def main(generator):
trn = generator.dataset
tst = YelpDataset(args.test_data, vocab)
trainer = BaseSeqTrainer(generator, trn, generator.device, args)
if args.save_ans:
if args.load:
print(trainer.evaluate(tst, args.load))
else:
logger.error("No model parameter configured!")
exit(0)
else:
trainer.train(tst, args.load)
def main(generator):
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
handlers=[
logging.FileHandler("{0}/{1}.log".format("log", args.logger)),
logging.StreamHandler()
])
logger.info('logger start')
trn = generator.dataset
tst = YelpDataset(args.test_data, vocab)
trainer = Seq2SeqTrainer(generator, trn, generator.device, args)
if args.save_ans:
if args.load:
print(trainer.evaluate(tst, args.load))
else:
logger.error("No model parameter configured!")
exit(0)
else:
trainer.train(tst, args.load)
def main(generator):
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
handlers=[
logging.FileHandler("{0}/{1}.log".format("log", args.logger)),
logging.StreamHandler()
])
logger = logging.getLogger(__name__)
logger.info('logger start')
g = generator
dataset = g.data_set
test_dataset = YelpDataset(args.test_data, dataset.vocab, args.test_tree)
logger.info("start network")
trainer = GeneratorTrainer(g, dataset, g.device, args)
if args.save_ans:
if args.load:
trainer.evaluate(test_dataset, args.load)
else:
logger.error("No model parameter configured!")
exit(0)
else:
trainer.train(test_dataset, args.load)
model = model.cuda()
print(args)
I = torch.zeros(args.batch_size, args.attention_hops, args.attention_hops)
for i in range(args.batch_size):
for j in range(args.attention_hops):
I.data[i][j][j] = 1
if args.cuda:
I = I.cuda()
criterion = nn.CrossEntropyLoss()
print('Begin to load data.')
import joblib
if args.load:
model.load_state_dict(torch.load(args.load))
if args.cuda:
model = model.cuda()
print('-' * 89)
if args.model == 'tree':
data_val = YelpDataset(args.test_data, vocab, args.test_tree)
cw_tree_attack(data_val)
elif args.model == 'random_attack':
data_val = YelpDataset(args.test_data, vocab)
print("random attack!")
cw_rand_words_attack(data_val)
else:
data_val = YelpDataset(args.test_data, vocab)
cw_word_attack(data_val)
print(trainer.evaluate(tst, args.load))
else:
logger.error("No model parameter configured!")
exit(0)
else:
trainer.train(tst, args.load)
if __name__ == '__main__':
args = get_args()
PAD_WORD = '<pad>'
UNK_WORD = '<unk>'
EOS_WORD = '<eos>'
SOS_WORD = '<sos>'
vocab = Vocab(filename=args.dictionary,
data=[PAD_WORD, UNK_WORD, EOS_WORD, SOS_WORD])
PAD = vocab.getIndex(PAD_WORD)
device = torch.device("cuda:0" if args.cuda else "cpu")
print('Loading word vectors from', args.word_vector)
embed = torch.load(args.word_vector)
trn = YelpDataset(args.train_data, vocab)
encoder = EncoderRNN(vocab, embed.size(1), args.nhid, device)
decoder = Decoder(embed.size(1), args.nhid, vocab.size())
generator = Seq2SeqGenerator(encoder, decoder, embed=embed,
dataset=trn).to(device)
main(generator)
logger.info("start network")
trainer = GeneratorTrainer(g, dataset, g.device, args)
if args.save_ans:
if args.load:
trainer.evaluate(test_dataset, args.load)
else:
logger.error("No model parameter configured!")
exit(0)
else:
trainer.train(test_dataset, args.load)
if __name__ == '__main__':
args = get_args()
PAD_WORD = '<pad>'
UNK_WORD = '<unk>'
EOS_WORD = '<eos>'
SOS_WORD = '<sos>'
vocab = Vocab(filename=args.dictionary,
data=[PAD_WORD, UNK_WORD, EOS_WORD, SOS_WORD])
PAD = vocab.getIndex(PAD_WORD)
device = torch.device("cuda:0" if args.cuda else "cpu")
print('Loading word vectors from', args.word_vector)
embed = torch.load(args.word_vector)
trn = YelpDataset(args.train_data, vocab, args.train_tree)
generator = Generator(args.train_data, vocab, embed, trn)
main(generator)