def save(self, dir_path):
dir_path = Path(dir_path)
dir_path.mkdir(parents=True, exist_ok=True)
opts.save(dir_path, self.options)
fields.save_vocabs(dir_path, self.fields_tuples)
models.save(dir_path, self.model)
optimizer.save(dir_path, self.optimizer)
scheduler.save(dir_path, self.scheduler)
def save(self, current_epoch):
epoch_dir = 'epoch_{}'.format(current_epoch)
output_path = Path(self.output_dir, epoch_dir)
output_path.mkdir(exist_ok=True)
logging.info('Saving training state to {}'.format(output_path))
models.save(output_path, self.model)
optimizer.save(output_path, self.optimizer)
scheduler.save(output_path, self.scheduler)
def run(options):
words_field = fields.WordsField()
tags_field = fields.TagsField()
fields_tuples = [('words', words_field), ('tags', tags_field)]
fields_tuples += features.build(options)
logging.info('Building train corpus: {}'.format(options.train_path))
train_dataset = dataset.build(options.train_path, fields_tuples, options)
logging.info('Building train iterator...')
train_iter = iterator.build(train_dataset,
options.gpu_id,
options.train_batch_size,
is_train=True)
dev_dataset = None
dev_iter = None
if options.dev_path is not None:
logging.info('Building dev dataset: {}'.format(options.dev_path))
dev_dataset = dataset.build(options.dev_path, fields_tuples, options)
logging.info('Building dev iterator...')
dev_iter = iterator.build(dev_dataset,
options.gpu_id,
options.dev_batch_size,
is_train=False)
test_dataset = None
test_iter = None
if options.test_path is not None:
logging.info('Building test dataset: {}'.format(options.test_path))
test_dataset = dataset.build(options.test_path, fields_tuples, options)
logging.info('Building test iterator...')
test_iter = iterator.build(test_dataset,
options.gpu_id,
options.dev_batch_size,
is_train=False)
datasets = [train_dataset, dev_dataset, test_dataset]
datasets = list(filter(lambda x: x is not None, datasets))
if options.load:
logging.info('Loading vocabularies...')
fields.load_vocabs(options.load, fields_tuples)
logging.info('Word vocab size: {}'.format(len(words_field.vocab)))
logging.info('Tag vocab size: {}'.format(len(tags_field.vocab)))
logging.info('Loading model...')
model = models.load(options.load, fields_tuples)
logging.info('Loading optimizer...')
optim = optimizer.load(options.load, model.parameters())
logging.info('Loading scheduler...')
sched = scheduler.load(options.load, optim)
else:
logging.info('Building vocabulary...')
fields.build_vocabs(fields_tuples, train_dataset, datasets, options)
logging.info('Word vocab size: {}'.format(len(words_field.vocab)))
logging.info('Tag vocab size: {}'.format(len(tags_field.vocab)))
logging.info('Building model...')
model = models.build(options, fields_tuples)
logging.info('Building optimizer...')
optim = optimizer.build(options, model.parameters())
logging.info('Building scheduler...')
sched = scheduler.build(options, optim)
logging.info('Building trainer...')
trainer = Trainer(train_iter,
model,
optim,
sched,
options,
dev_iter=dev_iter,
test_iter=test_iter)
if options.resume_epoch and options.load is None:
logging.info('Resuming training...')
trainer.resume(options.resume_epoch)
trainer.train()
if options.save:
logging.info('Saving path: {}'.format(options.save))
config_path = Path(options.save)
config_path.mkdir(parents=True, exist_ok=True)
logging.info('Saving config options...')
opts.save(config_path, options)
logging.info('Saving vocabularies...')
fields.save_vocabs(config_path, fields_tuples)
logging.info('Saving model...')
models.save(config_path, model)
logging.info('Saving optimizer...')
optimizer.save(config_path, optim)
logging.info('Saving scheduler...')
scheduler.save(config_path, sched)
return fields_tuples, model, optim, sched
def run(options):
logging.info('Running with options: {}'.format(options))
words_field = fields.WordsField()
tags_field = fields.TagsField()
fields_tuples = [('words', words_field), ('tags', tags_field)]
fields_tuples += features.build(options)
logging.info('Building train corpus: {}'.format(options.train_path))
train_dataset = dataset.build(options.train_path, fields_tuples, options)
logging.info('Building train iterator...')
train_iter = iterator.build(train_dataset,
options.gpu_id,
options.train_batch_size,
is_train=True)
dev_dataset = None
dev_iter = None
if options.dev_path is not None:
logging.info('Building dev dataset: {}'.format(options.dev_path))
dev_dataset = dataset.build(options.dev_path, fields_tuples, options)
logging.info('Building dev iterator...')
dev_iter = iterator.build(dev_dataset,
options.gpu_id,
options.dev_batch_size,
is_train=False)
test_dataset = None
test_iter = None
if options.test_path is not None:
logging.info('Building test dataset: {}'.format(options.test_path))
test_dataset = dataset.build(options.test_path, fields_tuples, options)
logging.info('Building test iterator...')
test_iter = iterator.build(test_dataset,
options.gpu_id,
options.dev_batch_size,
is_train=False)
datasets = [train_dataset, dev_dataset, test_dataset]
datasets = list(filter(lambda x: x is not None, datasets))
# BUILD
if not options.load:
logging.info('Building vocabulary...')
fields.build_vocabs(fields_tuples, train_dataset, datasets, options)
loss_weights = None
if options.loss_weights == 'balanced':
loss_weights = train_dataset.get_loss_weights()
logging.info('Building model...')
model = models.build(options, fields_tuples, loss_weights)
logging.info('Building optimizer...')
optim = optimizer.build(options, model.parameters())
logging.info('Building scheduler...')
sched = scheduler.build(options, optim)
# OR LOAD
else:
logging.info('Loading vocabularies...')
fields.load_vocabs(options.load, fields_tuples)
logging.info('Loading model...')
model = models.load(options.load, fields_tuples)
logging.info('Loading optimizer...')
optim = optimizer.load(options.load, model.parameters())
logging.info('Loading scheduler...')
sched = scheduler.load(options.load, optim)
# STATS
logging.info('Word vocab size: {}'.format(len(words_field.vocab)))
logging.info('Tag vocab size: {}'.format(len(tags_field.vocab) - 1))
logging.info('Number of training examples: {}'.format(len(train_dataset)))
if dev_dataset:
logging.info('Number of dev examples: {}'.format(len(dev_dataset)))
if test_dataset:
logging.info('Number of test examples: {}'.format(len(test_dataset)))
logging.info('Model info: ')
logging.info(str(model))
logging.info('Optimizer info: ')
logging.info(str(optim))
logging.info('Scheduler info: ')
logging.info(str(sched))
# TRAIN
logging.info('Building trainer...')
trainer = Trainer(train_iter,
model,
optim,
sched,
options,
dev_iter=dev_iter,
test_iter=test_iter)
if options.resume_epoch and options.load is None:
logging.info('Resuming training...')
trainer.resume(options.resume_epoch)
trainer.train()
# SAVE
if options.save:
logging.info('Saving path: {}'.format(options.save))
config_path = Path(options.save)
config_path.mkdir(parents=True, exist_ok=True)
logging.info('Saving config options...')
opts.save(config_path, options)
logging.info('Saving vocabularies...')
fields.save_vocabs(config_path, fields_tuples)
logging.info('Saving model...')
models.save(config_path, model)
logging.info('Saving optimizer...')
optimizer.save(config_path, optim)
logging.info('Saving scheduler...')
scheduler.save(config_path, sched)
return fields_tuples, model, optim, sched