def validate_config(self, config: TrainerConfigDict) -> None:
"""Checks and updates the config based on settings."""
# Call super's validation method.
super().validate_config(config)
if config["exploration_config"]["type"] == "ParameterNoise":
if config["batch_mode"] != "complete_episodes":
logger.warning(
"ParameterNoise Exploration requires `batch_mode` to be "
"'complete_episodes'. Setting batch_mode="
"complete_episodes.")
config["batch_mode"] = "complete_episodes"
if config.get("noisy", False):
raise ValueError(
"ParameterNoise Exploration and `noisy` network cannot be"
" used at the same time!")
validate_buffer_config(config)
# Multi-agent mode and multi-GPU optimizer.
if config["multiagent"]["policies"] and not config["simple_optimizer"]:
logger.info(
"In multi-agent mode, policies will be optimized sequentially"
" by the multi-GPU optimizer. Consider setting "
"`simple_optimizer=True` if this doesn't work for you.")
def validate_config(self, config: TrainerConfigDict) -> None:
# Call super's validation method.
super().validate_config(config)
validate_buffer_config(config)
if config["num_gpus"] > 1:
raise ValueError("`num_gpus` > 1 not yet supported for MARWIL!")
if config["postprocess_inputs"] is False and config["beta"] > 0.0:
raise ValueError(
"`postprocess_inputs` must be True for MARWIL (to "
"calculate accum., discounted returns)!")
def validate_config(self, config: AlgorithmConfigDict) -> None:
# Call super's validation method.
super().validate_config(config)
# TODO: Move this to super()?
validate_buffer_config(config)
if config["beta"] < 0.0 or config["beta"] > 1.0:
raise ValueError("`beta` must be within 0.0 and 1.0!")
if config["num_gpus"] > 1:
raise ValueError("`num_gpus` > 1 not yet supported for MARWIL!")
if config["postprocess_inputs"] is False and config["beta"] > 0.0:
raise ValueError(
"`postprocess_inputs` must be True for MARWIL (to "
"calculate accum., discounted returns)!")
def validate_config(self, config: TrainerConfigDict) -> None:
"""Checks and updates the config based on settings."""
# Call super's validation method.
super().validate_config(config)
validate_buffer_config(config)
def validate_config(self, config: TrainerConfigDict) -> None:
super().validate_config(config)
validate_buffer_config(config)
config = DQNConfig().training(replay_buffer_config={"type": ReplayBuffer}).to_dict()
another_config = (
DQNConfig().training(replay_buffer_config={"type": "ReplayBuffer"}).to_dict()
)
yet_another_config = (
DQNConfig()
.training(
replay_buffer_config={"type": "ray.rllib.utils.replay_buffers.ReplayBuffer"}
)
.to_dict()
)
validate_buffer_config(config)
validate_buffer_config(another_config)
validate_buffer_config(yet_another_config)
# After validation, all three configs yield the same effective config
assert config == another_config == yet_another_config
# __sphinx_doc_replay_buffer_type_specification__end__
# __sphinx_doc_replay_buffer_basic_interaction__begin__
# We choose fragments because it does not impose restrictions on our batch to be added
buffer = ReplayBuffer(capacity=2, storage_unit=StorageUnit.FRAGMENTS)
dummy_batch = SampleBatch({"a": [1], "b": [2]})
buffer.add(dummy_batch)
buffer.sample(2)
# Because elements can be sampled multiple times, we receive a concatenated version