def start():
# produce_data()
model = Bert_CRF()
model.load_state_dict(load_model(args.output_dir))
device = torch.device(args.device if torch.cuda.is_available()
and not args.no_cuda else "cpu")
model.to(device)
print('create_iter')
eval_iter = create_batch_iter("valid")
print('create_iter finished')
# ------------------判断CUDA模式----------------------
device = torch.device(args.device if torch.cuda.is_available()
and not args.no_cuda else "cpu")
# -----------------------验证----------------------------
model.eval()
count = 0
y_predicts, y_labels = [], []
eval_loss, eval_acc, eval_f1 = 0, 0, 0
with torch.no_grad():
for step, batch in enumerate(eval_iter):
batch = tuple(t.to(device) for t in batch)
input_ids, input_mask, segment_ids, label_ids, output_mask = batch
bert_encode = model(input_ids, segment_ids, input_mask).cpu()
eval_los = model.loss_fn(bert_encode=bert_encode,
tags=label_ids,
output_mask=output_mask)
eval_loss = eval_los + eval_loss
count += 1
predicts = model.predict(bert_encode, output_mask)
predict_tensor = predicts.cpu()
label_tensor = label_ids.cpu()
y_predicts.append(predicts)
y_labels.append(label_ids)
entity_precision, entity_recall, entity_f1 = score_predict(
label_tensor, predict_tensor)
print(
'\n step :%d - eval_loss: %4f - ent_p:%4f - ent_r:%4f - ent_f1:%4f\n'
% (step, eval_loss.item() / count, entity_precision,
entity_recall, entity_f1))
label_ids = label_ids.view(1, -1).squeeze()
predicts = predicts.view(1, -1).squeeze()
label_ids = label_ids[label_ids != -1]
predicts = predicts[predicts != -1]
assert len(label_ids) == len(predicts)
eval_predicted = torch.cat(y_predicts, dim=0).cpu()
eval_labeled = torch.cat(y_labels, dim=0).cpu()
entity_precision, entity_recall, entity_f1 = score_predict(
eval_labeled, eval_predicted)
print(
'\n\n- eval_loss: %4f - eval_acc:%4f - eval_f1:%4f - ent_p:%4f - ent_r:%4f - ent_f1:%4f\n'
% (eval_loss.item() / count, eval_acc, eval_f1, entity_precision,
entity_recall, entity_f1))
def __init__(self):
print('[INFO]加载分词器')
self.processor, self.bertTokenizer = init_params()
label_list = self.processor.get_labels()
self.label_map = {label: i for i, label in enumerate(label_list)}
self.tokenizer = BasicTokenizer()
print('[INFO]分词器加载完毕')
print('[INFO]加载模型')
self.model = Bert_CRF()
self.model.load_state_dict(load_model(args.output_dir))
self.device = torch.device(args.device if torch.cuda.is_available()
and not args.no_cuda else "cpu")
self.model.to(self.device)
self.model.eval()
print('[INFO]模型加载完毕')
def start():
# produce_data()
model = Bert_CRF()
print('create_iter')
train_iter, num_train_steps = create_batch_iter("train")
eval_iter = create_batch_iter("valid")
print('create_iter finished')
epoch_size = num_train_steps * args.train_batch_size * args.gradient_accumulation_steps / args.num_train_epochs
pbar = ProgressBar(epoch_size=epoch_size, batch_size=args.train_batch_size)
# for name, param in model.named_parameters():
# if param.requires_grad:
# print(name)
print('fit')
fit(model=model,
training_iter=train_iter,
eval_iter=eval_iter,
num_epoch=args.num_train_epochs,
pbar=pbar,
num_train_steps=num_train_steps,
verbose=1)