def create_trainer(args, converter, tag_index):
"""
Creates or loads a neural network according to the specified args.
:param tag_index: dict of tags.
"""
logger = logging.getLogger("Logger")
if args.load:
logger.info("Loading provided network...")
trainer = TaggerTrainer.load(args.load)
# change learning rate
trainer.learning_rate = args.learning_rate
trainer.threads = args.threads
else:
logger.info('Creating new network...')
# sum the number of features in all tables
input_size = converter.size() * (args.window * 2 + 1)
nn = SequenceNetwork(input_size, args.hidden, len(tag_index))
options = {
'learning_rate': args.learning_rate,
'eps': args.eps,
'ro': args.ro,
'verbose': args.verbose,
'left_context': args.window,
'right_context': args.window
}
trainer = TaggerTrainer(nn, converter, tag_index, options)
trainer.saver = saver(args.model, args.vectors, args.variant)
logger.info("... with the following parameters:")
logger.info(trainer.nn.description())
return trainer
def create_trainer(args, converter, tag_index):
"""
Creates or loads a neural network according to the specified args.
:param tag_index: dict of tags.
"""
logger = logging.getLogger("Logger")
if args.load:
logger.info("Loading provided network...")
trainer = TaggerTrainer.load(args.load)
# change learning rate
trainer.learning_rate = args.learning_rate
trainer.threads = args.threads
else:
logger.info('Creating new network...')
# sum the number of features in all tables
input_size = converter.size() * (args.window * 2 + 1)
nn = SequenceNetwork(input_size, args.hidden, len(tag_index))
options = {
'learning_rate': args.learning_rate,
'eps': args.eps,
'ro': args.ro,
'verbose': args.verbose,
'left_context': args.window,
'right_context': args.window
}
trainer = TaggerTrainer(nn, converter, tag_index, options)
trainer.saver = saver(args.model, args.vectors, args.variant)
logger.info("... with the following parameters:")
logger.info(trainer.nn.description())
return trainer
def create_trainer(args, converter, tags_dict):
"""
Creates or loads a neural network according to the specified args.
"""
logger = logging.getLogger("Logger")
if args.load:
logger.info("Loading provided network...")
trainer = TaggerTrainer.load(args.load)
trainer.learning_rate = args.learning_rate
trainer.threads = args.threads
else:
logger.info('Creating new network...')
trainer = TaggerTrainer(converter, args.learning_rate,
args.window/2, args.window/2,
args.hidden, tags_dict, args.verbose)
trainer.saver = saver(args.model, args.output)
logger.info("... with the following parameters:")
logger.info(trainer.nn.description())
return trainer
def create_trainer(args, converter, tags_dict):
"""
Creates or loads a neural network according to the specified args.
"""
logger = logging.getLogger("Logger")
if args.load:
logger.info("Loading provided network...")
trainer = TaggerTrainer.load(args.load)
trainer.learning_rate = args.learning_rate
trainer.threads = args.threads
else:
logger.info('Creating new network...')
trainer = TaggerTrainer(converter, args.learning_rate, args.window / 2,
args.window / 2, args.hidden, tags_dict,
args.verbose)
trainer.saver = saver(args.model, args.vectors, args.variant)
logger.info("... with the following parameters:")
logger.info(trainer.nn.description())
return trainer