中文命名实体识别，采用bilstm+crf模型基于Pytorch实现

import torch import torch.nn as nn import torch.optim as optim import numpy as np START_TAG = "<START>" STOP_TAG = "<STOP>" EMBEDDING_DIM = 5 HIDDEN_DIM = 4 def prepare_sequence(seq, to_ix): idxs = [to_ix[w] for w in seq] return torch.tensor(idxs, dtype=torch.long) class BiLSTM_CRF(nn.Module): def __init__(self, vocab_size, tag_to_ix, embedding_dim, hidden_dim): super(BiLSTM_CRF, self).__init__() self.embedding_dim = embedding_dim self.hidden_dim = hidden_dim self.vocab_size = vocab_size self.tag_to_ix = tag_to_ix self.tagset_size = len(tag_to_ix) self.word_embeds = nn.Embedding(vocab_size, embedding_dim) self.lstm = nn.LSTM(embedding_dim, hidden_dim // 2, num_layers=1, bidirectional=True) # Maps the output of the LSTM into tag space. self.hidden2tag = nn.Linear(hidden_dim, self.tagset_size) # Matrix of transition parameters. Entry i,j is the score of # transitioning *to* i *from* j. self.transitions = nn.Parameter( torch.randn(self.tagset_size, self.tagset_size)) # These two statements enforce the constraint that we never transfer # to the start tag and we never transfer from the stop tag self.transitions.data[tag_to_ix[START_TAG], :] = -10000 self.transitions.data[:, tag_to_ix[STOP_TAG]] = -10000 self.hidden = self.init_hidden() def init_hidden(self): return (torch.randn(2, 1, self.hidden_dim // 2), torch.randn(2, 1, self.hidden_dim // 2)) def _forward_alg(self, feats): # Do the forward algorithm to compute the partition function init_alphas = torch.full([self.tagset_size], -10000.) # START_TAG has all of the score. init_alphas[self.tag_to_ix[START_TAG]] = 0. # Wrap in a variable so that we will get automatic backprop # Iterate through the sentence forward_var_list=[] forward_var_list.append(init_alphas) for feat_index in range(feats.shape[0]): gamar_r_l = torch.stack([forward_var_list[feat_index]] * feats.shape[1]) t_r1_k = torch.unsqueeze(feats[feat_index],0).transpose(0,1) aa = gamar_r_l + t_r1_k + self.transitions forward_var_list.append(torch.logsumexp(aa,dim=1)) terminal_var = forward_var_list[-1] + self.transitions[self.tag_to_ix[STOP_TAG]] terminal_var = torch.unsqueeze(terminal_var,0) alpha = torch.logsumexp(terminal_var, dim=1)[0] return alpha def _get_lstm_features(self, sentence): self.hidden = self.init_hidden() embeds = self.word_embeds(sentence).view(len(sentence), 1, -1) lstm_out, self.hidden = self.lstm(embeds, self.hidden) lstm_out = lstm_out.view(len(sentence), self.hidden_dim) lstm_feats = self.hidden2tag(lstm_out) return lstm_feats def _score_sentence(self, feats, tags): # Gives the score of a provided tag sequence score = torch.zeros(1) tags = torch.cat([torch.tensor([self.tag_to_ix[START_TAG]], dtype=torch.long), tags]) for i, feat in enumerate(feats): score = score + \ self.transitions[tags[i + 1], tags[i]] + feat[tags[i + 1]] score = score + self.transitions[self.tag_to_ix[STOP_TAG], tags[-1]] return score def _viterbi_decode(self, feats): """ feats: [batch_size, seq_len, n_tag] :param feats: :return: """ backpointers = [] # Initialize the viterbi variables in log space init_vvars = torch.full((1, self.tagset_size), -10000.) init_vvars[0][self.tag_to_ix[START_TAG]] = 0 # forward_var at step i holds the viterbi variables for step i-1 forward_var_list = [] forward_var_list.append(init_vvars) for feat_index in range(feats.shape[0]): gamar_r_l = torch.stack([forward_var_list[feat_index]] * feats.shape[1]) gamar_r_l = torch.squeeze(gamar_r_l) next_tag_var = gamar_r_l + self.transitions viterbivars_t, bptrs_t = torch.max(next_tag_var,dim=1) t_r1_k = torch.unsqueeze(feats[feat_index], 0) forward_var_new = torch.unsqueeze(viterbivars_t,0) + t_r1_k forward_var_list.append(forward_var_new) backpointers.append(bptrs_t.tolist()) # Transition to STOP_TAG terminal_var = forward_var_list[-1] + self.transitions[self.tag_to_ix[STOP_TAG]] best_tag_id = torch.argmax(terminal_var).tolist() path_score = terminal_var[0][best_tag_id] # Follow the back pointers to decode the best path. best_path = [best_tag_id] for bptrs_t in reversed(backpointers): best_tag_id = bptrs_t[best_tag_id] best_path.append(best_tag_id) # Pop off the start tag (we dont want to return that to the caller) start = best_path.pop() assert start == self.tag_to_ix[START_TAG] # Sanity check best_path.reverse() return path_score, best_path def _viterbi_decode_parallel(self, feats): feats = feats.unsqueeze(dim=0) backpointers = [] init_vvars = torch.full((feats.shape[0], self.tagset_size), -10000.) init_vvars[:, self.tag_to_ix[START_TAG]] = 0 # forward_var at step i holds the viterbi variables for step i-1 forward_var_list = [] forward_var_list.append(init_vvars) for feat_index in range(feats.shape[1]): # gamr_r_l: [batch_size, n_tag, n_tag] gamar_r_l = torch.stack([forward_var_list[feat_index]]*feats.shape[2], dim=1) next_tag_var = gamar_r_l + self.transitions.unsqueeze(dim=0) viterbivars_t, bptrs_t = torch.max(next_tag_var, dim=2) # values, idx, [batch_size, n_tag] # feats: [batch_size, seq_len, n_tag] t_r1_k = feats[:, feat_index, :] # [batch_size, n_tag] forward_var_new = viterbivars_t + t_r1_k forward_var_list.append(forward_var_new) backpointers.append(np.array(bptrs_t.tolist())) # [batch_size, n_tag] terminal_var = forward_var_list[-1] + torch.unsqueeze(self.transitions[self.tag_to_ix[STOP_TAG]], 0) best_tag_id = torch.argmax(terminal_var, dim=1).tolist() path_score = terminal_var[range(terminal_var.shape[0]), best_tag_id] best_path = [best_tag_id] for bptrs_t in reversed(backpointers): best_tag_id = bptrs_t[range(bptrs_t.shape[0]), best_tag_id] best_path.append(best_tag_id.tolist()) # best_path: [seq_len+1, batch_size] start = best_path.pop() assert start[0] == self.tag_to_ix[START_TAG] best_path.reverse() best_path = np.array(best_path).transpose().tolist() return path_score, best_path def neg_log_likelihood(self, sentence, tags): feats = self._get_lstm_features(sentence) forward_score = self._forward_alg(feats) gold_score = self._score_sentence(feats, tags) return forward_score - gold_score def forward(self, sentence): # dont confuse this with _forward_alg above. # Get the emission scores from the BiLSTM lstm_feats = self._get_lstm_features(sentence) # Find the best path, given the features. score, tag_seq = self._viterbi_decode(lstm_feats) print(score, tag_seq, "*") score, tag_seq = self._viterbi_decode_parallel(lstm_feats) print(score, tag_seq, "*****") return score, tag_seq if __name__== '__main__': START_TAG = "<START>" STOP_TAG = "<STOP>" EMBEDDING_DIM = 300 HIDDEN_DIM = 256 # Make up some training data training_data = [( "the wall street journal reported today that apple corporation made money".split(), "B I I I O O O B I O O".split() ), ( "georgia tech is a univ