def show_samples(self,
dataset,
num_samples=10,
cuda=True,
logger=None,
decoding_strategy='sample',
indices=None,
beam_width=4):
if indices is None:
indices = list()
for _ in range(num_samples):
indices.append(random.randint(0, len(dataset) - 1))
collate_fn = get_collate_fn(cuda)
data = [dataset[ind] for ind in indices]
batch = collate_fn(data)
out = self.forward(batch[0],
decoding_strategy=decoding_strategy,
train=False,
beam_width=beam_width)
generated_utterance = out['utterance'].cpu().data
logger_fn = print
if logger:
logger_fn = logger
for i in range(len(indices)):
o = ''
for obs in data[i][0]['goldstandard']:
o += '(' + ','.join(
[dataset.map.landmark_dict.decode(o_ind)
for o_ind in obs]) + ') ,'
# a = ', '.join([i2act[a_ind] for a_ind in actions[i]])
a = ','.join([
dataset.act_dict.decode(a_ind)
for a_ind in data[i][0]['actions']
])
logger_fn('Observations: ' + o)
logger_fn('Actions: ' + a)
logger_fn('GT: ' +
dataset.dict.decode(batch[0]['utterance'][i, 1:]))
logger_fn('Sample: ' +
dataset.dict.decode(generated_utterance[i, :]))
logger_fn('-' * 80)
parser.add_argument('--report-every', type=int, default=5)
parser.add_argument('--num-epochs', type=int, default=500, help='Number of epochs')
args = parser.parse_args()
exp_dir = os.path.join(args.exp_dir, args.exp_name)
if not os.path.exists(exp_dir):
os.mkdir(exp_dir)
args = parser.parse_args()
logger = create_logger(os.path.join(exp_dir, 'log.txt'))
logger.info(args)
train_data = TalkTheWalkEmergent(args.data_dir, 'train', goldstandard_features=True, T=args.T)
train_loader = DataLoader(train_data, args.batch_sz, collate_fn=get_collate_fn(args.cuda), shuffle=True)
valid_data = TalkTheWalkEmergent(args.data_dir, 'valid', goldstandard_features=True, T=args.T)
valid_loader = DataLoader(valid_data, args.batch_sz, collate_fn=get_collate_fn(args.cuda))
test_data = TalkTheWalkEmergent(args.data_dir, 'test', goldstandard_features=True, T=args.T)
test_loader = DataLoader(test_data, args.batch_sz, collate_fn=get_collate_fn(args.cuda))
guide = GuideContinuous(args.vocab_sz, len(train_data.map.landmark_dict),
apply_masc=args.apply_masc, T=args.T)
tourist = TouristContinuous(args.vocab_sz, len(train_data.map.landmark_dict), len(train_data.act_dict),
apply_masc=args.apply_masc, T=args.T)
params = list(tourist.parameters()) + list(guide.parameters())
opt = optim.Adam(params)
logger = create_logger(os.path.join(exp_dir, 'log.txt'))
logger.info(args)
data = TalkTheWalkLandmarks(args.data_dir, args.resnet_features,
args.fasttext_features,
args.textrecog_features)
train_data, valid_data = create_split(data)
add_weights(train_data, valid_data)
train_data = DatasetHolder(train_data)
valid_data = DatasetHolder(valid_data)
train_loader = DataLoader(train_data,
args.batch_sz,
collate_fn=get_collate_fn(args.cuda),
shuffle=True)
valid_loader = DataLoader(valid_data,
args.batch_sz,
collate_fn=get_collate_fn(args.cuda))
target = numpy.array(
[valid_data[i]['target'] for i in range(len(valid_data))])
ones = numpy.ones_like(target)
rand = numpy.random.randint(2, size=target.shape)
logger.info('Baselines' + '-' * 70)
logger.info("All positive: {}, {}, {}".format(
f1_score(target, ones, average='weighted'),
precision_score(target, ones, average='weighted'),
recall_score(target, ones, average='weighted')))
data_dir = args.data_dir
if args.trajectories == 'all':
dictionary = Dictionary(file=os.path.join(data_dir, 'dict.txt'), min_freq=3)
train_data = TalkTheWalkEmergent(data_dir, 'train', T=args.T)
train_data.dict = dictionary
valid_data = TalkTheWalkEmergent(data_dir, 'valid', T=args.T)
valid_data.dict = dictionary
test_data = TalkTheWalkEmergent(data_dir, 'test', T=args.T)
test_data.dict = dictionary
elif args.trajectories == 'human':
train_data = TalkTheWalkLanguage(data_dir, 'train')
valid_data = TalkTheWalkLanguage(data_dir, 'valid')
test_data = TalkTheWalkLanguage(data_dir, 'test')
train_loader = DataLoader(train_data, args.batch_sz, collate_fn=get_collate_fn(args.cuda))
valid_loader = DataLoader(valid_data, args.batch_sz, collate_fn=get_collate_fn(args.cuda))
test_loader = DataLoader(test_data, args.batch_sz, collate_fn=get_collate_fn(args.cuda))
tourist = TouristLanguage.load(args.tourist_model)
if args.guide_model is not None:
guide = GuideLanguage.load(args.guide_model)
else:
guide = GuideLanguage(128, 256, len(train_data.dict), apply_masc=True, T=3)
if args.cuda:
tourist = tourist.cuda()
guide = guide.cuda()
if args.train_guide:
default=50,
help='Number of epochs')
args = parser.parse_args()
exp_dir = os.path.join(args.exp_dir, args.exp_name)
if not os.path.exists(exp_dir):
os.mkdir(exp_dir)
logger = create_logger(os.path.join(exp_dir, 'log.txt'))
logger.info(args)
train_data = TalkTheWalkLanguage(args.data_dir, 'train', args.last_turns)
train_loader = DataLoader(train_data,
args.batch_sz,
collate_fn=get_collate_fn(args.cuda),
shuffle=True)
valid_data = TalkTheWalkLanguage(args.data_dir, 'valid', args.last_turns)
valid_loader = DataLoader(valid_data,
args.batch_sz,
collate_fn=get_collate_fn(args.cuda))
test_data = TalkTheWalkLanguage(args.data_dir, 'test', args.last_turns)
test_loader = DataLoader(test_data,
args.batch_sz,
collate_fn=get_collate_fn(args.cuda))
guide = GuideLanguage(args.embed_sz,
args.hidden_sz,
len(train_data.dict),
