ray/rllib/agents/trainer.py

import copy
from datetime import datetime
import functools
import gym
import logging
import numpy as np
import os
import pickle
import tempfile
import time
from typing import Callable, Dict, List, Optional, Type, Union

import ray
from ray.actor import ActorHandle
from ray.exceptions import RayError
from ray.rllib.agents.callbacks import DefaultCallbacks
from ray.rllib.env.env_context import EnvContext
from ray.rllib.env.utils import gym_env_creator
from ray.rllib.evaluation.collectors.simple_list_collector import \
    SimpleListCollector
from ray.rllib.evaluation.metrics import collect_metrics
from ray.rllib.evaluation.rollout_worker import RolloutWorker
from ray.rllib.evaluation.worker_set import WorkerSet
from ray.rllib.models import MODEL_DEFAULTS
from ray.rllib.policy.policy import Policy, PolicySpec
from ray.rllib.policy.sample_batch import DEFAULT_POLICY_ID
from ray.rllib.utils import deep_update, FilterManager, merge_dicts
from ray.rllib.utils.annotations import Deprecated, DeveloperAPI, override, \
    PublicAPI
from ray.rllib.utils.debug import update_global_seed_if_necessary
from ray.rllib.utils.deprecation import deprecation_warning, DEPRECATED_VALUE
from ray.rllib.utils.error import EnvError, ERR_MSG_INVALID_ENV_DESCRIPTOR
from ray.rllib.utils.framework import try_import_tf
from ray.rllib.utils.from_config import from_config
from ray.rllib.utils.multi_agent import check_multi_agent
from ray.rllib.utils.spaces import space_utils
from ray.rllib.utils.typing import AgentID, EnvInfoDict, EnvType, EpisodeID, \
    PartialTrainerConfigDict, PolicyID, ResultDict, TensorStructType, \
    TrainerConfigDict
from ray.tune.logger import Logger, UnifiedLogger
from ray.tune.registry import ENV_CREATOR, register_env, _global_registry
from ray.tune.resources import Resources
from ray.tune.result import DEFAULT_RESULTS_DIR
from ray.tune.trainable import Trainable
from ray.tune.trial import ExportFormat
from ray.tune.utils.placement_groups import PlacementGroupFactory
from ray.util import log_once

tf1, tf, tfv = try_import_tf()

logger = logging.getLogger(__name__)

# Max number of times to retry a worker failure. We shouldn't try too many
# times in a row since that would indicate a persistent cluster issue.
MAX_WORKER_FAILURE_RETRIES = 3

# yapf: disable
# __sphinx_doc_begin__
COMMON_CONFIG: TrainerConfigDict = {
    # === Settings for Rollout Worker processes ===
    # Number of rollout worker actors to create for parallel sampling. Setting
    # this to 0 will force rollouts to be done in the trainer actor.
    "num_workers": 2,
    # Number of environments to evaluate vector-wise per worker. This enables
    # model inference batching, which can improve performance for inference
    # bottlenecked workloads.
    "num_envs_per_worker": 1,
    # When `num_workers` > 0, the driver (local_worker; worker-idx=0) does not
    # need an environment. This is because it doesn't have to sample (done by
    # remote_workers; worker_indices > 0) nor evaluate (done by evaluation
    # workers; see below).
    "create_env_on_driver": False,
    # Divide episodes into fragments of this many steps each during rollouts.
    # Sample batches of this size are collected from rollout workers and
    # combined into a larger batch of `train_batch_size` for learning.
    #
    # For example, given rollout_fragment_length=100 and train_batch_size=1000:
    #   1. RLlib collects 10 fragments of 100 steps each from rollout workers.
    #   2. These fragments are concatenated and we perform an epoch of SGD.
    #
    # When using multiple envs per worker, the fragment size is multiplied by
    # `num_envs_per_worker`. This is since we are collecting steps from
    # multiple envs in parallel. For example, if num_envs_per_worker=5, then
    # rollout workers will return experiences in chunks of 5*100 = 500 steps.
    #
    # The dataflow here can vary per algorithm. For example, PPO further
    # divides the train batch into minibatches for multi-epoch SGD.
    "rollout_fragment_length": 200,
    # How to build per-Sampler (RolloutWorker) batches, which are then
    # usually concat'd to form the train batch. Note that "steps" below can
    # mean different things (either env- or agent-steps) and depends on the
    # `count_steps_by` (multiagent) setting below.
    # truncate_episodes: Each produced batch (when calling
    #   RolloutWorker.sample()) will contain exactly `rollout_fragment_length`
    #   steps. This mode guarantees evenly sized batches, but increases
    #   variance as the future return must now be estimated at truncation
    #   boundaries.
    # complete_episodes: Each unroll happens exactly over one episode, from
    #   beginning to end. Data collection will not stop unless the episode
    #   terminates or a configured horizon (hard or soft) is hit.
    "batch_mode": "truncate_episodes",

    # === Settings for the Trainer process ===
    # Discount factor of the MDP.
    "gamma": 0.99,
    # The default learning rate.
    "lr": 0.0001,
    # Training batch size, if applicable. Should be >= rollout_fragment_length.
    # Samples batches will be concatenated together to a batch of this size,
    # which is then passed to SGD.
    "train_batch_size": 200,
    # Arguments to pass to the policy model. See models/catalog.py for a full
    # list of the available model options.
    "model": MODEL_DEFAULTS,
    # Arguments to pass to the policy optimizer. These vary by optimizer.
    "optimizer": {},

    # === Environment Settings ===
    # Number of steps after which the episode is forced to terminate. Defaults
    # to `env.spec.max_episode_steps` (if present) for Gym envs.
    "horizon": None,
    # Calculate rewards but don't reset the environment when the horizon is
    # hit. This allows value estimation and RNN state to span across logical
    # episodes denoted by horizon. This only has an effect if horizon != inf.
    "soft_horizon": False,
    # Don't set 'done' at the end of the episode.
    # In combination with `soft_horizon`, this works as follows:
    # - no_done_at_end=False soft_horizon=False:
    #   Reset env and add `done=True` at end of each episode.
    # - no_done_at_end=True soft_horizon=False:
    #   Reset env, but do NOT add `done=True` at end of the episode.
    # - no_done_at_end=False soft_horizon=True:
    #   Do NOT reset env at horizon, but add `done=True` at the horizon
    #   (pretending the episode has terminated).
    # - no_done_at_end=True soft_horizon=True:
    #   Do NOT reset env at horizon and do NOT add `done=True` at the horizon.
    "no_done_at_end": False,
    # The environment specifier:
    # This can either be a tune-registered env, via
    # `tune.register_env([name], lambda env_ctx: [env object])`,
    # or a string specifier of an RLlib supported type. In the latter case,
    # RLlib will try to interpret the specifier as either an openAI gym env,
    # a PyBullet env, a ViZDoomGym env, or a fully qualified classpath to an
    # Env class, e.g. "ray.rllib.examples.env.random_env.RandomEnv".
    "env": None,
    # The observation- and action spaces for the Policies of this Trainer.
    # Use None for automatically inferring these from the given env.
    "observation_space": None,
    "action_space": None,
    # Arguments dict passed to the env creator as an EnvContext object (which
    # is a dict plus the properties: num_workers, worker_index, vector_index,
    # and remote).
    "env_config": {},
    # If using num_envs_per_worker > 1, whether to create those new envs in
    # remote processes instead of in the same worker. This adds overheads, but
    # can make sense if your envs can take much time to step / reset
    # (e.g., for StarCraft). Use this cautiously; overheads are significant.
    "remote_worker_envs": False,
    # Timeout that remote workers are waiting when polling environments.
    # 0 (continue when at least one env is ready) is a reasonable default,
    # but optimal value could be obtained by measuring your environment
    # step / reset and model inference perf.
    "remote_env_batch_wait_ms": 0,
    # A callable taking the last train results, the base env and the env
    # context as args and returning a new task to set the env to.
    # The env must be a `TaskSettableEnv` sub-class for this to work.
    # See `examples/curriculum_learning.py` for an example.
    "env_task_fn": None,
    # If True, try to render the environment on the local worker or on worker
    # 1 (if num_workers > 0). For vectorized envs, this usually means that only
    # the first sub-environment will be rendered.
    # In order for this to work, your env will have to implement the
    # `render()` method which either:
    # a) handles window generation and rendering itself (returning True) or
    # b) returns a numpy uint8 image of shape [height x width x 3 (RGB)].
    "render_env": False,
    # If True, stores videos in this relative directory inside the default
    # output dir (~/ray_results/...). Alternatively, you can specify an
    # absolute path (str), in which the env recordings should be
    # stored instead.
    # Set to False for not recording anything.
    # Note: This setting replaces the deprecated `monitor` key.
    "record_env": False,
    # Whether to clip rewards during Policy's postprocessing.
    # None (default): Clip for Atari only (r=sign(r)).
    # True: r=sign(r): Fixed rewards -1.0, 1.0, or 0.0.
    # False: Never clip.
    # [float value]: Clip at -value and + value.
    # Tuple[value1, value2]: Clip at value1 and value2.
    "clip_rewards": None,
    # If True, RLlib will learn entirely inside a normalized action space
    # (0.0 centered with small stddev; only affecting Box components).
    # We will unsquash actions (and clip, just in case) to the bounds of
    # the env's action space before sending actions back to the env.
    "normalize_actions": True,
    # If True, RLlib will clip actions according to the env's bounds
    # before sending them back to the env.
    # TODO: (sven) This option should be obsoleted and always be False.
    "clip_actions": False,
    # Whether to use "rllib" or "deepmind" preprocessors by default
    # Set to None for using no preprocessor. In this case, the model will have
    # to handle possibly complex observations from the environment.
    "preprocessor_pref": "deepmind",

    # === Debug Settings ===
    # Set the ray.rllib.* log level for the agent process and its workers.
    # Should be one of DEBUG, INFO, WARN, or ERROR. The DEBUG level will also
    # periodically print out summaries of relevant internal dataflow (this is
    # also printed out once at startup at the INFO level). When using the
    # `rllib train` command, you can also use the `-v` and `-vv` flags as
    # shorthand for INFO and DEBUG.
    "log_level": "WARN",
    # Callbacks that will be run during various phases of training. See the
    # `DefaultCallbacks` class and `examples/custom_metrics_and_callbacks.py`
    # for more usage information.
    "callbacks": DefaultCallbacks,
    # Whether to attempt to continue training if a worker crashes. The number
    # of currently healthy workers is reported as the "num_healthy_workers"
    # metric.
    "ignore_worker_failures": False,
    # Log system resource metrics to results. This requires `psutil` to be
    # installed for sys stats, and `gputil` for GPU metrics.
    "log_sys_usage": True,
    # Use fake (infinite speed) sampler. For testing only.
    "fake_sampler": False,

    # === Deep Learning Framework Settings ===
    # tf: TensorFlow (static-graph)
    # tf2: TensorFlow 2.x (eager)
    # tfe: TensorFlow eager
    # torch: PyTorch
    "framework": "tf",
    # Enable tracing in eager mode. This greatly improves performance, but
    # makes it slightly harder to debug since Python code won't be evaluated
    # after the initial eager pass. Only possible if framework=tfe.
    "eager_tracing": False,

    # === Exploration Settings ===
    # Default exploration behavior, iff `explore`=None is passed into
    # compute_action(s).
    # Set to False for no exploration behavior (e.g., for evaluation).
    "explore": True,
    # Provide a dict specifying the Exploration object's config.
    "exploration_config": {
        # The Exploration class to use. In the simplest case, this is the name
        # (str) of any class present in the `rllib.utils.exploration` package.
        # You can also provide the python class directly or the full location
        # of your class (e.g. "ray.rllib.utils.exploration.epsilon_greedy.
        # EpsilonGreedy").
        "type": "StochasticSampling",
        # Add constructor kwargs here (if any).
    },
    # === Evaluation Settings ===
    # Evaluate with every `evaluation_interval` training iterations.
    # The evaluation stats will be reported under the "evaluation" metric key.
    # Note that evaluation is currently not parallelized, and that for Ape-X
    # metrics are already only reported for the lowest epsilon workers.
    "evaluation_interval": None,
    # Number of episodes to run in total per evaluation period.
    # If using multiple evaluation workers (evaluation_num_workers > 1),
    # episodes will be split amongst these.
    # If "auto":
    # - evaluation_parallel_to_training=True: Will run as many episodes as the
    #   training step takes.
    # - evaluation_parallel_to_training=False: Error.
    "evaluation_num_episodes": 10,
    # Whether to run evaluation in parallel to a Trainer.train() call
    # using threading. Default=False.
    # E.g. evaluation_interval=2 -> For every other training iteration,
    # the Trainer.train() and Trainer.evaluate() calls run in parallel.
    # Note: This is experimental. Possible pitfalls could be race conditions
    # for weight synching at the beginning of the evaluation loop.
    "evaluation_parallel_to_training": False,
    # Internal flag that is set to True for evaluation workers.
    "in_evaluation": False,
    # Typical usage is to pass extra args to evaluation env creator
    # and to disable exploration by computing deterministic actions.
    # IMPORTANT NOTE: Policy gradient algorithms are able to find the optimal
    # policy, even if this is a stochastic one. Setting "explore=False" here
    # will result in the evaluation workers not using this optimal policy!
    "evaluation_config": {
        # Example: overriding env_config, exploration, etc:
        # "env_config": {...},
        # "explore": False
    },
    # Number of parallel workers to use for evaluation. Note that this is set
    # to zero by default, which means evaluation will be run in the trainer
    # process (only if evaluation_interval is not None). If you increase this,
    # it will increase the Ray resource usage of the trainer since evaluation
    # workers are created separately from rollout workers (used to sample data
    # for training).
    "evaluation_num_workers": 0,
    # Customize the evaluation method. This must be a function of signature
    # (trainer: Trainer, eval_workers: WorkerSet) -> metrics: dict. See the
    # Trainer.evaluate() method to see the default implementation. The
    # trainer guarantees all eval workers have the latest policy state before
    # this function is called.
    "custom_eval_function": None,

    # === Advanced Rollout Settings ===
    # Use a background thread for sampling (slightly off-policy, usually not
    # advisable to turn on unless your env specifically requires it).
    "sample_async": False,

    # The SampleCollector class to be used to collect and retrieve
    # environment-, model-, and sampler data. Override the SampleCollector base
    # class to implement your own collection/buffering/retrieval logic.
    "sample_collector": SimpleListCollector,

    # Element-wise observation filter, either "NoFilter" or "MeanStdFilter".
    "observation_filter": "NoFilter",
    # Whether to synchronize the statistics of remote filters.
    "synchronize_filters": True,
    # Configures TF for single-process operation by default.
    "tf_session_args": {
        # note: overridden by `local_tf_session_args`
        "intra_op_parallelism_threads": 2,
        "inter_op_parallelism_threads": 2,
        "gpu_options": {
            "allow_growth": True,
        },
        "log_device_placement": False,
        "device_count": {
            "CPU": 1
        },
        # Required by multi-GPU (num_gpus > 1).
        "allow_soft_placement": True,
    },
    # Override the following tf session args on the local worker
    "local_tf_session_args": {
        # Allow a higher level of parallelism by default, but not unlimited
        # since that can cause crashes with many concurrent drivers.
        "intra_op_parallelism_threads": 8,
        "inter_op_parallelism_threads": 8,
    },
    # Whether to LZ4 compress individual observations.
    "compress_observations": False,
    # Wait for metric batches for at most this many seconds. Those that
    # have not returned in time will be collected in the next train iteration.
    "collect_metrics_timeout": 180,
    # Smooth metrics over this many episodes.
    "metrics_smoothing_episodes": 100,
    # Minimum time per train iteration (frequency of metrics reporting).
    "min_iter_time_s": 0,
    # Minimum env steps to optimize for per train call. This value does
    # not affect learning, only the length of train iterations.
    "timesteps_per_iteration": 0,
    # This argument, in conjunction with worker_index, sets the random seed of
    # each worker, so that identically configured trials will have identical
    # results. This makes experiments reproducible.
    "seed": None,
    # Any extra python env vars to set in the trainer process, e.g.,
    # {"OMP_NUM_THREADS": "16"}
    "extra_python_environs_for_driver": {},
    # The extra python environments need to set for worker processes.
    "extra_python_environs_for_worker": {},

    # === Resource Settings ===
    # Number of GPUs to allocate to the trainer process. Note that not all
    # algorithms can take advantage of trainer GPUs. Support for multi-GPU
    # is currently only available for tf-[PPO/IMPALA/DQN/PG].
    # This can be fractional (e.g., 0.3 GPUs).
    "num_gpus": 0,
    # Set to True for debugging (multi-)?GPU funcitonality on a CPU machine.
    # GPU towers will be simulated by graphs located on CPUs in this case.
    # Use `num_gpus` to test for different numbers of fake GPUs.
    "_fake_gpus": False,
    # Number of CPUs to allocate per worker.
    "num_cpus_per_worker": 1,
    # Number of GPUs to allocate per worker. This can be fractional. This is
    # usually needed only if your env itself requires a GPU (i.e., it is a
    # GPU-intensive video game), or model inference is unusually expensive.
    "num_gpus_per_worker": 0,
    # Any custom Ray resources to allocate per worker.
    "custom_resources_per_worker": {},
    # Number of CPUs to allocate for the trainer. Note: this only takes effect
    # when running in Tune. Otherwise, the trainer runs in the main program.
    "num_cpus_for_driver": 1,
    # The strategy for the placement group factory returned by
    # `Trainer.default_resource_request()`. A PlacementGroup defines, which
    # devices (resources) should always be co-located on the same node.
    # For example, a Trainer with 2 rollout workers, running with
    # num_gpus=1 will request a placement group with the bundles:
    # [{"gpu": 1, "cpu": 1}, {"cpu": 1}, {"cpu": 1}], where the first bundle is
    # for the driver and the other 2 bundles are for the two workers.
    # These bundles can now be "placed" on the same or different
    # nodes depending on the value of `placement_strategy`:
    # "PACK": Packs bundles into as few nodes as possible.
    # "SPREAD": Places bundles across distinct nodes as even as possible.
    # "STRICT_PACK": Packs bundles into one node. The group is not allowed
    #   to span multiple nodes.
    # "STRICT_SPREAD": Packs bundles across distinct nodes.
    "placement_strategy": "PACK",

    # === Offline Datasets ===
    # Specify how to generate experiences:
    #  - "sampler": Generate experiences via online (env) simulation (default).
    #  - A local directory or file glob expression (e.g., "/tmp/*.json").
    #  - A list of individual file paths/URIs (e.g., ["/tmp/1.json",
    #    "s3://bucket/2.json"]).
    #  - A dict with string keys and sampling probabilities as values (e.g.,
    #    {"sampler": 0.4, "/tmp/*.json": 0.4, "s3://bucket/expert.json": 0.2}).
    #  - A callable that returns a ray.rllib.offline.InputReader.
    #  - A string key that indexes a callable with tune.registry.register_input
    "input": "sampler",
    # Arguments accessible from the IOContext for configuring custom input
    "input_config": {},
    # True, if the actions in a given offline "input" are already normalized
    # (between -1.0 and 1.0). This is usually the case when the offline
    # file has been generated by another RLlib algorithm (e.g. PPO or SAC),
    # while "normalize_actions" was set to True.
    "actions_in_input_normalized": False,
    # Specify how to evaluate the current policy. This only has an effect when
    # reading offline experiences ("input" is not "sampler").
    # Available options:
    #  - "wis": the weighted step-wise importance sampling estimator.
    #  - "is": the step-wise importance sampling estimator.
    #  - "simulation": run the environment in the background, but use
    #    this data for evaluation only and not for learning.
    "input_evaluation": ["is", "wis"],
    # Whether to run postprocess_trajectory() on the trajectory fragments from
    # offline inputs. Note that postprocessing will be done using the *current*
    # policy, not the *behavior* policy, which is typically undesirable for
    # on-policy algorithms.
    "postprocess_inputs": False,
    # If positive, input batches will be shuffled via a sliding window buffer
    # of this number of batches. Use this if the input data is not in random
    # enough order. Input is delayed until the shuffle buffer is filled.
    "shuffle_buffer_size": 0,
    # Specify where experiences should be saved:
    #  - None: don't save any experiences
    #  - "logdir" to save to the agent log dir
    #  - a path/URI to save to a custom output directory (e.g., "s3://bucket/")
    #  - a function that returns a rllib.offline.OutputWriter
    "output": None,
    # What sample batch columns to LZ4 compress in the output data.
    "output_compress_columns": ["obs", "new_obs"],
    # Max output file size before rolling over to a new file.
    "output_max_file_size": 64 * 1024 * 1024,

    # === Settings for Multi-Agent Environments ===
    "multiagent": {
        # Map of type MultiAgentPolicyConfigDict from policy ids to tuples
        # of (policy_cls, obs_space, act_space, config). This defines the
        # observation and action spaces of the policies and any extra config.
        "policies": {},
        # Keep this many policies in the "policy_map" (before writing
        # least-recently used ones to disk/S3).
        "policy_map_capacity": 100,
        # Where to store overflowing (least-recently used) policies?
        # Could be a directory (str) or an S3 location. None for using
        # the default output dir.
        "policy_map_cache": None,
        # Function mapping agent ids to policy ids.
        "policy_mapping_fn": None,
        # Optional list of policies to train, or None for all policies.
        "policies_to_train": None,
        # Optional function that can be used to enhance the local agent
        # observations to include more state.
        # See rllib/evaluation/observation_function.py for more info.
        "observation_fn": None,
        # When replay_mode=lockstep, RLlib will replay all the agent
        # transitions at a particular timestep together in a batch. This allows
        # the policy to implement differentiable shared computations between
        # agents it controls at that timestep. When replay_mode=independent,
        # transitions are replayed independently per policy.
        "replay_mode": "independent",
        # Which metric to use as the "batch size" when building a
        # MultiAgentBatch. The two supported values are:
        # env_steps: Count each time the env is "stepped" (no matter how many
        #   multi-agent actions are passed/how many multi-agent observations
        #   have been returned in the previous step).
        # agent_steps: Count each individual agent step as one step.
        "count_steps_by": "env_steps",
    },

    # === Logger ===
    # Define logger-specific configuration to be used inside Logger
    # Default value None allows overwriting with nested dicts
    "logger_config": None,

    # === API deprecations/simplifications/changes ===
    # Experimental flag.
    # If True, TFPolicy will handle more than one loss/optimizer.
    # Set this to True, if you would like to return more than
    # one loss term from your `loss_fn` and an equal number of optimizers
    # from your `optimizer_fn`.
    # In the future, the default for this will be True.
    "_tf_policy_handles_more_than_one_loss": False,
    # Experimental flag.
    # If True, no (observation) preprocessor will be created and
    # observations will arrive in model as they are returned by the env.
    # In the future, the default for this will be True.
    "_disable_preprocessor_api": False,

    # === Deprecated keys ===
    # Uses the sync samples optimizer instead of the multi-gpu one. This is
    # usually slower, but you might want to try it if you run into issues with
    # the default optimizer.
    # This will be set automatically from now on.
    "simple_optimizer": DEPRECATED_VALUE,
    # Whether to write episode stats and videos to the agent log dir. This is
    # typically located in ~/ray_results.
    "monitor": DEPRECATED_VALUE,
}
# __sphinx_doc_end__
# yapf: enable


@DeveloperAPI
def with_common_config(
        extra_config: PartialTrainerConfigDict) -> TrainerConfigDict:
    """Returns the given config dict merged with common agent confs.

    Args:
        extra_config (PartialTrainerConfigDict): A user defined partial config
            which will get merged with COMMON_CONFIG and returned.

    Returns:
        TrainerConfigDict: The merged config dict resulting of COMMON_CONFIG
            plus `extra_config`.
    """
    return Trainer.merge_trainer_configs(
        COMMON_CONFIG, extra_config, _allow_unknown_configs=True)


@PublicAPI
class Trainer(Trainable):
    """A trainer coordinates the optimization of one or more RL policies.

    All RLlib trainers extend this base class, e.g., the A3CTrainer implements
    the A3C algorithm for single and multi-agent training.

    Trainer objects retain internal model state between calls to train(), so
    you should create a new trainer instance for each training session.

    Attributes:
        env_creator (func): Function that creates a new training env.
        config (obj): Algorithm-specific configuration data.
        logdir (str): Directory in which training outputs should be placed.
    """
    # Whether to allow unknown top-level config keys.
    _allow_unknown_configs = False

    # List of top-level keys with value=dict, for which new sub-keys are
    # allowed to be added to the value dict.
    _allow_unknown_subkeys = [
        "tf_session_args", "local_tf_session_args", "env_config", "model",
        "optimizer", "multiagent", "custom_resources_per_worker",
        "evaluation_config", "exploration_config",
        "extra_python_environs_for_driver", "extra_python_environs_for_worker",
        "input_config"
    ]

    # List of top level keys with value=dict, for which we always override the
    # entire value (dict), iff the "type" key in that value dict changes.
    _override_all_subkeys_if_type_changes = ["exploration_config"]

    @PublicAPI
    def __init__(self,
                 config: TrainerConfigDict = None,
                 env: str = None,
                 logger_creator: Callable[[], Logger] = None):
        """Initialize an RLLib trainer.

        Args:
            config (dict): Algorithm-specific configuration data.
            env (str): Name of the environment to use. Note that this can also
                be specified as the `env` key in config.
            logger_creator (func): Function that creates a ray.tune.Logger
                object. If unspecified, a default logger is created.
        """

        # User provided config (this is w/o the default Trainer's
        # `COMMON_CONFIG` (see above)). Will get merged with COMMON_CONFIG
        # in self.setup().
        config = config or {}

        # Trainers allow env ids to be passed directly to the constructor.
        self._env_id = self._register_if_needed(
            env or config.get("env"), config)

        # Placeholder for a local replay buffer instance.
        self.local_replay_buffer = None

        # Create a default logger creator if no logger_creator is specified
        if logger_creator is None:
            # Default logdir prefix containing the agent's name and the
            # env id.
            timestr = datetime.today().strftime("%Y-%m-%d_%H-%M-%S")
            logdir_prefix = "{}_{}_{}".format(self._name, self._env_id,
                                              timestr)
            if not os.path.exists(DEFAULT_RESULTS_DIR):
                os.makedirs(DEFAULT_RESULTS_DIR)
            logdir = tempfile.mkdtemp(
                prefix=logdir_prefix, dir=DEFAULT_RESULTS_DIR)

            # Allow users to more precisely configure the created logger
            # via "logger_config.type".
            if config.get(
                    "logger_config") and "type" in config["logger_config"]:

                def default_logger_creator(config):
                    """Creates a custom logger with the default prefix."""
                    cfg = config["logger_config"].copy()
                    cls = cfg.pop("type")
                    # Provide default for logdir, in case the user does
                    # not specify this in the "logger_config" dict.
                    logdir_ = cfg.pop("logdir", logdir)
                    return from_config(cls=cls, _args=[cfg], logdir=logdir_)

            # If no `type` given, use tune's UnifiedLogger as last resort.
            else:

                def default_logger_creator(config):
                    """Creates a Unified logger with the default prefix."""
                    return UnifiedLogger(config, logdir, loggers=None)

            logger_creator = default_logger_creator

        super().__init__(config, logger_creator)

    @classmethod
    @override(Trainable)
    def default_resource_request(
            cls, config: PartialTrainerConfigDict) -> \
            Union[Resources, PlacementGroupFactory]:
        cf = dict(cls._default_config, **config)

        eval_config = cf["evaluation_config"]

        # TODO(ekl): add custom resources here once tune supports them
        # Return PlacementGroupFactory containing all needed resources
        # (already properly defined as device bundles).
        return PlacementGroupFactory(
            bundles=[{
                # Driver.
                "CPU": cf["num_cpus_for_driver"],
                "GPU": 0 if cf["_fake_gpus"] else cf["num_gpus"],
            }] + [
                {
                    # RolloutWorkers.
                    "CPU": cf["num_cpus_per_worker"],
                    "GPU": cf["num_gpus_per_worker"],
                } for _ in range(cf["num_workers"])
            ] + ([
                {
                    # Evaluation workers.
                    # Note: The local eval worker is located on the driver CPU.
                    "CPU": eval_config.get("num_cpus_per_worker",
                                           cf["num_cpus_per_worker"]),
                    "GPU": eval_config.get("num_gpus_per_worker",
                                           cf["num_gpus_per_worker"]),
                } for _ in range(cf["evaluation_num_workers"])
            ] if cf["evaluation_interval"] else []),
            strategy=config.get("placement_strategy", "PACK"))

    @override(Trainable)
    @PublicAPI
    def train(self) -> ResultDict:
        """Overrides super.train to synchronize global vars."""

        result = None
        for _ in range(1 + MAX_WORKER_FAILURE_RETRIES):
            try:
                result = Trainable.train(self)
            except RayError as e:
                if self.config["ignore_worker_failures"]:
                    logger.exception(
                        "Error in train call, attempting to recover")
                    self._try_recover()
                else:
                    logger.info(
                        "Worker crashed during call to train(). To attempt to "
                        "continue training without the failed worker, set "
                        "`'ignore_worker_failures': True`.")
                    raise e
            except Exception as e:
                time.sleep(0.5)  # allow logs messages to propagate
                raise e
            else:
                break
        if result is None:
            raise RuntimeError("Failed to recover from worker crash")

        if hasattr(self, "workers") and isinstance(self.workers, WorkerSet):
            self._sync_filters_if_needed(self.workers)

        return result

    def _sync_filters_if_needed(self, workers: WorkerSet):
        if self.config.get("observation_filter", "NoFilter") != "NoFilter":
            FilterManager.synchronize(
                workers.local_worker().filters,
                workers.remote_workers(),
                update_remote=self.config["synchronize_filters"])
            logger.debug("synchronized filters: {}".format(
                workers.local_worker().filters))

    @override(Trainable)
    def log_result(self, result: ResultDict):
        self.callbacks.on_train_result(trainer=self, result=result)
        # log after the callback is invoked, so that the user has a chance
        # to mutate the result
        Trainable.log_result(self, result)

    @override(Trainable)
    def setup(self, config: PartialTrainerConfigDict):
        env = self._env_id
        if env:
            config["env"] = env
            # An already registered env.
            if _global_registry.contains(ENV_CREATOR, env):
                self.env_creator = _global_registry.get(ENV_CREATOR, env)
            # A class specifier.
            elif "." in env:

                def env_creator_from_classpath(env_context):
                    try:
                        env_obj = from_config(env, env_context)
                    except ValueError:
                        raise EnvError(
                            ERR_MSG_INVALID_ENV_DESCRIPTOR.format(env))
                    return env_obj

                self.env_creator = env_creator_from_classpath
            # Try gym/PyBullet/Vizdoom.
            else:
                self.env_creator = functools.partial(
                    gym_env_creator, env_descriptor=env)
        else:
            self.env_creator = lambda env_config: None

        # Merge the supplied config with the class default, but store the
        # user-provided one.
        self.raw_user_config = config
        self.config = self.merge_trainer_configs(self._default_config, config,
                                                 self._allow_unknown_configs)

        # Check and resolve DL framework settings.
        # Enable eager/tracing support.
        if tf1 and self.config["framework"] in ["tf2", "tfe"]:
            if self.config["framework"] == "tf2" and tfv < 2:
                raise ValueError("`framework`=tf2, but tf-version is < 2.0!")
            if not tf1.executing_eagerly():
                tf1.enable_eager_execution()
            logger.info("Executing eagerly, with eager_tracing={}".format(
                self.config["eager_tracing"]))
        if tf1 and not tf1.executing_eagerly() and \
                self.config["framework"] != "torch":
            logger.info("Tip: set framework=tfe or the --eager flag to enable "
                        "TensorFlow eager execution")

        # Set Trainer's seed after we have - if necessary - enabled
        # tf eager-execution.
        update_global_seed_if_necessary(
            config.get("framework"), config.get("seed"))

        self._validate_config(self.config, trainer_obj_or_none=self)
        if not callable(self.config["callbacks"]):
            raise ValueError(
                "`callbacks` must be a callable method that "
                "returns a subclass of DefaultCallbacks, got {}".format(
                    self.config["callbacks"]))
        self.callbacks = self.config["callbacks"]()
        log_level = self.config.get("log_level")
        if log_level in ["WARN", "ERROR"]:
            logger.info("Current log_level is {}. For more information, "
                        "set 'log_level': 'INFO' / 'DEBUG' or use the -v and "
                        "-vv flags.".format(log_level))
        if self.config.get("log_level"):
            logging.getLogger("ray.rllib").setLevel(self.config["log_level"])

        self._init(self.config, self.env_creator)

        # Evaluation setup.
        self.evaluation_workers = None
        self.evaluation_metrics = {}
        # Do automatic evaluation from time to time.
        if self.config.get("evaluation_interval"):
            # Update env_config with evaluation settings:
            extra_config = copy.deepcopy(self.config["evaluation_config"])
            # Assert that user has not unset "in_evaluation".
            assert "in_evaluation" not in extra_config or \
                extra_config["in_evaluation"] is True
            evaluation_config = merge_dicts(self.config, extra_config)
            # Validate evaluation config.
            self._validate_config(evaluation_config, trainer_obj_or_none=self)
            # Switch on complete_episode rollouts (evaluations are
            # always done on n complete episodes) and set the
            # `in_evaluation` flag. Also, make sure our rollout fragments
            # are short so we don't have more than one episode in one rollout.
            evaluation_config.update({
                "batch_mode": "complete_episodes",
                "rollout_fragment_length": 1,
                "in_evaluation": True,
            })
            logger.debug("using evaluation_config: {}".format(extra_config))
            # Create a separate evaluation worker set for evaluation.
            # If evaluation_num_workers=0, use the evaluation set's local
            # worker for evaluation, otherwise, use its remote workers
            # (parallelized evaluation).
            self.evaluation_workers = self._make_workers(
                env_creator=self.env_creator,
                validate_env=None,
                policy_class=self._policy_class,
                config=evaluation_config,
                num_workers=self.config["evaluation_num_workers"])

    @override(Trainable)
    def cleanup(self):
        if hasattr(self, "workers"):
            self.workers.stop()
        if hasattr(self, "optimizer") and self.optimizer:
            self.optimizer.stop()

    @override(Trainable)
    def save_checkpoint(self, checkpoint_dir: str) -> str:
        checkpoint_path = os.path.join(checkpoint_dir,
                                       "checkpoint-{}".format(self.iteration))
        pickle.dump(self.__getstate__(), open(checkpoint_path, "wb"))

        return checkpoint_path

    @override(Trainable)
    def load_checkpoint(self, checkpoint_path: str):
        extra_data = pickle.load(open(checkpoint_path, "rb"))
        self.__setstate__(extra_data)

    @DeveloperAPI
    def _make_workers(
            self, *, env_creator: Callable[[EnvContext], EnvType],
            validate_env: Optional[Callable[[EnvType, EnvContext], None]],
            policy_class: Type[Policy], config: TrainerConfigDict,
            num_workers: int) -> WorkerSet:
        """Default factory method for a WorkerSet running under this Trainer.

        Override this method by passing a custom `make_workers` into
        `build_trainer`.

        Args:
            env_creator (callable): A function that return and Env given an env
                config.
            validate_env (Optional[Callable[[EnvType, EnvContext], None]]):
                Optional callable to validate the generated environment (only
                on worker=0).
            policy (Type[Policy]): The Policy class to use for creating the
                policies of the workers.
            config (TrainerConfigDict): The Trainer's config.
            num_workers (int): Number of remote rollout workers to create.
                0 for local only.

        Returns:
            WorkerSet: The created WorkerSet.
        """
        return WorkerSet(
            env_creator=env_creator,
            validate_env=validate_env,
            policy_class=policy_class,
            trainer_config=config,
            num_workers=num_workers,
            logdir=self.logdir)

    @DeveloperAPI
    def _init(self, config: TrainerConfigDict,
              env_creator: Callable[[EnvContext], EnvType]):
        """Subclasses should override this for custom initialization."""
        raise NotImplementedError

    @Deprecated(new="Trainer.evaluate", error=False)
    def _evaluate(self) -> dict:
        return self.evaluate()

    @PublicAPI
    def evaluate(self, episodes_left_fn: Optional[Callable[[int], int]] = None
                 ) -> dict:
        """Evaluates current policy under `evaluation_config` settings.

        Note that this default implementation does not do anything beyond
        merging evaluation_config with the normal trainer config.

        Args:
            episodes_left_fn (Optional[Callable[[int], int]]): An optional
                callable taking the already run num episodes as only arg
                and returning the number of episodes left to run. It's used
                to find out whether evaluation should continue.
        """
        # In case we are evaluating (in a thread) parallel to training,
        # we may have to re-enable eager mode here (gets disabled in the
        # thread).
        if self.config.get("framework") in ["tf2", "tfe"] and \
                not tf.executing_eagerly():
            tf1.enable_eager_execution()

        # Call the `_before_evaluate` hook.
        self._before_evaluate()

        if self.evaluation_workers is not None:
            # Sync weights to the evaluation WorkerSet.
            self._sync_weights_to_workers(worker_set=self.evaluation_workers)
            self._sync_filters_if_needed(self.evaluation_workers)

        if self.config["custom_eval_function"]:
            logger.info("Running custom eval function {}".format(
                self.config["custom_eval_function"]))
            metrics = self.config["custom_eval_function"](
                self, self.evaluation_workers)
            if not metrics or not isinstance(metrics, dict):
                raise ValueError("Custom eval function must return "
                                 "dict of metrics, got {}.".format(metrics))
        else:
            # How many episodes do we need to run?
            # In "auto" mode (only for parallel eval + training): Run one
            # episode per eval worker.
            num_episodes = self.config["evaluation_num_episodes"] if \
                self.config["evaluation_num_episodes"] != "auto" else \
                (self.config["evaluation_num_workers"] or 1)

            # Default done-function returns True, whenever num episodes
            # have been completed.
            if episodes_left_fn is None:

                def episodes_left_fn(num_episodes_done):
                    return num_episodes - num_episodes_done

            logger.info(
                f"Evaluating current policy for {num_episodes} episodes.")

            metrics = None
            # No evaluation worker set ->
            # Do evaluation using the local worker. Expect error due to the
            # local worker not having an env.
            if self.evaluation_workers is None:
                try:
                    for _ in range(num_episodes):
                        self.workers.local_worker().sample()
                    metrics = collect_metrics(self.workers.local_worker())
                except ValueError as e:
                    if "RolloutWorker has no `input_reader` object" in \
                            e.args[0]:
                        raise ValueError(
                            "Cannot evaluate w/o an evaluation worker set in "
                            "the Trainer or w/o an env on the local worker!\n"
                            "Try one of the following:\n1) Set "
                            "`evaluation_interval` >= 0 to force creating a "
                            "separate evaluation worker set.\n2) Set "
                            "`create_env_on_driver=True` to force the local "
                            "(non-eval) worker to have an environment to "
                            "evaluate on.")
                    else:
                        raise e

            # Evaluation worker set only has local worker.
            elif self.config["evaluation_num_workers"] == 0:
                for _ in range(num_episodes):
                    self.evaluation_workers.local_worker().sample()

            # Evaluation worker set has n remote workers.
            else:
                # How many episodes have we run (across all eval workers)?
                num_episodes_done = 0
                round_ = 0
                while True:
                    episodes_left_to_do = episodes_left_fn(num_episodes_done)
                    if episodes_left_to_do <= 0:
                        break

                    round_ += 1
                    batches = ray.get([
                        w.sample.remote() for i, w in enumerate(
                            self.evaluation_workers.remote_workers())
                        if i < episodes_left_to_do
                    ])
                    # Per our config for the evaluation workers
                    # (`rollout_fragment_length=1` and
                    # `batch_mode=complete_episode`), we know that we'll have
                    # exactly one episode per returned batch.
                    num_episodes_done += len(batches)
                    logger.info(
                        f"Ran round {round_} of parallel evaluation "
                        f"({num_episodes_done}/{num_episodes} episodes done)")
            if metrics is None:
                metrics = collect_metrics(
                    self.evaluation_workers.local_worker(),
                    self.evaluation_workers.remote_workers())
        return {"evaluation": metrics}

    @DeveloperAPI
    def _before_evaluate(self):
        """Pre-evaluation callback."""
        pass

    @DeveloperAPI
    def _sync_weights_to_workers(
            self,
            *,
            worker_set: Optional[WorkerSet] = None,
            workers: Optional[List[RolloutWorker]] = None,
    ) -> None:
        """Sync "main" weights to given WorkerSet or list of workers."""
        assert worker_set is not None
        # Broadcast the new policy weights to all evaluation workers.
        logger.info("Synchronizing weights to workers.")
        weights = ray.put(self.workers.local_worker().save())
        worker_set.foreach_worker(lambda w: w.restore(ray.get(weights)))

    @PublicAPI
    def compute_single_action(
            self,
            observation: TensorStructType,
            state: List[TensorStructType] = None,
            prev_action: TensorStructType = None,
            prev_reward: float = None,
            info: EnvInfoDict = None,
            policy_id: PolicyID = DEFAULT_POLICY_ID,
            full_fetch: bool = False,
            explore: bool = None,
            unsquash_actions: Optional[bool] = None,
            clip_actions: Optional[bool] = None,
    ) -> TensorStructType:
        """Computes an action for the specified policy on the local worker.

        Note that you can also access the policy object through
        self.get_policy(policy_id) and call compute_single_action() on it
        directly.

        Args:
            observation (TensorStructType): observation from the environment.
            state (List[TensorStructType]): RNN hidden state, if any. If state
                is not None, then all of compute_single_action(...) is returned
                (computed action, rnn state(s), logits dictionary).
                Otherwise compute_single_action(...)[0] is returned
                (computed action).
            prev_action (TensorStructType): Previous action value, if any.
            prev_reward (float): Previous reward, if any.
            info (EnvInfoDict): info object, if any
            policy_id (PolicyID): Policy to query (only applies to
                multi-agent).
            full_fetch (bool): Whether to return extra action fetch results.
                This is always set to True if RNN state is specified.
            explore (bool): Whether to pick an exploitation or exploration
                action (default: None -> use self.config["explore"]).
            unsquash_actions (bool): Should actions be unsquashed according to
                 the env's/Policy's action space?
            clip_actions (bool): Should actions be clipped according to the
                env's/Policy's action space?

        Returns:
            any: The computed action if full_fetch=False, or
            tuple: The full output of policy.compute_actions() if
                full_fetch=True or we have an RNN-based Policy.

        Raises:
            KeyError: If the `policy_id` cannot be found in this Trainer's
                local worker.
        """
        policy = self.get_policy(policy_id)
        if policy is None:
            raise KeyError(
                f"PolicyID '{policy_id}' not found in PolicyMap of the "
                f"Trainer's local worker!")

        local_worker = self.workers.local_worker()

        if state is None:
            state = []

        # Check the preprocessor and preprocess, if necessary.
        pp = local_worker.preprocessors[policy_id]
        if pp and type(pp).__name__ != "NoPreprocessor":
            observation = pp.transform(observation)
        filtered_observation = local_worker.filters[policy_id](
            observation, update=False)

        # Compute the action.
        result = policy.compute_single_action(
            filtered_observation,
            state,
            prev_action,
            prev_reward,
            info,
            unsquash_actions=unsquash_actions,
            clip_actions=clip_actions,
            explore=explore)

        # Return 3-Tuple: Action, states, and extra-action fetches.
        if state or full_fetch:
            return result
        # Ensure backward compatibility.
        else:
            return result[0]

    @Deprecated(new="compute_single_action", error=False)
    def compute_action(self, *args, **kwargs):
        return self.compute_single_action(*args, **kwargs)

    @PublicAPI
    def compute_actions(
            self,
            observations: TensorStructType,
            state: List[TensorStructType] = None,
            prev_action: TensorStructType = None,
            prev_reward: TensorStructType = None,
            info=None,
            policy_id=DEFAULT_POLICY_ID,
            full_fetch=False,
            explore=None,
            normalize_actions=None,
            clip_actions=None,
    ):
        """Computes an action for the specified policy on the local Worker.

        Note that you can also access the policy object through
        self.get_policy(policy_id) and call compute_actions() on it directly.

        Args:
            observation (obj): observation from the environment.
            state (dict): RNN hidden state, if any. If state is not None,
                then all of compute_single_action(...) is returned
                (computed action, rnn state(s), logits dictionary).
                Otherwise compute_single_action(...)[0] is returned
                (computed action).
            prev_action (obj): previous action value, if any
            prev_reward (int): previous reward, if any
            info (dict): info object, if any
            policy_id (str): Policy to query (only applies to multi-agent).
            full_fetch (bool): Whether to return extra action fetch results.
                This is always set to True if RNN state is specified.
            explore (bool): Whether to pick an exploitation or exploration
                action (default: None -> use self.config["explore"]).
            normalize_actions (bool): Should actions be unsquashed according
                to the env's/Policy's action space?
            clip_actions (bool): Should actions be clipped according to the
                env's/Policy's action space?

        Returns:
            any: The computed action if full_fetch=False, or
            tuple: The full output of policy.compute_actions() if
                full_fetch=True or we have an RNN-based Policy.
        """
        # Preprocess obs and states.
        state_defined = state is not None
        policy = self.get_policy(policy_id)
        filtered_obs, filtered_state = [], []
        for agent_id, ob in observations.items():
            worker = self.workers.local_worker()
            preprocessed = worker.preprocessors[policy_id].transform(ob)
            filtered = worker.filters[policy_id](preprocessed, update=False)
            filtered_obs.append(filtered)
            if state is None:
                continue
            elif agent_id in state:
                filtered_state.append(state[agent_id])
            else:
                filtered_state.append(policy.get_initial_state())

        # Batch obs and states
        obs_batch = np.stack(filtered_obs)
        if state is None:
            state = []
        else:
            state = list(zip(*filtered_state))
            state = [np.stack(s) for s in state]

        # Batch compute actions
        actions, states, infos = policy.compute_actions(
            obs_batch,
            state,
            prev_action,
            prev_reward,
            info,
            normalize_actions=normalize_actions,
            clip_actions=clip_actions,
            explore=explore)

        # Unbatch actions for the environment
        atns, actions = space_utils.unbatch(actions), {}
        for key, atn in zip(observations, atns):
            actions[key] = atn

        # Unbatch states into a dict
        unbatched_states = {}
        for idx, agent_id in enumerate(observations):
            unbatched_states[agent_id] = [s[idx] for s in states]

        # Return only actions or full tuple
        if state_defined or full_fetch:
            return actions, unbatched_states, infos
        else:
            return actions

    @property
    def _name(self) -> str:
        """Subclasses should override this to declare their name."""
        raise NotImplementedError

    @property
    def _default_config(self) -> TrainerConfigDict:
        """Subclasses should override this to declare their default config."""
        raise NotImplementedError

    @PublicAPI
    def get_policy(self, policy_id: PolicyID = DEFAULT_POLICY_ID) -> Policy:
        """Return policy for the specified id, or None.

        Args:
            policy_id (PolicyID): ID of the policy to return.
        """
        return self.workers.local_worker().get_policy(policy_id)

    @PublicAPI
    def get_weights(self, policies: List[PolicyID] = None) -> dict:
        """Return a dictionary of policy ids to weights.

        Args:
            policies (list): Optional list of policies to return weights for,
                or None for all policies.
        """
        return self.workers.local_worker().get_weights(policies)

    @PublicAPI
    def set_weights(self, weights: Dict[PolicyID, dict]):
        """Set policy weights by policy id.

        Args:
            weights (dict): Map of policy ids to weights to set.
        """
        self.workers.local_worker().set_weights(weights)

    @PublicAPI
    def add_policy(
            self,
            policy_id: PolicyID,
            policy_cls: Type[Policy],
            *,
            observation_space: Optional[gym.spaces.Space] = None,
            action_space: Optional[gym.spaces.Space] = None,
            config: Optional[PartialTrainerConfigDict] = None,
            policy_mapping_fn: Optional[Callable[[AgentID, EpisodeID],
                                                 PolicyID]] = None,
            policies_to_train: Optional[List[PolicyID]] = None,
            evaluation_workers: bool = True,
    ) -> Policy:
        """Adds a new policy to this Trainer.

        Args:
            policy_id (PolicyID): ID of the policy to add.
            policy_cls (Type[Policy]): The Policy class to use for
                constructing the new Policy.
            observation_space (Optional[gym.spaces.Space]): The observation
                space of the policy to add.
            action_space (Optional[gym.spaces.Space]): The action space
                of the policy to add.
            config (Optional[PartialTrainerConfigDict]): The config overrides
                for the policy to add.
            policy_mapping_fn (Optional[Callable[[AgentID], PolicyID]]): An
                optional (updated) policy mapping function to use from here on.
                Note that already ongoing episodes will not change their
                mapping but will use the old mapping till the end of the
                episode.
            policies_to_train (Optional[List[PolicyID]]): An optional list of
                policy IDs to be trained. If None, will keep the existing list
                in place. Policies, whose IDs are not in the list will not be
                updated.
            evaluation_workers (bool): Whether to add the new policy also
                to the evaluation WorkerSet.

        Returns:
            Policy: The newly added policy (the copy that got added to the
                local worker).
        """

        def fn(worker: RolloutWorker):
            # `foreach_worker` function: Adds the policy the the worker (and
            # maybe changes its policy_mapping_fn - if provided here).
            worker.add_policy(
                policy_id=policy_id,
                policy_cls=policy_cls,
                observation_space=observation_space,
                action_space=action_space,
                config=config,
                policy_mapping_fn=policy_mapping_fn,
                policies_to_train=policies_to_train,
            )

        # Run foreach_worker fn on all workers (incl. evaluation workers).
        self.workers.foreach_worker(fn)
        if evaluation_workers and self.evaluation_workers is not None:
            self.evaluation_workers.foreach_worker(fn)

        # Return newly added policy (from the local rollout worker).
        return self.get_policy(policy_id)

    @PublicAPI
    def remove_policy(
            self,
            policy_id: PolicyID = DEFAULT_POLICY_ID,
            *,
            policy_mapping_fn: Optional[Callable[[AgentID], PolicyID]] = None,
            policies_to_train: Optional[List[PolicyID]] = None,
            evaluation_workers: bool = True,
    ) -> None:
        """Removes a new policy from this Trainer.

        Args:
            policy_id (Optional[PolicyID]): ID of the policy to be removed.
            policy_mapping_fn (Optional[Callable[[AgentID], PolicyID]]): An
                optional (updated) policy mapping function to use from here on.
                Note that already ongoing episodes will not change their
                mapping but will use the old mapping till the end of the
                episode.
            policies_to_train (Optional[List[PolicyID]]): An optional list of
                policy IDs to be trained. If None, will keep the existing list
                in place. Policies, whose IDs are not in the list will not be
                updated.
            evaluation_workers (bool): Whether to also remove the policy from
                the evaluation WorkerSet.
        """

        def fn(worker):
            worker.remove_policy(
                policy_id=policy_id,
                policy_mapping_fn=policy_mapping_fn,
                policies_to_train=policies_to_train,
            )

        self.workers.foreach_worker(fn)
        if evaluation_workers and self.evaluation_workers is not None:
            self.evaluation_workers.foreach_worker(fn)

    @DeveloperAPI
    def export_policy_model(self,
                            export_dir: str,
                            policy_id: PolicyID = DEFAULT_POLICY_ID,
                            onnx: Optional[int] = None):
        """Export policy model with given policy_id to local directory.

        Args:
            export_dir (string): Writable local directory.
            policy_id (string): Optional policy id to export.
            onnx (int): If given, will export model in ONNX format. The
                value of this parameter set the ONNX OpSet version to use.

        Example:
            >>> trainer = MyTrainer()
            >>> for _ in range(10):
            >>>     trainer.train()
            >>> trainer.export_policy_model("/tmp/export_dir")
        """
        self.workers.local_worker().export_policy_model(
            export_dir, policy_id, onnx)

    @DeveloperAPI
    def export_policy_checkpoint(self,
                                 export_dir: str,
                                 filename_prefix: str = "model",
                                 policy_id: PolicyID = DEFAULT_POLICY_ID):
        """Export tensorflow policy model checkpoint to local directory.

        Args:
            export_dir (string): Writable local directory.
            filename_prefix (string): file name prefix of checkpoint files.
            policy_id (string): Optional policy id to export.

        Example:
            >>> trainer = MyTrainer()
            >>> for _ in range(10):
            >>>     trainer.train()
            >>> trainer.export_policy_checkpoint("/tmp/export_dir")
        """
        self.workers.local_worker().export_policy_checkpoint(
            export_dir, filename_prefix, policy_id)

    @DeveloperAPI
    def import_policy_model_from_h5(self,
                                    import_file: str,
                                    policy_id: PolicyID = DEFAULT_POLICY_ID):
        """Imports a policy's model with given policy_id from a local h5 file.

        Args:
            import_file (str): The h5 file to import from.
            policy_id (string): Optional policy id to import into.

        Example:
            >>> trainer = MyTrainer()
            >>> trainer.import_policy_model_from_h5("/tmp/weights.h5")
            >>> for _ in range(10):
            >>>     trainer.train()
        """
        self.workers.local_worker().import_policy_model_from_h5(
            import_file, policy_id)

    @DeveloperAPI
    def collect_metrics(self,
                        selected_workers: List[ActorHandle] = None) -> dict:
        """Collects metrics from the remote workers of this agent.

        This is the same data as returned by a call to train().
        """
        return self.optimizer.collect_metrics(
            self.config["collect_metrics_timeout"],
            min_history=self.config["metrics_smoothing_episodes"],
            selected_workers=selected_workers)

    @classmethod
    def resource_help(cls, config: TrainerConfigDict) -> str:
        return ("\n\nYou can adjust the resource requests of RLlib agents by "
                "setting `num_workers`, `num_gpus`, and other configs. See "
                "the DEFAULT_CONFIG defined by each agent for more info.\n\n"
                "The config of this agent is: {}".format(config))

    @classmethod
    def merge_trainer_configs(cls,
                              config1: TrainerConfigDict,
                              config2: PartialTrainerConfigDict,
                              _allow_unknown_configs: Optional[bool] = None
                              ) -> TrainerConfigDict:
        config1 = copy.deepcopy(config1)
        if "callbacks" in config2 and type(config2["callbacks"]) is dict:
            legacy_callbacks_dict = config2["callbacks"]

            def make_callbacks():
                # Deprecation warning will be logged by DefaultCallbacks.
                return DefaultCallbacks(
                    legacy_callbacks_dict=legacy_callbacks_dict)

            config2["callbacks"] = make_callbacks
        if _allow_unknown_configs is None:
            _allow_unknown_configs = cls._allow_unknown_configs
        return deep_update(config1, config2, _allow_unknown_configs,
                           cls._allow_unknown_subkeys,
                           cls._override_all_subkeys_if_type_changes)

    @staticmethod
    def _validate_config(config: PartialTrainerConfigDict,
                         trainer_obj_or_none: Optional["Trainer"] = None):
        model_config = config.get("model")
        if model_config is None:
            config["model"] = model_config = {}

        # Monitor should be replaced by `record_env`.
        if config.get("monitor", DEPRECATED_VALUE) != DEPRECATED_VALUE:
            deprecation_warning("monitor", "record_env", error=False)
            config["record_env"] = config.get("monitor", False)
        # Empty string would fail some if-blocks checking for this setting.
        # Set to True instead, meaning: use default output dir to store
        # the videos.
        if config.get("record_env") == "":
            config["record_env"] = True

        # DefaultCallbacks if callbacks - for whatever reason - set to
        # None.
        if config["callbacks"] is None:
            config["callbacks"] = DefaultCallbacks

        # Multi-GPU settings.
        simple_optim_setting = config.get("simple_optimizer", DEPRECATED_VALUE)
        if simple_optim_setting != DEPRECATED_VALUE:
            deprecation_warning(old="simple_optimizer", error=False)

        # Loop through all policy definitions in multi-agent policies.
        policies, is_multi_agent = check_multi_agent(config)

        for pid, policy_spec in policies.copy().items():
            # Policy IDs must be strings.
            if not isinstance(pid, str):
                raise ValueError("Policy keys must be strs, got {}".format(
                    type(pid)))

            # Convert to PolicySpec if plain list/tuple.
            if not isinstance(policy_spec, PolicySpec):
                # Values must be lists/tuples of len 4.
                if not isinstance(policy_spec, (list, tuple)) or \
                        len(policy_spec) != 4:
                    raise ValueError(
                        "Policy specs must be tuples/lists of "
                        "(cls or None, obs_space, action_space, config), "
                        f"got {policy_spec}")
                policies[pid] = PolicySpec(*policy_spec)

            # Config is None -> Set to {}.
            if policies[pid].config is None:
                policies[pid] = policies[pid]._replace(config={})
            # Config not a dict.
            elif not isinstance(policies[pid].config, dict):
                raise ValueError(
                    f"Multiagent policy config for {pid} must be a dict, "
                    f"but got {type(policies[pid].config)}!")

        framework = config.get("framework")
        # Multi-GPU setting: Must use MultiGPUTrainOneStep.
        if config.get("num_gpus", 0) > 1:
            if framework in ["tfe", "tf2"]:
                raise ValueError("`num_gpus` > 1 not supported yet for "
                                 "framework={}!".format(framework))
            elif simple_optim_setting is True:
                raise ValueError(
                    "Cannot use `simple_optimizer` if `num_gpus` > 1! "
                    "Consider not setting `simple_optimizer` in your config.")
            config["simple_optimizer"] = False
        # Auto-setting: Use simple-optimizer for tf-eager or multiagent,
        # otherwise: MultiGPUTrainOneStep (if supported by the algo's execution
        # plan).
        elif simple_optim_setting == DEPRECATED_VALUE:
            # tf-eager: Must use simple optimizer.
            if framework not in ["tf", "torch"]:
                config["simple_optimizer"] = True
            # Multi-agent case: Try using MultiGPU optimizer (only
            # if all policies used are DynamicTFPolicies or TorchPolicies).
            elif is_multi_agent:
                from ray.rllib.policy.dynamic_tf_policy import DynamicTFPolicy
                from ray.rllib.policy.torch_policy import TorchPolicy
                default_policy_cls = None if trainer_obj_or_none is None else \
                    getattr(trainer_obj_or_none, "_policy_class", None)
                if any((p[0] or default_policy_cls) is None
                       or not issubclass(p[0] or default_policy_cls,
                                         (DynamicTFPolicy, TorchPolicy))
                       for p in config["multiagent"]["policies"].values()):
                    config["simple_optimizer"] = True
                else:
                    config["simple_optimizer"] = False
            else:
                config["simple_optimizer"] = False

        # User manually set simple-optimizer to False -> Error if tf-eager.
        elif simple_optim_setting is False:
            if framework in ["tfe", "tf2"]:
                raise ValueError("`simple_optimizer=False` not supported for "
                                 "framework={}!".format(framework))

        # Offline RL settings.
        if isinstance(config["input_evaluation"], tuple):
            config["input_evaluation"] = list(config["input_evaluation"])
        elif not isinstance(config["input_evaluation"], list):
            raise ValueError(
                "`input_evaluation` must be a list of strings, got {}!".format(
                    config["input_evaluation"]))

        # Check model config.
        # If no preprocessing, propagate into model's config as well
        # (so model will know, whether inputs are preprocessed or not).
        if config["_disable_preprocessor_api"] is True:
            model_config["_disable_preprocessor_api"] = True

        # Prev_a/r settings.
        prev_a_r = model_config.get("lstm_use_prev_action_reward",
                                    DEPRECATED_VALUE)
        if prev_a_r != DEPRECATED_VALUE:
            deprecation_warning(
                "model.lstm_use_prev_action_reward",
                "model.lstm_use_prev_action and model.lstm_use_prev_reward",
                error=False)
            model_config["lstm_use_prev_action"] = prev_a_r
            model_config["lstm_use_prev_reward"] = prev_a_r

        # Check batching/sample collection settings.
        if config["batch_mode"] not in [
                "truncate_episodes", "complete_episodes"
        ]:
            raise ValueError("`batch_mode` must be one of [truncate_episodes|"
                             "complete_episodes]! Got {}".format(
                                 config["batch_mode"]))

        # Check multi-agent batch count mode.
        if config["multiagent"].get("count_steps_by", "env_steps") not in \
                ["env_steps", "agent_steps"]:
            raise ValueError(
                "`count_steps_by` must be one of [env_steps|agent_steps]! "
                "Got {}".format(config["multiagent"]["count_steps_by"]))

        # Evaluation settings.
        # If `evaluation_num_workers` > 0, warn if `evaluation_interval` is
        # None (also set `evaluation_interval` to 1).
        if config["evaluation_num_workers"] > 0 and \
                not config["evaluation_interval"]:
            logger.warning(
                "You have specified {} evaluation workers, but no evaluation "
                "interval! Will set the interval to 1 (each `train()` call). "
                "If this is too frequent, set `evaluation_interval` to some "
                "larger value.".format(config["evaluation_num_workers"]))
            config["evaluation_interval"] = 1
        # If `evaluation_num_workers=0` and
        # `evaluation_parallel_to_training=True`, warn that you need
        # at least one remote eval worker for parallel training and
        # evaluation, and set `evaluation_parallel_to_training` to False.
        elif config["evaluation_num_workers"] == 0 and \
                config.get("evaluation_parallel_to_training", False):
            logger.warning(
                "`evaluation_parallel_to_training` can only be done if "
                "`evaluation_num_workers` > 0! Setting "
                "`evaluation_parallel_to_training` to False.")
            config["evaluation_parallel_to_training"] = False

        # If `evaluation_num_episodes=auto`, error if
        # `evaluation_parallel_to_training=False`.
        if config["evaluation_num_episodes"] == "auto":
            if not config["evaluation_parallel_to_training"]:
                raise ValueError(
                    "`evaluation_num_episodes=auto` not supported for "
                    "`evaluation_parallel_to_training=False`!")
        # Make sure, it's an int otherwise.
        elif not isinstance(config["evaluation_num_episodes"], int):
            raise ValueError(
                "`evaluation_num_episodes` ({}) must be an int and "
                ">0!".format(config["evaluation_num_episodes"]))

    def _try_recover(self):
        """Try to identify and remove any unhealthy workers.

        This method is called after an unexpected remote error is encountered
        from a worker. It issues check requests to all current workers and
        removes any that respond with error. If no healthy workers remain,
        an error is raised.
        """

        assert hasattr(self, "execution_plan")
        workers = self.workers

        logger.info("Health checking all workers...")
        checks = []
        for ev in workers.remote_workers():
            _, obj_ref = ev.sample_with_count.remote()
            checks.append(obj_ref)

        healthy_workers = []
        for i, obj_ref in enumerate(checks):
            w = workers.remote_workers()[i]
            try:
                ray.get(obj_ref)
                healthy_workers.append(w)
                logger.info("Worker {} looks healthy".format(i + 1))
            except RayError:
                logger.exception("Removing unhealthy worker {}".format(i + 1))
                try:
                    w.__ray_terminate__.remote()
                except Exception:
                    logger.exception("Error terminating unhealthy worker")

        if len(healthy_workers) < 1:
            raise RuntimeError(
                "Not enough healthy workers remain to continue.")

        logger.warning("Recreating execution plan after failure")
        workers.reset(healthy_workers)
        self.train_exec_impl = self.execution_plan(workers, self.config)

    @override(Trainable)
    def _export_model(self, export_formats: List[str],
                      export_dir: str) -> Dict[str, str]:
        ExportFormat.validate(export_formats)
        exported = {}
        if ExportFormat.CHECKPOINT in export_formats:
            path = os.path.join(export_dir, ExportFormat.CHECKPOINT)
            self.export_policy_checkpoint(path)
            exported[ExportFormat.CHECKPOINT] = path
        if ExportFormat.MODEL in export_formats:
            path = os.path.join(export_dir, ExportFormat.MODEL)
            self.export_policy_model(path)
            exported[ExportFormat.MODEL] = path
        if ExportFormat.ONNX in export_formats:
            path = os.path.join(export_dir, ExportFormat.ONNX)
            self.export_policy_model(
                path, onnx=int(os.getenv("ONNX_OPSET", "11")))
            exported[ExportFormat.ONNX] = path
        return exported

    def import_model(self, import_file: str):
        """Imports a model from import_file.

        Note: Currently, only h5 files are supported.

        Args:
            import_file (str): The file to import the model from.

        Returns:
            A dict that maps ExportFormats to successfully exported models.
        """
        # Check for existence.
        if not os.path.exists(import_file):
            raise FileNotFoundError(
                "`import_file` '{}' does not exist! Can't import Model.".
                format(import_file))
        # Get the format of the given file.
        import_format = "h5"  # TODO(sven): Support checkpoint loading.

        ExportFormat.validate([import_format])
        if import_format != ExportFormat.H5:
            raise NotImplementedError
        else:
            return self.import_policy_model_from_h5(import_file)

    def __getstate__(self) -> dict:
        state = {}
        if hasattr(self, "workers"):
            state["worker"] = self.workers.local_worker().save()
        if hasattr(self, "optimizer") and hasattr(self.optimizer, "save"):
            state["optimizer"] = self.optimizer.save()
        # TODO: Experimental functionality: Store contents of replay buffer
        #  to checkpoint, only if user has configured this.
        if self.local_replay_buffer is not None and \
                self.config.get("store_buffer_in_checkpoints"):
            state["local_replay_buffer"] = \
                self.local_replay_buffer.get_state()
        return state

    def __setstate__(self, state: dict):
        if "worker" in state and hasattr(self, "workers"):
            self.workers.local_worker().restore(state["worker"])
            remote_state = ray.put(state["worker"])
            for r in self.workers.remote_workers():
                r.restore.remote(remote_state)
        # Restore optimizer data, if necessary.
        if "optimizer" in state and hasattr(self, "optimizer"):
            self.optimizer.restore(state["optimizer"])
        # If necessary, restore replay data as well.
        if self.local_replay_buffer is not None:
            # TODO: Experimental functionality: Restore contents of replay
            #  buffer from checkpoint, only if user has configured this.
            if self.config.get("store_buffer_in_checkpoints"):
                if "local_replay_buffer" in state:
                    self.local_replay_buffer.set_state(
                        state["local_replay_buffer"])
                else:
                    logger.warning(
                        "`store_buffer_in_checkpoints` is True, but no replay "
                        "data found in state!")
            elif "local_replay_buffer" in state and \
                    log_once("no_store_buffer_in_checkpoints_but_data_found"):
                logger.warning(
                    "`store_buffer_in_checkpoints` is False, but some replay "
                    "data found in state!")

    @staticmethod
    def with_updates(**overrides) -> Type["Trainer"]:
        raise NotImplementedError(
            "`with_updates` may only be called on Trainer sub-classes "
            "that were generated via the `ray.rllib.agents.trainer_template."
            "build_trainer()` function!")

    def _register_if_needed(self, env_object: Union[str, EnvType, None],
                            config):
        if isinstance(env_object, str):
            return env_object
        elif isinstance(env_object, type):
            name = env_object.__name__

            # Add convenience `_get_spaces` method.

            def _get_spaces(s):
                return s.observation_space, s.action_space

            env_object._get_spaces = _get_spaces

            if config.get("remote_worker_envs"):
                register_env(
                    name,
                    lambda cfg: ray.remote(num_cpus=0)(env_object).remote(cfg))
            else:
                register_env(name, lambda cfg: env_object(cfg))
            return name
        elif env_object is None:
            return None
        raise ValueError(
            "{} is an invalid env specification. ".format(env_object) +
            "You can specify a custom env as either a class "
            "(e.g., YourEnvCls) or a registered env id (e.g., \"your_env\").")

    def __repr__(self):
        return self._name