def main():
global opt
train_from = opt.train_from
if opt.train_from:
train_from = True
checkpoint = torch.load(opt.train_from)
_ep = opt.epochs
opt = checkpoint['opt']
opt.epochs = _ep
print("Loading dictrionary from ", opt.dict)
vocab_dict = torch.load(opt.dict)
print("Loading train data from ", opt.traindata)
train_data = torch.load(opt.traindata)
print("Loading valid data from ", opt.validdata)
valid_data = torch.load(opt.validdata)
train_dataset = reader.Dataset(train_data, opt.batch_size, True)
valid_dataset = reader.Dataset(valid_data,
opt.batch_size,
True,
volatile=True)
print(' * vocabulary size = %d' % (vocab_dict.size()))
print(' * number of training samples. %d' % len(train_data['answers']))
print(' * maximum batch size. %d' % opt.batch_size)
print('Building model...')
model = reader.AoAReader(vocab_dict,
dropout_rate=opt.dropout,
embed_dim=opt.embed_size,
hidden_dim=opt.gru_size)
# no way on CPU
model.cuda()
if train_from:
print('Loading model from checkpoint at %s' % opt.train_from)
chk_model = checkpoint['model']
model.load_state_dict(chk_model)
opt.start_epoch = checkpoint['epoch'] + 1
optimizer = torch.optim.Adam(model.parameters(),
lr=opt.learning_rate,
weight_decay=opt.weight_decay)
if train_from:
optimizer.load_state_dict(checkpoint['optimizer'])
nParams = sum([p.nelement() for p in model.parameters()])
print('* number of parameters: %d' % nParams)
trainModel(model, train_dataset, valid_dataset, optimizer)
def main():
print("Loading dict", testopt.dict)
vocab_dict = torch.load(testopt.dict)
print("Loading test data")
test_data = torch.load(testopt.testdata)
print("Loading model from ", testopt.model)
ckp = torch.load(testopt.model)
opt = ckp['opt']
model_state = ckp['model']
if opt.gpu:
torch.cuda.set_device(opt.gpu)
test_dataset = reader.Dataset(test_data,
opt.batch_size,
True,
volatile=True)
print(' * vocabulary size = %d' % (vocab_dict.size()))
print(' * number of test samples. %d' % len(test_data['candidates']))
print(' * maximum batch size. %d' % opt.batch_size)
print('Building model...')
model = reader.AoAReader(vocab_dict,
dropout_rate=opt.dropout,
embed_dim=opt.embed_size,
hidden_dim=opt.gru_size)
# no way on CPU
model.cuda()
# load state
model.load_state_dict(model_state)
print('Evaluate on test data')
answers = evalulate(model, test_dataset, vocab_dict)
with open(testopt.out, 'w') as out:
print('\n'.join(answers), file=out)