def test(args):
source_vocab = Vocab.load(args.model_path + SRC_VOCAB_NAME)
target_vocab = Vocab.load(args.model_path + TAR_VOCAB_NAME)
vocab_size, hidden_size, maxout_hidden_size, embed_size = Backup.load(
args.model_path + HPARAM_NAME)
att_encdec = ABED(vocab_size, hidden_size, maxout_hidden_size, embed_size)
if args.use_gpu:
att_encdec.to_gpu()
serializers.load_hdf5(args.model_path + str(args.epochs) + '.attencdec',
att_encdec)
with open(args.output + str(args.epochs), 'w') as fp:
source_gen = word_list(args.source)
target_gen = word_list(args.target)
batch_gen = batch(sort(source_gen, target_gen, 100 * args.minibatch),
args.minibatch)
for source_batch, target_batch in batch_gen:
source_batch = fill_batch_end(source_batch)
target_batch = fill_batch_end(target_batch)
if args.beam_search:
hyp_batch = forward_beam(source_batch, None, source_vocab,
target_vocab, att_encdec, False,
args.limit, args.beam_size)
else:
hyp_batch = forward(source_batch, None, source_vocab,
target_vocab, att_encdec, False,
args.limit)
for i, hyp in enumerate(hyp_batch):
hyp.append(END)
hyp = hyp[:hyp.index(END)]
show(source_batch[i], target_batch[i], hyp, "TEST")
fwrite(source_batch[i], target_batch[i], hyp, fp)
def test(args):
source_vocab = Vocab.load(args.model_path+SRC_VOCAB_NAME)
target_vocab= Vocab.load(args.model_path+TAR_VOCAB_NAME)
vocab_size, hidden_size, maxout_hidden_size, embed_size = Backup.load(args.model_path+HPARAM_NAME)
att_encdec = ABED(vocab_size, hidden_size, maxout_hidden_size, embed_size)
if args.use_gpu:
att_encdec.to_gpu()
serializers.load_hdf5(args.model_path+str(args.epochs)+'.attencdec', att_encdec)
with open(args.output+str(args.epochs), 'w') as fp:
source_gen = word_list(args.source)
target_gen = word_list(args.target)
batch_gen = batch(sort(source_gen, target_gen, 100*args.minibatch), args.minibatch)
for source_batch, target_batch in batch_gen:
source_batch = fill_batch_end(source_batch)
target_batch = fill_batch_end(target_batch)
if args.beam_search:
hyp_batch = forward_beam(source_batch, None, source_vocab, target_vocab, att_encdec, False, args.limit, args.beam_size)
else:
hyp_batch = forward(source_batch, None, source_vocab, target_vocab, att_encdec, False, args.limit)
for i, hyp in enumerate(hyp_batch):
hyp.append(END)
hyp = hyp[:hyp.index(END)]
show(source_batch[i], target_batch[i], hyp, "TEST")
fwrite(source_batch[i], target_batch[i], hyp, fp)
def train(args):
source_vocab = Vocab(args.source, args.vocab)
target_vocab = Vocab(args.target, args.vocab)
att_encdec = ABED(args.vocab, args.hidden_size, args.maxout_hidden_size,
args.embed_size)
if args.use_gpu:
att_encdec.to_gpu()
if args.source_validation:
if os.path.exists(PLOT_DIR) == False: os.mkdir(PLOT_DIR)
fp_loss = open(PLOT_DIR + "loss", "w")
fp_loss_val = open(PLOT_DIR + "loss_val", "w")
opt = optimizers.AdaDelta(args.rho, args.eps)
opt.setup(att_encdec)
opt.add_hook(optimizer.WeightDecay(DECAY_COEFF))
opt.add_hook(optimizer.GradientClipping(CLIP_THR))
for epoch in xrange(args.epochs):
print "--- epoch: %s/%s ---" % (epoch + 1, args.epochs)
source_gen = word_list(args.source)
target_gen = word_list(args.target)
batch_gen = batch(sort(source_gen, target_gen, 100 * args.minibatch),
args.minibatch)
n = 0
total_loss = 0.0
for source_batch, target_batch in batch_gen:
n += len(source_batch)
source_batch = fill_batch_end(source_batch)
target_batch = fill_batch_end(target_batch)
hyp_batch, loss = forward(source_batch, target_batch, source_vocab,
target_vocab, att_encdec, True, 0)
total_loss += loss.data * len(source_batch)
closed_test(source_batch, target_batch, hyp_batch)
loss.backward()
opt.update()
print "[n=%s]" % (n)
print "[total=%s]" % (n)
prefix = args.model_path + '%s' % (epoch + 1)
serializers.save_hdf5(prefix + '.attencdec', att_encdec)
if args.source_validation:
total_loss_val, n_val = validation_test(args, att_encdec,
source_vocab, target_vocab)
fp_loss.write("\t".join([str(epoch), str(total_loss / n) + "\n"]))
fp_loss_val.write("\t".join(
[str(epoch), str(total_loss_val / n_val) + "\n"]))
fp_loss.flush()
fp_loss_val.flush()
hyp_params = att_encdec.get_hyper_params()
Backup.dump(hyp_params, args.model_path + HPARAM_NAME)
source_vocab.save(args.model_path + SRC_VOCAB_NAME)
target_vocab.save(args.model_path + TAR_VOCAB_NAME)
hyp_params = att_encdec.get_hyper_params()
Backup.dump(hyp_params, args.model_path + HPARAM_NAME)
source_vocab.save(args.model_path + SRC_VOCAB_NAME)
target_vocab.save(args.model_path + TAR_VOCAB_NAME)
if args.source_validation:
fp_loss.close()
fp_loss_val.close()
def train(args):
source_vocab = Vocab(args.source, args.vocab)
target_vocab = Vocab(args.target, args.vocab)
att_encdec = ABED(args.vocab, args.hidden_size, args.maxout_hidden_size, args.embed_size)
if args.use_gpu:
att_encdec.to_gpu()
if args.source_validation:
if os.path.exists(PLOT_DIR)==False: os.mkdir(PLOT_DIR)
fp_loss = open(PLOT_DIR+"loss", "w")
fp_loss_val = open(PLOT_DIR+"loss_val", "w")
opt = optimizers.AdaDelta(args.rho, args.eps)
opt.setup(att_encdec)
opt.add_hook(optimizer.WeightDecay(DECAY_COEFF))
opt.add_hook(optimizer.GradientClipping(CLIP_THR))
for epoch in xrange(args.epochs):
print "--- epoch: %s/%s ---"%(epoch+1, args.epochs)
source_gen = word_list(args.source)
target_gen = word_list(args.target)
batch_gen = batch(sort(source_gen, target_gen, 100*args.minibatch), args.minibatch)
n = 0
total_loss = 0.0
for source_batch, target_batch in batch_gen:
n += len(source_batch)
source_batch = fill_batch_end(source_batch)
target_batch = fill_batch_end(target_batch)
hyp_batch, loss = forward(source_batch, target_batch, source_vocab, target_vocab, att_encdec, True, 0)
total_loss += loss.data*len(source_batch)
closed_test(source_batch, target_batch, hyp_batch)
loss.backward()
opt.update()
print "[n=%s]"%(n)
print "[total=%s]"%(n)
prefix = args.model_path + '%s'%(epoch+1)
serializers.save_hdf5(prefix+'.attencdec', att_encdec)
if args.source_validation:
total_loss_val, n_val = validation_test(args, att_encdec, source_vocab, target_vocab)
fp_loss.write("\t".join([str(epoch), str(total_loss/n)+"\n"]))
fp_loss_val.write("\t".join([str(epoch), str(total_loss_val/n_val)+"\n"]))
fp_loss.flush()
fp_loss_val.flush()
hyp_params = att_encdec.get_hyper_params()
Backup.dump(hyp_params, args.model_path+HPARAM_NAME)
source_vocab.save(args.model_path+SRC_VOCAB_NAME)
target_vocab.save(args.model_path+TAR_VOCAB_NAME)
hyp_params = att_encdec.get_hyper_params()
Backup.dump(hyp_params, args.model_path+HPARAM_NAME)
source_vocab.save(args.model_path+SRC_VOCAB_NAME)
target_vocab.save(args.model_path+TAR_VOCAB_NAME)
if args.source_validation:
fp_loss.close()
fp_loss_val.close()
def validation_test(args, encdec, src_vocab, tar_vocab):
src_gen = word_list(args.source_validation)
tar_gen = word_list(args.target_validation)
batch_gen = batch(sort(src_gen, tar_gen, 100*args.minibatch), args.minibatch)
total_loss = 0.0
n = 0
for src_batch, tar_batch in batch_gen:
n += len(src_batch)
src_batch= fill_batch_end(src_batch)
tar_batch = fill_batch_end(tar_batch)
hyp_batch, loss = forward(src_batch, tar_batch, src_vocab, tar_vocab, encdec, True, 0)
total_loss += loss.data*len(src_batch)
for i, hyp in enumerate(hyp_batch):
hyp.append(END)
hyp = hyp[:hyp.index(END)]
tar = tar_batch[i][:tar_batch[i].index(END)]
show(src_batch[i], tar, hyp, "VALIDATION")
return total_loss, n
def validation_test(args, encdec, src_vocab, tar_vocab):
src_gen = word_list(args.source_validation)
tar_gen = word_list(args.target_validation)
batch_gen = batch(sort(src_gen, tar_gen, 100 * args.minibatch),
args.minibatch)
total_loss = 0.0
n = 0
for src_batch, tar_batch in batch_gen:
n += len(src_batch)
src_batch = fill_batch_end(src_batch)
tar_batch = fill_batch_end(tar_batch)
hyp_batch, loss = forward(src_batch, tar_batch, src_vocab, tar_vocab,
encdec, True, 0)
total_loss += loss.data * len(src_batch)
for i, hyp in enumerate(hyp_batch):
hyp.append(END)
hyp = hyp[:hyp.index(END)]
tar = tar_batch[i][:tar_batch[i].index(END)]
show(src_batch[i], tar, hyp, "VALIDATION")
return total_loss, n