'nasbench101': NasBench101SearchSpace(),
'nasbench201': NasBench201SearchSpace(),
'darts': DartsSearchSpace()
}
#load_labeled = (True if config.search_space == 'darts' else False)
load_labeled = False
dataset_api = get_dataset_api(config.search_space, config.dataset)
utils.set_seed(config.seed)
search_space = supported_search_spaces[config.search_space]
optimizer = supported_optimizers[config.optimizer]
optimizer.adapt_search_space(search_space, dataset_api=dataset_api)
trainer = Trainer(optimizer, config, lightweight_output=True)
if config.optimizer == 'bananas':
trainer.search(resume_from="")
trainer.evaluate(resume_from="", dataset_api=dataset_api)
elif config.optimizer in ['oneshot', 'rsws']:
predictor = OneShotPredictor(config, trainer, model_path=config.model_path)
predictor_evaluator = PredictorEvaluator(predictor, config=config)
predictor_evaluator.adapt_search_space(search_space,
load_labeled=load_labeled,
dataset_api=dataset_api)
# evaluate the predictor
predictor_evaluator.evaluate()
utils.log_args(config)
supported_optimizers = {
'darts': DARTSOptimizer(config),
'gdas': GDASOptimizer(config),
'oneshot': OneShotNASOptimizer(config),
'rsws': RandomNASOptimizer(config),
're': RegularizedEvolution(config),
'rs': RandomSearch(config),
'ls': RandomSearch(config),
'bananas': Bananas(config),
'bp': BasePredictor(config)
}
search_space = NasBench201SearchSpace()
dataset_api = get_dataset_api('nasbench201', config.dataset)
optimizer = supported_optimizers[config.optimizer]
optimizer.adapt_search_space(search_space)
trainer = Trainer(optimizer, config, lightweight_output=True)
#trainer.search()
#if not config.eval_only:
# checkpoint = utils.get_last_checkpoint(config) if config.resume else ""
# trainer.search(resume_from=checkpoint)
#checkpoint = utils.get_last_checkpoint(config, search=False) if config.resume else ""
#trainer.evaluate(resume_from=checkpoint, dataset_api=dataset_api)
import logging from naslib.defaults.trainer import Trainer from naslib.optimizers import DARTSOptimizer, GDASOptimizer, RandomSearch from naslib.search_spaces import DartsSearchSpace, SimpleCellSearchSpace from naslib.utils import set_seed, setup_logger, get_config_from_args config = get_config_from_args() # use --help so see the options set_seed(config.seed) logger = setup_logger(config.save + "/log.log") logger.setLevel(logging.INFO) # default DEBUG is very verbose search_space = DartsSearchSpace() # use SimpleCellSearchSpace() for less heavy search optimizer = DARTSOptimizer(config) optimizer.adapt_search_space(search_space) trainer = Trainer(optimizer, config) trainer.search() # Search for an architecture trainer.evaluate() # Evaluate the best architecture
if config.optimizer == 'darts':
optimizer = DARTSOptimizer(config)
elif config.optimizer == 'gdas':
optimizer = GDASOptimizer(config)
elif config.optimizer == 'liu_et_al':
optimizer = DARTSOptimizer(
config) # hack to instanciate the trainer (is ignored during eval)
best_arch = LiuFinalArch()
elif config.optimizer == 'random':
optimizer = DARTSOptimizer(
config) # hack to instanciate the trainer (is ignored during eval)
best_arch = sample_random_architecture(search_space,
search_space.OPTIMIZER_SCOPE)
else:
raise ValueError("Unknown optimizer : {}".format(config.optimizer))
optimizer.adapt_search_space(search_space)
trainer = Trainer(optimizer, config)
if config.eval_only and not best_arch:
trainer.evaluate(resume_from=utils.get_last_checkpoint(
config, search=False) if config.resume else "")
elif best_arch:
best_arch.parse()
trainer.evaluate(best_arch=best_arch)
else:
trainer.search(
resume_from=utils.get_last_checkpoint(config) if config.resume else "")
trainer.evaluate(resume_from=utils.get_last_checkpoint(
config, search=False) if config.resume else "")
logger.setLevel(logging.INFO) # default DEBUG is too verbose
utils.log_args(config)
supported_optimizers = {
'darts': DARTSOptimizer(config),
'gdas': GDASOptimizer(config),
'oneshot': OneShotNASOptimizer(config),
'rsws': RandomNASOptimizer(config),
're': RegularizedEvolution(config),
'rs': RandomSearch(config),
'ls': RandomSearch(config),
'bananas': Bananas(config),
'bp': BasePredictor(config)
}
if config.dataset == 'cifar100': DartsSearchSpace.NUM_CLASSES = 100
search_space = DartsSearchSpace()
optimizer = supported_optimizers[config.optimizer]
optimizer.adapt_search_space(search_space)
trainer = Trainer(optimizer, config)
#trainer.search(resume_from=utils.get_last_checkpoint(config) if config.resume else "")
#if config.eval_only:
#trainer.evaluate(resume_from=utils.get_last_checkpoint(config, search=False) if config.resume else "")
#else:
#trainer.search(resume_from=utils.get_last_checkpoint(config) if config.resume else "")
#trainer.evaluate(resume_from=utils.get_last_checkpoint(config, search=False) if config.resume else "")
config = utils.get_config_from_args()
utils.set_seed(config.seed)
logger = setup_logger(config.save + "/log.log")
logger.setLevel(logging.INFO) # default DEBUG is too verbose
utils.log_args(config)
supported_optimizers = {
'darts': DARTSOptimizer(config),
'gdas': GDASOptimizer(config),
're': RegularizedEvolution(config),
'rs': RandomSearch(config),
}
search_space = NasBench201SearchSpace()
optimizer = supported_optimizers[config.optimizer]
optimizer.adapt_search_space(search_space)
trainer = Trainer(optimizer, config)
if not config.eval_only:
checkpoint = utils.get_last_checkpoint(config) if config.resume else ""
trainer.search(resume_from=checkpoint)
checkpoint = utils.get_last_checkpoint(config,
search=False) if config.resume else ""
trainer.evaluate(resume_from=checkpoint)