def _initialize(self):
"""Initialize non-common things."""
self.save_distillation_dir = None
if not self.hyper_params.is_student:
# Since raining teacher do not require DistillationBuffer,
# it overloads DQNAgent._initialize.
print("[INFO] Teacher mode.")
DQNAgent._initialize(self)
self.make_distillation_dir()
else:
# Training student or generating distillation data(test).
print("[INFO] Student mode.")
self.softmax_tau = 0.01
build_args = dict(
hyper_params=self.hyper_params,
log_cfg=self.log_cfg,
env_name=self.env_info.name,
state_size=self.env_info.observation_space.shape,
output_size=self.env_info.action_space.n,
is_test=self.is_test,
load_from=self.load_from,
)
self.learner = build_learner(self.learner_cfg, build_args)
self.dataset_path = self.hyper_params.dataset_path
self.memory = DistillationBuffer(self.hyper_params.batch_size,
self.dataset_path)
if self.is_test:
self.make_distillation_dir()
def _initialize(self):
"""Initialize non-common things."""
self.save_distillation_dir = None
if not self.args.student:
# Since raining teacher do not require DistillationBuffer,
# it overloads DQNAgent._initialize.
DQNAgent._initialize(self)
self.make_distillation_dir()
else:
# Training student or generating distillation data(test).
self.softmax_tau = 0.01
self.learner = build_learner(self.learner_cfg)
self.dataset_path = self.hyper_params.dataset_path
self.memory = DistillationBuffer(self.hyper_params.batch_size,
self.dataset_path)
if self.args.test:
self.make_distillation_dir()