def _output_records(self, step_name, records):
"""Dump records."""
columns = ["worker_id", "performance", "desc"]
outputs = []
for record in records:
record = record.serialize()
_record = {}
for key in columns:
_record[key] = record[key]
outputs.append(deepcopy(_record))
data = pd.DataFrame(outputs)
step_path = FileOps.join_path(TaskOps().local_output_path, step_name)
FileOps.make_dir(step_path)
_file = FileOps.join_path(step_path, "output.csv")
try:
data.to_csv(_file, index=False)
except Exception:
logging.error("Failed to save output file, file={}".format(_file))
for record in outputs:
worker_id = record["worker_id"]
worker_path = TaskOps().get_local_worker_path(step_name, worker_id)
outputs_globs = []
outputs_globs += glob.glob(FileOps.join_path(worker_path, "desc_*.json"))
outputs_globs += glob.glob(FileOps.join_path(worker_path, "hps_*.json"))
outputs_globs += glob.glob(FileOps.join_path(worker_path, "model_*"))
outputs_globs += glob.glob(FileOps.join_path(worker_path, "performance_*.json"))
for _file in outputs_globs:
if os.path.isfile(_file):
FileOps.copy_file(_file, step_path)
elif os.path.isdir(_file):
FileOps.copy_folder(_file, FileOps.join_path(step_path, os.path.basename(_file)))
def _get_model_desc(self):
model_desc = self.trainer.model_desc
if not model_desc:
if ModelConfig.model_desc_file is not None:
desc_file = ModelConfig.model_desc_file
desc_file = desc_file.replace("{local_base_path}",
self.trainer.local_base_path)
if ":" not in desc_file:
desc_file = os.path.abspath(desc_file)
if ":" in desc_file:
local_desc_file = FileOps.join_path(
self.trainer.local_output_path,
os.path.basename(desc_file))
FileOps.copy_file(desc_file, local_desc_file)
desc_file = local_desc_file
model_desc = Config(desc_file)
logger.info("net_desc:{}".format(model_desc))
elif ModelConfig.model_desc is not None:
model_desc = ModelConfig.model_desc
elif ModelConfig.models_folder is not None:
folder = ModelConfig.models_folder.replace(
"{local_base_path}", self.trainer.local_base_path)
pattern = FileOps.join_path(folder, "desc_*.json")
desc_file = glob.glob(pattern)[0]
model_desc = Config(desc_file)
return model_desc
def _new_model_init(self):
"""Init new model.
:return: initial model after loading pretrained model
:rtype: torch.nn.Module
"""
init_model_file = self.config.init_model_file
if ":" in init_model_file:
local_path = FileOps.join_path(
self.trainer.get_local_worker_path(),
os.path.basename(init_model_file))
FileOps.copy_file(init_model_file, local_path)
self.config.init_model_file = local_path
network_desc = copy.deepcopy(self.base_net_desc)
network_desc.backbone.cfgs = network_desc.backbone.base_cfgs
model_init = NetworkDesc(network_desc).to_model()
return model_init
def _save_descript(self):
"""Save result descript."""
template_file = self.config.darts_template_file
genotypes = self.search_alg.codec.calc_genotype(
self._get_arch_weights())
if template_file == "{default_darts_cifar10_template}":
template = DartsNetworkTemplateConfig.cifar10
elif template_file == "{default_darts_cifar100_template}":
template = DartsNetworkTemplateConfig.cifar100
elif template_file == "{default_darts_imagenet_template}":
template = DartsNetworkTemplateConfig.imagenet
else:
dst = FileOps.join_path(self.trainer.get_local_worker_path(),
os.path.basename(template_file))
FileOps.copy_file(template_file, dst)
template = Config(dst)
model_desc = self._gen_model_desc(genotypes, template)
self.trainer.config.codec = model_desc
def _get_model_desc(self):
model_desc = self.model_desc
self.saved_folder = self.get_local_worker_path(self.step_name,
self.worker_id)
if not model_desc:
if os.path.exists(
FileOps.join_path(self.saved_folder,
'desc_{}.json'.format(self.worker_id))):
model_config = Config(
FileOps.join_path(self.saved_folder,
'desc_{}.json'.format(self.worker_id)))
if "type" not in model_config and "modules" not in model_config:
model_config = ModelConfig.model_desc
model_desc = model_config
elif ModelConfig.model_desc_file is not None:
desc_file = ModelConfig.model_desc_file
desc_file = desc_file.replace("{local_base_path}",
self.local_base_path)
if ":" not in desc_file:
desc_file = os.path.abspath(desc_file)
if ":" in desc_file:
local_desc_file = FileOps.join_path(
self.local_output_path, os.path.basename(desc_file))
FileOps.copy_file(desc_file, local_desc_file)
desc_file = local_desc_file
model_desc = Config(desc_file)
logger.info("net_desc:{}".format(model_desc))
elif ModelConfig.model_desc is not None:
model_desc = ModelConfig.model_desc
elif ModelConfig.models_folder is not None:
folder = ModelConfig.models_folder.replace(
"{local_base_path}", self.local_base_path)
pattern = FileOps.join_path(folder, "desc_*.json")
desc_file = glob.glob(pattern)[0]
model_desc = Config(desc_file)
elif PipeStepConfig.pipe_step.get("models_folder") is not None:
folder = PipeStepConfig.pipe_step.get("models_folder").replace(
"{local_base_path}", self.local_base_path)
desc_file = FileOps.join_path(
folder, "desc_{}.json".format(self.worker_id))
model_desc = Config(desc_file)
logger.info("Load model from model folder {}.".format(folder))
return model_desc
def _copy_needed_file(self):
if self.config.pareto_front_file is None:
raise FileNotFoundError(
"Config item paretor_front_file not found in config file.")
init_pareto_front_file = self.config.pareto_front_file.replace(
"{local_base_path}", self.local_base_path)
self.pareto_front_file = FileOps.join_path(self.local_output_path,
self.step_name,
"pareto_front.csv")
FileOps.make_base_dir(self.pareto_front_file)
FileOps.copy_file(init_pareto_front_file, self.pareto_front_file)
if self.config.random_file is None:
raise FileNotFoundError(
"Config item random_file not found in config file.")
init_random_file = self.config.random_file.replace(
"{local_base_path}", self.local_base_path)
self.random_file = FileOps.join_path(self.local_output_path,
self.step_name, "random.csv")
FileOps.copy_file(init_random_file, self.random_file)
def _get_pretrained_model_file(self):
if not self.trainer.load_weights_file:
return None
model_file = self.trainer.config.kwargs.get("pretrained_model_file")
if model_file:
return model_file
if ModelConfig.pretrained_model_file:
model_file = ModelConfig.pretrained_model_file
model_file = model_file.replace("{local_base_path}",
self.trainer.local_base_path)
model_file = model_file.replace("{worker_id}",
str(self.trainer.worker_id))
if ":" not in model_file:
model_file = os.path.abspath(model_file)
if ":" in model_file:
local_model_file = FileOps.join_path(
self.trainer.local_output_path,
os.path.basename(model_file))
FileOps.copy_file(model_file, local_model_file)
model_file = local_model_file
return model_file
else:
return None
def _save_best_model(self):
"""Save best model."""
if vega.is_torch_backend():
torch.save(self.trainer.model.state_dict(),
self.trainer.weights_file)
elif vega.is_tf_backend():
worker_path = self.trainer.get_local_worker_path()
model_id = "model_{}".format(self.trainer.worker_id)
weights_folder = FileOps.join_path(worker_path, model_id)
FileOps.make_dir(weights_folder)
checkpoint_file = tf.train.latest_checkpoint(worker_path)
ckpt_globs = glob.glob("{}.*".format(checkpoint_file))
for _file in ckpt_globs:
FileOps.copy_file(
_file,
FileOps.join_path(weights_folder,
os.path.split(_file)[-1]))
FileOps.copy_file(FileOps.join_path(worker_path, 'checkpoint'),
weights_folder)
if self.trainer.save_ext_model:
self._save_pb_model(weights_folder, model_id)
self.trainer.ext_model = FileOps.join_path(
weights_folder, '{}.pb'.format(model_id))
elif vega.is_ms_backend():
worker_path = self.trainer.get_local_worker_path()
save_path = os.path.join(
worker_path, "model_{}.ckpt".format(self.trainer.worker_id))
for file in os.listdir(worker_path):
if file.startswith("CKP") and file.endswith(".ckpt"):
self.weights_file = FileOps.join_path(worker_path, file)
os.rename(self.weights_file, save_path)
if self.trainer.save_ext_model:
model_id = "model_{}".format(self.trainer.worker_id)
self._save_om_model(worker_path, model_id)
self.trainer.ext_model = FileOps.join_path(
worker_path, '{}.om'.format(model_id))
