def _prepare_config(self, str_optimizer: str, train_mode: TrainMode):
self.general_config = GeneralConfig(train_mode,
0.0001,
self.summary_interval,
self.ckp_interval,
config_name=str_optimizer,
model_name=self.dataset_name)
# Creates configuration for 5 mega-batches
if train_mode == TrainMode.INCREMENTAL:
for i in range(5):
train_conf = MegabatchConfig(60, batch_size=128)
self.general_config.add_train_conf(train_conf)
elif train_mode == TrainMode.ACUMULATIVE:
train_confs = [
MegabatchConfig(60, batch_size=128),
MegabatchConfig(50, batch_size=128),
MegabatchConfig(30, batch_size=128),
MegabatchConfig(30, batch_size=128),
MegabatchConfig(30, batch_size=128)
]
self.general_config.train_configurations = train_confs
else:
raise OptionNotSupportedError(
"The requested Experiment class: {} doesn't support the requested training"
" mode: {}".format(self.__class__, train_mode))
def load_training_data(self):
if self.train_mode == TrainMode.INCREMENTAL:
return self.curr_path, None
elif self.train_mode == TrainMode.ACUMULATIVE:
return self.accumulative_path, None
else:
raise OptionNotSupportedError(
"The requested Reader class: {} doesn't support the requested training"
" mode: {}".format(self.__class__, self.train_mode))
def _prepare_config(self, str_optimizer: str, train_mode: TrainMode):
self.general_config = CRILConfig(train_mode, 0.0001, self.summary_interval, self.ckp_interval,
config_name=str_optimizer, model_name=self.dataset_name,
n_candidates=40, memory_size=50, buffer_size=1)
# Creates configuration for 5 mega-batches
if train_mode == TrainMode.INCREMENTAL or train_mode == TrainMode.ACUMULATIVE:
for i in range(5):
train_conf = MegabatchConfig(100, batch_size=256)
self.general_config.add_train_conf(train_conf)
else:
raise OptionNotSupportedError("The requested Experiment class: {} doesn't support the requested training"
" mode: {}".format(self.__class__, train_mode))
def reload_training_data(self):
tr_filenames, tr_labels = self._find_image_files(
self.curr_path, self.categories)
if self.train_mode == TrainMode.INCREMENTAL:
self.train_filenames, self.train_labels = tr_filenames, tr_labels
elif self.train_mode == TrainMode.ACUMULATIVE:
self.train_filenames.extend(tr_filenames)
self.train_labels.extend(tr_labels)
else:
raise OptionNotSupportedError(
"The requested Reader class: {} doesn't support the requested training"
" mode: {}".format(self.__class__, self.train_mode))
def __get_not_supported_dataset(*args):
raise OptionNotSupportedError(
"The requested dataset doesn't have default paths in the current version of the "
"program.")
