def register_custom_model(model_name, model_class):
"""Register a custom model class by name.
The model can be later used by specifying {"custom_model": model_name}
in the model config.
Args:
model_name (str): Name to register the model under.
model_class (type): Python class of the model.
"""
_global_registry.register(RLLIB_MODEL, model_name, model_class)
def register_custom_preprocessor(preprocessor_name, preprocessor_class):
"""Register a custom preprocessor class by name.
The preprocessor can be later used by specifying
{"custom_preprocessor": preprocesor_name} in the model config.
Args:
preprocessor_name (str): Name to register the preprocessor under.
preprocessor_class (type): Python class of the preprocessor.
"""
_global_registry.register(RLLIB_PREPROCESSOR, preprocessor_name,
preprocessor_class)
def register_custom_action_dist(action_dist_name, action_dist_class):
"""Register a custom action distribution class by name.
The model can be later used by specifying
{"custom_action_dist": action_dist_name} in the model config.
Args:
model_name (str): Name to register the action distribution under.
model_class (type): Python class of the action distribution.
"""
_global_registry.register(RLLIB_ACTION_DIST, action_dist_name,
action_dist_class)
def register_custom_preprocessor(preprocessor_name, preprocessor_class):
"""Register a custom preprocessor class by name.
The preprocessor can be later used by specifying
{"custom_preprocessor": preprocesor_name} in the model config.
Args:
preprocessor_name (str): Name to register the preprocessor under.
preprocessor_class (type): Python class of the preprocessor.
"""
_global_registry.register(RLLIB_PREPROCESSOR, preprocessor_name,
preprocessor_class)
def register_custom_model(model_name: str, model_class: type) -> None:
"""Register a custom model class by name.
The model can be later used by specifying {"custom_model": model_name}
in the model config.
Args:
model_name (str): Name to register the model under.
model_class (type): Python class of the model.
"""
if issubclass(model_class, tf.keras.Model):
deprecation_warning(old="register_custom_model", error=False)
_global_registry.register(RLLIB_MODEL, model_name, model_class)
def testErrorHandling(self):
ray.init(num_cpus=4, num_gpus=2)
runner = TrialRunner()
kwargs = {
"stopping_criterion": {"training_iteration": 1},
"resources": Resources(cpu=1, gpu=1),
}
_global_registry.register(TRAINABLE_CLASS, "asdf", None)
trials = [Trial("asdf", **kwargs), Trial("__fake", **kwargs)]
for t in trials:
runner.add_trial(t)
runner.step()
self.assertEqual(trials[0].status, Trial.ERROR)
self.assertEqual(trials[1].status, Trial.PENDING)
runner.step()
self.assertEqual(trials[0].status, Trial.ERROR)
self.assertEqual(trials[1].status, Trial.RUNNING)
def testStartFailure(self):
_global_registry.register(TRAINABLE_CLASS, "asdf", None)
trial = Trial("asdf", resources=Resources(1, 0))
self.trial_executor.start_trial(trial)
self.assertEqual(Trial.ERROR, trial.status)
def action_adapter(model_action):
assert model_action in [0, 1, 2, 3]
return ACTIONS[model_action]
# throttle, brake, steering = model_action
# return np.array([throttle, brake, steering * np.pi * 0.25])
class TrainingModel(GTrXLNet):
NAME = "GTrXLNet"
# ModelCatalog.register_custom_model(TrainingModel.NAME, TrainingModel)
_global_registry.register(RLLIB_MODEL, TrainingModel.NAME, TrainingModel)
class RLLibTFSavedModelAgent(Agent):
def __init__(self, path_to_model, observation_space):
path_to_model = str(path_to_model) # might be a str or a Path, normalize to str
self._prep = ModelCatalog.get_preprocessor_for_space(observation_space)
self._sess = tf.compat.v1.Session(graph=tf.Graph())
tf.compat.v1.saved_model.load(
self._sess, export_dir=path_to_model, tags=["serve"]
)
self._output_node = self._sess.graph.get_tensor_by_name("default_policy/add:0")
self._input_node = self._sess.graph.get_tensor_by_name(
"default_policy/observation:0"
)
from grl.rl_apps.scenarios.trainer_configs.defaults import GRL_DEFAULT_OPENSPIEL_POKER_DQN_PARAMS, \
GRL_DEFAULT_POKER_PPO_PARAMS
from grl.rllib_tools.action_dists import TorchGaussianSquashedGaussian
from grl.rllib_tools.models.valid_actions_fcnet import get_valid_action_fcn_class_for_env
from grl.rllib_tools.valid_actions_epsilon_greedy import ValidActionsEpsilonGreedy
class _PokerAndOshiBetaTorchDist(TorchBeta):
def __init__(self, inputs, model):
super(_PokerAndOshiBetaTorchDist, self).__init__(inputs,
model,
low=-1.0,
high=1.0)
_global_registry.register(RLLIB_ACTION_DIST, "PokerAndOshiBetaTorchDist",
_PokerAndOshiBetaTorchDist)
_global_registry.register(RLLIB_ACTION_DIST, "TorchGaussianSquashedGaussian",
TorchGaussianSquashedGaussian)
def psro_kuhn_dqn_params_openspiel(env: MultiAgentEnv) -> Dict[str, Any]:
return GRL_DEFAULT_OPENSPIEL_POKER_DQN_PARAMS
def psro_leduc_dqn_params_openspiel(env: MultiAgentEnv) -> Dict[str, Any]:
return merge_dicts(
GRL_DEFAULT_OPENSPIEL_POKER_DQN_PARAMS,
{
# === Exploration Settings ===
"exploration_config": {
# The Exploration class to use.
def testStartFailure(self):
_global_registry.register(TRAINABLE_CLASS, "asdf", None)
trial = Trial("asdf", resources=Resources(1, 0))
self.trial_executor.start_trial(trial)
self.assertEqual(Trial.ERROR, trial.status)
