add_general_options(parser)
task_class = add_task_option(parser)
task_class.add_options(parser)
trainer_class = add_trainer_option(parser)
trainer_class.add_inference_options(parser)
add_log_options(parser)
parser.add_argument('-load_from',
type=str,
required=True,
help='Path to one or more pretrained models.')
parser.add_argument('-output', help="Path to output the predictions")
args = parser.parse_args()
logger = setup_logging_from_args(args, 'evaluate')
logger.debug('Torch version: {}'.format(torch.__version__))
logger.debug(args)
torch.manual_seed(args.seed)
np.random.seed(args.seed)
task = task_class.setup_task(args) # type: Task
logger.info('Loading checkpoint {}'.format(args.load_from))
checkpoint = load_checkpoint(args.load_from)
trainer = trainer_class(args, for_training=False, checkpoint=checkpoint)
results = trainer.solve(task)
parser = argparse.ArgumentParser(description="train.py")
add_general_options(parser)
task_class = add_task_option(parser)
task_class.add_options(parser)
trainer_class = add_trainer_option(parser)
trainer_class.add_eval_options(parser)
add_log_options(parser)
parser.add_argument('-load_from', type=str, nargs='+', required=True,
help='Path to one or more pretrained models.')
parser.add_argument('-output',
help="Path to output the predictions")
args = parser.parse_args()
logger = custom_logging.setup_logging_from_args(args, 'evaluate')
logger.debug('Torch version: {}'.format(torch.__version__))
logger.debug(args)
torch.manual_seed(args.seed)
task = task_class.setup_task(args) # type: Task
trainer = trainer_class(args) # type: Trainer
models = []
for filename in args.load_from:
logger.info('Loading checkpoint {}'.format(filename))
models.append(trainer.load_checkpoint(convert.load_checkpoint(filename)))
add_general_options(parser)
task_class = add_task_option(parser)
task_class.add_options(parser)
trainer_class = add_trainer_option(parser)
trainer_class.add_training_options(parser)
add_log_options(parser)
parser.add_argument('-load_from',
type=str,
help='If training from a checkpoint then this is the'
'path to the pretrained model.')
parser.add_argument('-reset_optim', action='store_true')
args = parser.parse_args()
logger = setup_logging_from_args(args, 'train')
logger.debug('Torch version: {}'.format(torch.__version__))
logger.debug(args)
torch.manual_seed(args.seed)
np.random.seed(args.seed)
task = task_class.setup_task(args) # type: Task
if args.load_from is not None:
logger.info("Loading checkpoint {}".format(args.load_from))
checkpoint = load_checkpoint(args.load_from)
else:
checkpoint = None
parser = argparse.ArgumentParser(description="train.py")
add_general_options(parser)
task_class = add_task_option(parser)
task_class.add_options(parser)
trainer_class = add_trainer_option(parser)
trainer_class.add_eval_options(parser)
add_log_options(parser)
parser.add_argument('-load_from',
type=str,
required=True,
help='Path to a pretrained model.')
parser.add_argument('-output', help="Path to output the predictions")
args = parser.parse_args()
logger = custom_logging.setup_logging_from_args(args, 'validate')
logger.debug('Torch version: {}'.format(torch.__version__))
logger.debug(args)
torch.manual_seed(args.seed)
task = task_class.setup_task(args) # type: Task
trainer = trainer_class(args) # type: Trainer
model = trainer.load_checkpoint(convert.load_checkpoint(args.load_from))
val_loss = trainer.evaluate(model, task)
import argparse
import torch
from nmtg import custom_logging
from nmtg.average_checkpoints import average_checkpoints
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('checkpoints',
nargs='+',
help='Which checkpoints to average')
parser.add_argument('-output',
type=str,
required=True,
help='Output filename')
parser.add_argument('-method',
choices=['mean', 'gmean'],
default='mean',
help='Method of averaging')
custom_logging.add_log_options(parser)
args = parser.parse_args()
logger = custom_logging.setup_logging_from_args(args,
'average_checkpoints.py')
checkpoint = average_checkpoints(args.checkpoints, args.method)
logger.info('Saving checkpoint to {}'.format(args.output))
torch.save(checkpoint, args.output)