def __init__(self, config, char2id, tag2id, emb_matrix):
super(NERLSTM_CRF, self).__init__()
self.hidden_dim = config.hidden_dim
self.vocab_size = len(char2id)
self.seg_dize = 5
self.tag_to_ix = tag2id
self.tagset_size = len(tag2id)
""" pdding_idx=0，也就是让pad标记不更新 """
self.char_emb = nn.Embedding.from_pretrained(emb_matrix,
freeze=False,
padding_idx=0)
self.seg_emb = nn.Embedding(self.seg_size,
config.seg_dim,
padding_idx=0)
self.emb_dim = config.char_dim + config.seg_dim
self.dropout = nn.Dropout(config.dropout)
self.lstm = nn.LSTM(self.emb_dim,
self.hidden_dim // 2,
num_layers=1,
bidirectional=True,
batch_first=True)
""" 得到发射矩阵 """
self.hidden2tag = nn.Linear(self.hidden_dim, self.tagset_size)
self.crf = CRF(self.tagset_size, batch_first=True)
class NERLSTM_CRF(nn.Module):
def __init__(self, config, char2id, tag2id, emb_matrix, device):
super(NERLSTM_CRF, self).__init__()
self.hidden_dim = config.hidden_dim
self.vocab_size = len(char2id)
self.seg_size = 5
self.tag_to_ix = tag2id
self.tagset_size = len(tag2id)
""" pdding_idx=0，也就是让pad标记不更新 """
self.char_emb = nn.Embedding.from_pretrained(
emb_matrix,freeze=False, padding_idx=0
)
self.seg_emb = nn.Embedding(
self.seg_size, config.seg_dim, padding_idx=0
)
self.emb_dim = config.char_dim + config.seg_dim
self.dropout = nn.Dropout(config.dropout)
self.lstm = nn.LSTM(
self.emb_dim, self.hidden_dim // 2, num_layers=2,
bidirectional=True, batch_first=True
)
""" 得到发射矩阵 """
self.hidden2tag = nn.Linear(self.hidden_dim, self.tagset_size)
self.crf = CRF(tag2id, self.tagset_size, device,batch_first=True)
def forward(self, char_ids,seg_ids,mask=None):
""" 把字向量（100维）和词长度特征向量（20维），拼接 """
embedding = torch.cat(
(self.char_emb(char_ids),self.seg_emb(seg_ids)), 2
)
outputs, hidden = self.lstm(embedding)
outputs = self.dropout(outputs)
outputs = self.hidden2tag(outputs)
""" 预测时，得到维特比解码的路径 """
return self.crf.decode(outputs, mask)
def _get_lstm_features(self, char_ids, seg_ids):
embedding = torch.cat(
(self.char_emb(char_ids), self.seg_emb(seg_ids)), 2
)
# embedding:batch,seq,hidden: 128,37,129
outputs, hidden = self.lstm(embedding)
outputs = self.dropout(outputs)
outputs = self.hidden2tag(outputs)
return outputs
def neg_log_likelihood(self, char_ids, seg_ids, tag_ids, mask=None):
feats = self._get_lstm_features(char_ids, seg_ids)
""" 训练时，得到损失 """
return self.crf(feats, tag_ids, mask)
if args.model_type == 'single':
test = test.to_col()
shuffle_test = shuffle_test.to_col()
org_tests.append(test)
shuffle_tests.append(shuffle_test)
val_dataset = ConcatDataset(org_tests)
t2 = time()
print("Done ({} sec.)".format(int(t2 - t1)))
# create models
classifier = build_sherlock(sherlock_feature_groups,
num_classes=len(valid_types),
topic_dim=topic_dim).to(device)
model = CRF(len(valid_types), batch_first=True).to(device)
if args.model_type == 'single':
# load pre-trained model
model_loc = join(os.environ['BASEPATH'], 'model',
'pre_trained_sherlock', TYPENAME)
classifier.load_state_dict(
torch.load(join(model_loc, args.model_path), map_location=device))
classifier.eval()
# eval
with torch.no_grad():
result_list = []
# get base accuracy
report_b = eval_batch_col(classifier, val_dataset, batch_size,