loss = base_model.update(batch)
train_loss += loss
if global_step % opt['log_step'] == 0:
duration = time.time() - start_time
print(
format_str.format(datetime.now(), global_step, max_steps,
epoch, opt['num_epoch'], loss, duration,
current_lr))
# eval on dev
print("Evaluating on dev set...")
predictions = []
dev_loss = 0
for _, batch in enumerate(dev_batch):
preds, _, loss = base_model.predict(batch)
predictions += preds
dev_loss += loss
predictions = [id2label[p] for p in predictions]
dev_p, dev_r, dev_f1 = scorer.score(dev_batch.gold(), predictions)
max_dev_f1, max_dev_id = (dev_f1,
epoch) if max_dev_f1 < dev_f1 else (max_dev_f1,
max_dev_id)
train_loss = train_loss / train_batch.num_examples * opt[
'batch_size'] # avg loss per batch
dev_loss = dev_loss / dev_batch.num_examples * opt['batch_size']
# eval on test
print("Evaluating on test set...")
predictions = []
print("Loading data from {} with batch size {}...".format(
data_file, opt['batch_size']))
batch = DataLoader(data_file,
opt['batch_size'],
opt,
vocab,
evaluation=True,
pattern_file=opt['pattern_file'])
helper.print_config(opt)
id2label = dict([(v, k) for k, v in constant.LABEL_TO_ID.items()])
predictions = []
all_probs = []
for i, b in enumerate(batch):
preds, probs, _ = base_model.predict(b)
predictions += preds
all_probs += probs
predictions = [id2label[p] for p in predictions]
_, _, _ = scorer.score(batch.gold(), predictions, verbose=True)
# save probability scores
# if len(args.out) > 0:
# outfile = 'saved_models/' + args.model_id + '/' + args.out
# with open(outfile, 'w') as fw:
# for prob in all_probs:
# fw.write(json.dumps([round(p, 4) for p in prob]))
# fw.write('\r\n')
# print("Prediction scores saved to {}.".format(outfile))
print("Evaluation ended.")