def __init__(self,
obs_space,
action_space,
config,
existing_inputs=None):
if get_default_config:
config = dict(get_default_config(), **config)
if before_init:
before_init(self, obs_space, action_space, config)
def before_loss_init_wrapper(policy, obs_space, action_space,
config):
if before_loss_init:
before_loss_init(policy, obs_space, action_space, config)
if extra_action_fetches_fn is None:
self._extra_action_fetches = {}
else:
self._extra_action_fetches = extra_action_fetches_fn(self)
DynamicTFPolicy.__init__(self,
obs_space,
action_space,
config,
loss_fn,
stats_fn=stats_fn,
grad_stats_fn=grad_stats_fn,
before_loss_init=before_loss_init_wrapper,
existing_inputs=existing_inputs)
if after_init:
after_init(self, obs_space, action_space, config)
def __init__(
self,
obs_space,
action_space,
config,
existing_model=None,
existing_inputs=None,
):
if get_default_config:
config = dict(get_default_config(), **config)
if validate_spaces:
validate_spaces(self, obs_space, action_space, config)
if before_init:
before_init(self, obs_space, action_space, config)
def before_loss_init_wrapper(policy, obs_space, action_space,
config):
if before_loss_init:
before_loss_init(policy, obs_space, action_space, config)
if extra_action_out_fn is None or policy._is_tower:
extra_action_fetches = {}
else:
extra_action_fetches = extra_action_out_fn(policy)
if hasattr(policy, "_extra_action_fetches"):
policy._extra_action_fetches.update(extra_action_fetches)
else:
policy._extra_action_fetches = extra_action_fetches
DynamicTFPolicy.__init__(
self,
obs_space=obs_space,
action_space=action_space,
config=config,
loss_fn=loss_fn,
stats_fn=stats_fn,
grad_stats_fn=grad_stats_fn,
before_loss_init=before_loss_init_wrapper,
make_model=make_model,
action_sampler_fn=action_sampler_fn,
action_distribution_fn=action_distribution_fn,
existing_inputs=existing_inputs,
existing_model=existing_model,
get_batch_divisibility_req=get_batch_divisibility_req,
)
if after_init:
after_init(self, obs_space, action_space, config)
# Got to reset global_timestep again after this fake run-through.
self.global_timestep = 0
def __init__(self,
obs_space,
action_space,
config,
existing_model=None,
existing_inputs=None):
if get_default_config:
config = dict(get_default_config(), **config)
if validate_spaces:
validate_spaces(self, obs_space, action_space, config)
if before_init:
before_init(self, obs_space, action_space, config)
def before_loss_init_wrapper(policy, obs_space, action_space,
config):
if before_loss_init:
before_loss_init(policy, obs_space, action_space, config)
if extra_action_fetches_fn is None:
policy._extra_action_fetches = {}
else:
policy._extra_action_fetches = extra_action_fetches_fn(
policy)
policy._extra_action_fetches = extra_action_fetches_fn(
policy)
DynamicTFPolicy.__init__(
self,
obs_space=obs_space,
action_space=action_space,
config=config,
loss_fn=loss_fn,
stats_fn=stats_fn,
grad_stats_fn=grad_stats_fn,
before_loss_init=before_loss_init_wrapper,
make_model=make_model,
action_sampler_fn=action_sampler_fn,
action_distribution_fn=action_distribution_fn,
existing_inputs=existing_inputs,
existing_model=existing_model,
view_requirements_fn=view_requirements_fn,
get_batch_divisibility_req=get_batch_divisibility_req,
obs_include_prev_action_reward=obs_include_prev_action_reward)
if after_init:
after_init(self, obs_space, action_space, config)
# Got to reset global_timestep again after this fake run-through.
self.global_timestep = 0
def __init__(self,
obs_space,
action_space,
config,
existing_model=None,
existing_inputs=None):
if get_default_config:
config = dict(get_default_config(), **config)
if before_init:
before_init(self, obs_space, action_space, config)
def before_loss_init_wrapper(policy, obs_space, action_space,
config):
if before_loss_init:
before_loss_init(policy, obs_space, action_space, config)
if extra_action_fetches_fn is None:
self._extra_action_fetches = {}
else:
self._extra_action_fetches = extra_action_fetches_fn(self)
t = time.time()
# DEBUG
print("\ttf_policy_template.py: Initializing DynamicTFPolicy.")
DynamicTFPolicy.__init__(
self,
obs_space,
action_space,
config,
loss_fn,
stats_fn=stats_fn,
grad_stats_fn=grad_stats_fn,
before_loss_init=before_loss_init_wrapper,
make_model=make_model,
action_sampler_fn=action_sampler_fn,
existing_model=existing_model,
existing_inputs=existing_inputs,
get_batch_divisibility_req=get_batch_divisibility_req,
obs_include_prev_action_reward=obs_include_prev_action_reward)
# DEBUG
print(
"\ttf_policy_template.py: Done initializing DynamicTFPolicy.")
print("DTFP: %fs" % (time.time() - t))
if after_init:
after_init(self, obs_space, action_space, config)
def __init__(self,
obs_space,
action_space,
config,
existing_model=None,
existing_inputs=None):
if get_default_config:
config = dict(get_default_config(), **config)
if validate_spaces:
validate_spaces(self, obs_space, action_space, config)
if before_init:
before_init(self, obs_space, action_space, config)
def before_loss_init_wrapper(policy, obs_space, action_space,
config):
if before_loss_init:
before_loss_init(policy, obs_space, action_space, config)
if extra_action_fetches_fn is None:
self._extra_action_fetches = {}
else:
self._extra_action_fetches = extra_action_fetches_fn(self)
DynamicTFPolicy.__init__(
self,
obs_space=obs_space,
action_space=action_space,
config=config,
loss_fn=loss_fn,
stats_fn=stats_fn,
grad_stats_fn=grad_stats_fn,
before_loss_init=before_loss_init_wrapper,
make_model=make_model,
action_sampler_fn=action_sampler_fn,
action_distribution_fn=action_distribution_fn,
existing_model=existing_model,
existing_inputs=existing_inputs,
get_batch_divisibility_req=get_batch_divisibility_req,
obs_include_prev_action_reward=obs_include_prev_action_reward)
if after_init:
after_init(self, obs_space, action_space, config)
