def __init__(self,
model_dir,
initial_iter=0,
is_training_action=True,
**_unused):
self.initial_iter = initial_iter
self.file_name_prefix = make_model_name(model_dir)
# randomly initialise or restoring model
if is_training_action and initial_iter == 0:
SESS_STARTED.connect(self.rand_init_model)
else:
SESS_STARTED.connect(self.restore_model)
def __init__(self,
model_dir,
save_every_n=0,
max_checkpoints=1,
is_training_action=True,
**_unused):
self.save_every_n = save_every_n
self.max_checkpoints = max_checkpoints
self.file_name_prefix = make_model_name(model_dir)
self.saver = None
# initialise the saver after the graph finalised
SESS_STARTED.connect(self.init_saver)
# save the training model at a positive frequency
if self.save_every_n > 0:
ITER_FINISHED.connect(self.save_model_interval)
# always save the final training model before exiting
if is_training_action:
SESS_FINISHED.connect(self.save_model)