.split(','))
log_template = ' '.join(
'{:>6.0f},{:>5.0f},{:>9.0f},{:>5.0f}/{:<5.0f} {:>7.0f}%,{:>8.6f},{:12.4f},{}'
.split(','))
best_snapshot_prefix = os.path.join(args.save_path, 'best_snapshot')
print(header)
for epoch in range(1, args.epochs + 1):
if early_stop:
print("Early stopping. Epoch: {}, Best Dev. Acc: {}".format(
epoch, best_dev_acc))
break
n_correct, n_total = 0, 0
for batch_idx, batch in enumerate(train_loader.next_batch()):
iterations += 1
model.train()
optimizer.zero_grad()
pos_score, neg_score = model(batch)
n_correct += (torch.sum(torch.gt(pos_score, neg_score),
0).data == neg_score.size(0)).sum()
n_total += pos_score.size(1)
train_acc = 100. * n_correct / n_total
ones = torch.autograd.Variable(
torch.ones(pos_score.size(0) * pos_score.size(1)))
if args.cuda:
ones = ones.cuda()
#-*- coding: utf-8 -*- # Author: QuYingqi # mail: [email protected] # Created Time: 2017-11-09 import sys import torch from seqRankingLoader import SeqRankingLoader import numpy as np sys.path.append('../vocab') rel_vocab = torch.load('../vocab/vocab.rel.pt') neg_range = len(rel_vocab) print(neg_range) word_vocab = torch.load('../vocab/vocab.word.pt') loader = SeqRankingLoader('data/test.relation_ranking.pt', 5, neg_range, 0) batch_size = loader.batch_size for i, batch in enumerate(loader.next_batch(False)): if i >= 1: break seqs, pos_rel, neg_rel = batch seqs_trans = np.transpose(seqs.cpu().data.numpy()) pos_rel_trans = pos_rel.cpu().data.numpy() neg_rel_trans = np.transpose(neg_rel.cpu().data.numpy()) for j in range(5): question = ' '.join(word_vocab.convert_to_word(seqs_trans[j])) print(question) pos_rel_ = rel_vocab.convert_to_word([pos_rel_trans[j]]) print(pos_rel_) neg_rel_ = ' | '.join(rel_vocab.convert_to_word(neg_rel_trans[j])) print(neg_rel_)