torch.manual_seed(123)
batch_size = 32
hidden_size = 1024
dropout = 0.0
feature_fname = 'mfcc_delta_features.pt'
logging.basicConfig(level=logging.INFO)
factors = [3, 9, 27, 81, 243]
lz = len(str(abs(factors[-1])))
for ds_factor in factors:
logging.info('Loading data')
data = dict(train=D.flickr8k_loader(split='train',
batch_size=batch_size,
shuffle=True,
feature_fname=feature_fname,
downsampling_factor=ds_factor),
val=D.flickr8k_loader(split='val',
batch_size=batch_size,
shuffle=False,
feature_fname=feature_fname))
fd = D.Flickr8KData
fd.init_vocabulary(data['train'].dataset)
# Saving config
pickle.dump(
dict(feature_fname=feature_fname,
label_encoder=fd.get_label_encoder(),
language='en'), open('config.pkl', 'wb'))
import platalea.asr as M
import platalea.dataset as D
torch.manual_seed(123)
batch_size = 8
hidden_size = 1024
dropout = 0.0
feature_fname = 'mfcc_delta_features.pt'
logging.basicConfig(level=logging.INFO)
logging.info('Loading data')
data = dict(train=D.flickr8k_loader(split='train',
batch_size=batch_size,
shuffle=True,
feature_fname=feature_fname),
val=D.flickr8k_loader(split='val',
batch_size=batch_size,
shuffle=False,
feature_fname=feature_fname))
fd = D.Flickr8KData
fd.init_vocabulary(data['train'].dataset)
# Saving config
pickle.dump(
dict(feature_fname=feature_fname,
label_encoder=fd.get_label_encoder(),
language='en'), open('config.pkl', 'wb'))
config = dict(
torch.manual_seed(args.seed)
random.seed(args.seed)
logging.basicConfig(level=logging.INFO)
# Logging the arguments
logging.info('Arguments: {}'.format(args))
batch_size = 8
hidden_size = 1024
dropout = 0.0
logging.info('Loading data')
data = dict(
train=D.flickr8k_loader(
args.flickr8k_root, args.flickr8k_meta, args.flickr8k_language,
args.audio_features_fn, split='train', batch_size=batch_size,
shuffle=True, downsampling_factor=args.downsampling_factor),
val=D.flickr8k_loader(
args.flickr8k_root, args.flickr8k_meta, args.flickr8k_language,
args.audio_features_fn, split='val', batch_size=batch_size,
shuffle=False))
if args.downsampling_factor_text:
ds_factor_text = args.downsampling_factor_text
step_st = args.downsampling_factor_text
# The downsampling factor for text is applied on top of the main
# downsampling factor that is applied to all data
if args.downsampling_factor:
ds_factor_text *= args.downsampling_factor
data_st = dict(
train=D.flickr8k_loader(
args.enable_help()
args.parse()
# Setting general configuration
torch.manual_seed(args.seed)
random.seed(args.seed)
batch_size = 32
hidden_size = 1024
dropout = 0.0
logging.info('Loading data')
data = dict(train=D.flickr8k_loader(args.flickr8k_root,
args.flickr8k_meta,
args.flickr8k_language,
args.audio_features_fn,
split='train',
batch_size=batch_size,
shuffle=True),
val=D.flickr8k_loader(args.flickr8k_root,
args.flickr8k_meta,
args.flickr8k_language,
args.audio_features_fn,
split='val',
batch_size=batch_size,
shuffle=False))
logging.info('Building model')
net = M.TextImage(M.get_default_config())
run_config = dict(max_lr=2 * 1e-4, epochs=args.epochs)
import platalea.asr as M
import platalea.dataset as D
torch.manual_seed(123)
batch_size = 8
hidden_size = 1024
dropout = 0.0
feature_fname = 'mfcc_delta_features.pt'
logging.basicConfig(level=logging.INFO)
logging.info('Loading data')
data = dict(
train=D.flickr8k_loader(split='train', batch_size=batch_size, shuffle=True,
feature_fname=feature_fname, language='jp'),
val=D.flickr8k_loader(split='val', batch_size=batch_size, shuffle=False,
feature_fname=feature_fname, language='jp'))
fd = D.Flickr8KData
fd.init_vocabulary(data['train'].dataset)
# Saving config
pickle.dump(dict(feature_fname=feature_fname,
label_encoder=fd.get_label_encoder(),
language='jp'),
open('config.pkl', 'wb'))
config = dict(
SpeechEncoder=dict(
conv=dict(in_channels=39, out_channels=64, kernel_size=6, stride=2,
padding=0, bias=False),
parser = configargparse.get_argument_parser('platalea')
parser.add_argument(
'--epochs',
action='store',
default=32,
dest='epochs',
type=int,
help='number of epochs after which to stop training (default: 32)')
config_args, unknown_args = parser.parse_known_args()
logging.basicConfig(level=logging.INFO)
logging.info('Loading data')
data = dict(train=D.flickr8k_loader(split='train', batch_size=32,
shuffle=True),
val=D.flickr8k_loader(split='val', batch_size=32, shuffle=False))
D.Flickr8KData.init_vocabulary(data['train'].dataset)
config = dict(SpeechEncoder=dict(conv=dict(in_channels=39,
out_channels=64,
kernel_size=6,
stride=2,
padding=0,
bias=False),
rnn=dict(input_size=64,
hidden_size=1024,
num_layers=4,
bidirectional=True,
dropout=0),
att=dict(in_size=2048, hidden_size=128)),
logging.basicConfig(level=logging.INFO)
# Parse command line parameters
parser = argparse.ArgumentParser()
parser.add_argument('path', metavar='path', help='Model\'s path')
parser.add_argument('-b', help='Use beam decoding', dest='use_beam_decoding',
action='store_true', default=False)
args = parser.parse_args()
# Loading config
conf = pickle.load(open('config.pkl', 'rb'))
logging.info('Loading data')
data = dict(
train=D.flickr8k_loader(split='train', batch_size=batch_size,
shuffle=False, feature_fname=conf['feature_fname'],
language=conf['language']),
val=D.flickr8k_loader(split='val', batch_size=batch_size, shuffle=False,
feature_fname=conf['feature_fname'],
language=conf['language']))
fd = D.Flickr8KData
fd.le = conf['label_encoder']
net = torch.load(args.path)
trn = {}
logging.info('Extracting transcriptions')
for set_id, set_name in [('train', 'Training'), ('val', 'Validation')]:
import torch
import platalea.basic as M
import platalea.dataset as D
from utils.copy_best import copy_best
torch.manual_seed(123)
logging.basicConfig(level=logging.INFO)
factors = [3, 9, 27, 81, 243]
lz = len(str(abs(factors[-1])))
for ds_factor in factors:
logging.info('Loading data')
data = dict(train=D.flickr8k_loader(split='train',
batch_size=32,
shuffle=True,
downsampling_factor=ds_factor),
val=D.flickr8k_loader(split='val',
batch_size=32,
shuffle=False))
D.Flickr8KData.init_vocabulary(data['train'].dataset)
config = dict(SpeechEncoder=dict(conv=dict(in_channels=39,
out_channels=64,
kernel_size=6,
stride=2,
padding=0,
bias=False),
rnn=dict(input_size=64,
hidden_size=1024,
num_layers=4,
def train(args):
# Setting general configuration
torch.manual_seed(args.seed)
random.seed(args.seed)
# Logging the arguments
logging.info('Arguments: {}'.format(args))
logging.info('Loading data')
if args.dataset_name == 'flickr8k':
data = dict(train=D.flickr8k_loader(
args.flickr8k_root,
args.flickr8k_meta,
args.flickr8k_language,
args.audio_features_fn,
split='train',
batch_size=32,
shuffle=True,
downsampling_factor=args.downsampling_factor),
val=D.flickr8k_loader(args.flickr8k_root,
args.flickr8k_meta,
args.flickr8k_language,
args.audio_features_fn,
split='val',
batch_size=32,
shuffle=False))
elif args.dataset_name == "spokencoco":
data = dict(train=D.spokencoco_loader(
args.spokencoco_root,
args.spokencoco_meta,
args.audio_features_fn,
split='train',
batch_size=32,
shuffle=True,
downsampling_factor=args.downsampling_factor,
debug=args.debug),
val=D.spokencoco_loader(args.spokencoco_root,
args.spokencoco_meta,
args.audio_features_fn,
split='val',
batch_size=32,
shuffle=False,
debug=args.debug))
else:
raise ValueError(
"dataset_name should be in ['flickr8k', 'spokencoco']")
config = dict(
SpeechEncoder=dict(conv=dict(in_channels=39,
out_channels=64,
kernel_size=6,
stride=2,
padding=0,
bias=False),
rnn=dict(input_size=64,
hidden_size=args.hidden_size_factor,
num_layers=4,
bidirectional=True,
dropout=0),
att=dict(in_size=2 * args.hidden_size_factor,
hidden_size=128)),
ImageEncoder=dict(linear=dict(in_size=2048,
out_size=2 * args.hidden_size_factor),
norm=True),
margin_size=0.2)
logging.info('Building model')
net = M.SpeechImage(config)
run_config = dict(
max_lr=args.cyclic_lr_max,
min_lr=args.cyclic_lr_min,
epochs=args.epochs,
l2_regularization=args.l2_regularization,
)
logging.info('Training')
old_time = datetime.datetime.now()
logging.info(f'Start of training - {old_time}')
M.experiment(net, data, run_config, wandb_mode='disabled')
new_time = datetime.datetime.now()
logging.info(f'End of training - {new_time}')
diff_time = new_time - old_time
logging.info(f'Total duration: {diff_time}')
# Parse command line parameters
parser = argparse.ArgumentParser()
parser.add_argument('path', metavar='path', help='Model\'s path')
parser.add_argument('-b',
help='Use beam decoding',
dest='use_beam_decoding',
action='store_true',
default=False)
args = parser.parse_args()
# Loading config
conf = pickle.load(open('config.pkl', 'rb'))
logging.info('Loading data')
data = dict(val=D.flickr8k_loader(split='val',
batch_size=batch_size,
shuffle=False,
feature_fname=conf['feature_fname']))
fd = D.Flickr8KData
fd.le = conf['label_encoder']
logging.info('Loading model')
net = torch.load(args.path)
logging.info('Evaluating')
with torch.no_grad():
net.eval()
if args.use_beam_decoding:
result = platalea.score.score_asr(net,
data['val'].dataset,
beam_size=10)
else:
result = platalea.score.score_asr(net, data['val'].dataset)
dest='use_test_set',
action='store_true',
default=False)
args.enable_help()
args.parse()
batch_size = 16
logging.basicConfig(level=logging.INFO)
logging.info('Loading data')
if args.use_test_set:
data = D.flickr8k_loader(args.flickr8k_root,
args.flickr8k_meta,
args.flickr8k_language,
args.audio_features_fn,
split='test',
batch_size=batch_size,
shuffle=False)
else:
data = D.flickr8k_loader(args.flickr8k_root,
args.flickr8k_meta,
args.flickr8k_language,
args.audio_features_fn,
split='val',
batch_size=batch_size,
shuffle=False)
logging.info('Loading model')
net = torch.load(args.path)
logging.info('Evaluating')
