.Tensor] = None, token_type_ids: Optional[torch.Tensor] = None, position_ids: Optional[torch.Tensor] = None, head_mask: Optional[torch.Tensor] = None, inputs_embeds: Optional[torch.Tensor] = None, labels: Optional[torch.Tensor] = None, output_attentions: Optional[bool] = None, output_hidden_states: Optional[bool] = None, return_dict: Optional[bool] = None, ) -> Union[Tuple[torch.Tensor], SequenceClassifierOutput]: r""" labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If `config.num_labels > 1` a classification loss is computed (Cross-Entropy). """ return_dict = return_dict if return_dict is not None else self.config.use_return_dict 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, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) logits = None if self.architecture == 'SEQ_CLS': pooled_output = outputs[1] pooled_output = self.dropout(pooled_output) logits = self.classification_layers(pooled_output) else: sequence_output = outputs[0] sequence_output = self.dropout(sequence_output) entity_mark_1_s, entity_1, entity_mark_2_s, entity_2 = self._get_entities_and_start_markers_indices( input_ids) e1_start_mark_tensors = sequence_output[torch.arange(sequence_output.size(0)), entity_mark_1_s] e2_start_mark_tensors = sequence_output[torch.arange(sequence_output.size(0)), entity_mark_2_s] e1_tensor = sequence_output[torch.arange(sequence_output.size(0)), entity_1] e2_tensor = sequence_output[torch.arange(sequence_output.size(0)), entity_2] if self.pool_tokens: e1_start_mark_tensors = self.ems_1_pooler(e1_start_mark_tensors) e2_start_mark_tensors = self.ems_2_pooler(e2_start_mark_tensors) e1_tensor = self.e_1_pooler(e1_tensor) e2_tensor = self.e_2_pooler(e2_tensor) if self.architecture == 'ESS': e_start_markers_cat = torch.cat((e1_start_mark_tensors, e2_start_mark_tensors), 1) logits = self.classification_layers(e_start_markers_cat) if self.architecture == 'EF': events_cat = torch.cat((e1_tensor, e2_tensor), 1) logits = self.classification_layers(events_cat) if self.architecture == 'EMP': e1_and_start_mark = self.e_1_linear(torch.cat((e1_start_mark_tensors, e1_tensor), 1)) e2_and_start_mark = self.e_2_linear(torch.cat((e2_start_mark_tensors, e2_tensor), 1)) both_e_cat = torch.cat((e1_and_start_mark, e2_and_start_mark), 1) logits = self.classification_layers(both_e_cat) loss = None if labels is not None: loss_fct = CrossEntropyLoss() loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1)) if not return_dict: output = (logits,) + outputs[2:] return ((loss,) + output) if loss is not None else output return SequenceClassifierOutput( loss=loss, logits=logits, hidden_states=outputs.hidden_states, attentions=outputs.attentions, )