Source code for nnabla_rl.algorithms.categorical_dqn

# Copyright 2020,2021 Sony Corporation.
# Copyright 2021,2022,2023,2024 Sony Group Corporation.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from dataclasses import dataclass
from typing import Any, Dict, Union

import gym
import numpy as np

import nnabla as nn
import nnabla.solvers as NS
import nnabla_rl.environment_explorers as EE
import nnabla_rl.model_trainers as MT
from nnabla_rl.algorithm import Algorithm, AlgorithmConfig, eval_api
from nnabla_rl.algorithms.common_utils import _GreedyActionSelector
from nnabla_rl.builders import ExplorerBuilder, ModelBuilder, ReplayBufferBuilder, SolverBuilder
from nnabla_rl.environment_explorer import EnvironmentExplorer
from nnabla_rl.environment_explorers.epsilon_greedy_explorer import epsilon_greedy_action_selection
from nnabla_rl.environments.environment_info import EnvironmentInfo
from nnabla_rl.model_trainers.model_trainer import ModelTrainer, TrainingBatch
from nnabla_rl.models import C51ValueDistributionFunction, ValueDistributionFunction
from nnabla_rl.replay_buffer import ReplayBuffer
from nnabla_rl.utils import context
from nnabla_rl.utils.data import marshal_experiences
from nnabla_rl.utils.misc import sync_model


[docs]@dataclass class CategoricalDQNConfig(AlgorithmConfig): """CategoricalDQNConfig List of configurations for CategoricalDQN algorithm. Args: gamma (float): discount factor of rewards. Defaults to 0.99. learning_rate (float): learning rate which is set to all solvers. \ You can customize/override the learning rate for each solver by implementing the \ (:py:class:`SolverBuilder <nnabla_rl.builders.SolverBuilder>`) by yourself. \ Defaults to 0.001. batch_size (int): training batch size. Defaults to 32. num_steps (int): number of steps for N-step Q targets. Defaults to 1. start_timesteps (int): the timestep when training starts.\ The algorithm will collect experiences from the environment by acting randomly until this timestep. Defaults to 50000. replay_buffer_size (int): the capacity of replay buffer. Defaults to 1000000. learner_update_frequency (float): the interval of learner update. Defaults to 4 target_update_frequency (float): the interval of target q-function update. Defaults to 10000. max_explore_steps (int): the number of steps decaying the epsilon value.\ The epsilon will be decayed linearly \ :math:`\\epsilon=\\epsilon_{init} - step\\times\\frac{\\epsilon_{init} - \ \\epsilon_{final}}{max\\_explore\\_steps}`.\ Defaults to 1000000. initial_epsilon (float): the initial epsilon value for ε-greedy explorer. Defaults to 1.0. final_epsilon (float): the last epsilon value for ε-greedy explorer. Defaults to 0.01. test_epsilon (float): the epsilon value on testing. Defaults to 0.001. v_min (float): lower limit of the value used in value distribution function. Defaults to -10.0. v_max (float): upper limit of the value used in value distribution function. Defaults to 10.0. num_atoms (int): the number of bins used in value distribution function. Defaults to 51. loss_reduction_method (str): KL loss reduction method. "sum" or "mean" is supported. Defaults to mean. unroll_steps (int): Number of steps to unroll tranining network. The network will be unrolled even though the provided model doesn't have RNN layers. Defaults to 1. burn_in_steps (int): Number of burn-in steps to initiaze recurrent layer states during training. This flag does not take effect if given model is not an RNN model. Defaults to 0. reset_rnn_on_terminal (bool): Reset recurrent internal states to zero during training if episode ends. This flag does not take effect if given model is not an RNN model. Defaults to True. """ gamma: float = 0.99 learning_rate: float = 0.00025 batch_size: int = 32 num_steps: int = 1 start_timesteps: int = 50000 replay_buffer_size: int = 1000000 learner_update_frequency: int = 4 target_update_frequency: int = 10000 max_explore_steps: int = 1000000 initial_epsilon: float = 1.0 final_epsilon: float = 0.01 test_epsilon: float = 0.001 v_min: float = -10.0 v_max: float = 10.0 num_atoms: int = 51 loss_reduction_method: str = "mean" # rnn model support unroll_steps: int = 1 burn_in_steps: int = 0 reset_rnn_on_terminal: bool = True def __post_init__(self): """__post_init__ Check set values are in valid range. """ self._assert_between(self.gamma, 0.0, 1.0, "gamma") self._assert_positive(self.learning_rate, "learning_rate") self._assert_positive(self.batch_size, "batch_size") self._assert_positive(self.num_steps, "num_steps") self._assert_positive(self.learner_update_frequency, "learner_update_frequency") self._assert_positive(self.target_update_frequency, "target_update_frequency") self._assert_positive(self.start_timesteps, "start_timesteps") self._assert_positive(self.replay_buffer_size, "replay_buffer_size") self._assert_smaller_than(self.start_timesteps, self.replay_buffer_size, "start_timesteps") self._assert_positive(self.max_explore_steps, "max_explore_steps") self._assert_between(self.initial_epsilon, 0.0, 1.0, "initial_epsilon") self._assert_between(self.final_epsilon, 0.0, 1.0, "final_epsilon") self._assert_between(self.test_epsilon, 0.0, 1.0, "test_epsilon") self._assert_positive(self.num_atoms, "num_atoms") self._assert_positive(self.unroll_steps, "unroll_steps") self._assert_positive_or_zero(self.burn_in_steps, "burn_in_steps")
class DefaultValueDistFunctionBuilder(ModelBuilder[ValueDistributionFunction]): def build_model( # type: ignore[override] self, scope_name: str, env_info: EnvironmentInfo, algorithm_config: CategoricalDQNConfig, **kwargs, ) -> ValueDistributionFunction: return C51ValueDistributionFunction( scope_name, env_info.action_dim, algorithm_config.num_atoms, algorithm_config.v_min, algorithm_config.v_max ) class DefaultReplayBufferBuilder(ReplayBufferBuilder): def build_replay_buffer( # type: ignore[override] self, env_info: EnvironmentInfo, algorithm_config: CategoricalDQNConfig, **kwargs ) -> ReplayBuffer: return ReplayBuffer(capacity=algorithm_config.replay_buffer_size) class DefaultSolverBuilder(SolverBuilder): def build_solver( # type: ignore[override] self, env_info: EnvironmentInfo, algorithm_config: CategoricalDQNConfig, **kwargs ) -> nn.solver.Solver: return NS.Adam(alpha=algorithm_config.learning_rate, eps=1e-2 / algorithm_config.batch_size) class DefaultExplorerBuilder(ExplorerBuilder): def build_explorer( # type: ignore[override] self, env_info: EnvironmentInfo, algorithm_config: CategoricalDQNConfig, algorithm: "CategoricalDQN", **kwargs, ) -> EnvironmentExplorer: explorer_config = EE.LinearDecayEpsilonGreedyExplorerConfig( warmup_random_steps=algorithm_config.start_timesteps, initial_step_num=algorithm.iteration_num, initial_epsilon=algorithm_config.initial_epsilon, final_epsilon=algorithm_config.final_epsilon, max_explore_steps=algorithm_config.max_explore_steps, ) explorer = EE.LinearDecayEpsilonGreedyExplorer( greedy_action_selector=algorithm._exploration_action_selector, random_action_selector=algorithm._random_action_selector, env_info=env_info, config=explorer_config, ) return explorer
[docs]class CategoricalDQN(Algorithm): """Categorical DQN algorithm. This class implements the Categorical DQN algorithm proposed by M. Bellemare, et al. in the paper: "A Distributional Perspective on Reinfocement Learning" For details see: https://arxiv.org/abs/1707.06887 Args: env_or_env_info \ (gym.Env or :py:class:`EnvironmentInfo <nnabla_rl.environments.environment_info.EnvironmentInfo>`): the environment to train or environment info config (:py:class:`CategoricalDQNConfig <nnabla_rl.algorithms.categorical_dqn.CategoricalDQNConfig>`): configuration of the CategoricalDQN algorithm value_distribution_builder (:py:class:`ModelBuilder[ValueDistributionFunctionFunction] \ <nnabla_rl.builders.ModelBuilder>`): builder of value distribution function models value_distribution_solver_builder (:py:class:`SolverBuilder <nnabla_rl.builders.SolverBuilder>`): builder of value distribution function solvers replay_buffer_builder (:py:class:`ReplayBufferBuilder <nnabla_rl.builders.ReplayBufferBuilder>`): builder of replay_buffer explorer_builder (:py:class:`ExplorerBuilder <nnabla_rl.builders.ExplorerBuilder>`): builder of environment explorer """ # type declarations to type check with mypy # NOTE: declared variables are instance variable and NOT class variable, unless it is marked with ClassVar # See https://mypy.readthedocs.io/en/stable/class_basics.html for details _config: CategoricalDQNConfig _atom_p: ValueDistributionFunction _atom_p_solver: nn.solver.Solver _target_atom_p: ValueDistributionFunction _replay_buffer: ReplayBuffer _explorer_builder: ExplorerBuilder _environment_explorer: EnvironmentExplorer _model_trainer: ModelTrainer _evaluation_actor: _GreedyActionSelector _exploration_actor: _GreedyActionSelector _model_trainer_state: Dict[str, Any] def __init__( self, env_or_env_info: Union[gym.Env, EnvironmentInfo], config: CategoricalDQNConfig = CategoricalDQNConfig(), value_distribution_builder: ModelBuilder[ValueDistributionFunction] = DefaultValueDistFunctionBuilder(), value_distribution_solver_builder: SolverBuilder = DefaultSolverBuilder(), replay_buffer_builder: ReplayBufferBuilder = DefaultReplayBufferBuilder(), explorer_builder: ExplorerBuilder = DefaultExplorerBuilder(), ): super(CategoricalDQN, self).__init__(env_or_env_info, config=config) self._explorer_builder = explorer_builder with nn.context_scope(context.get_nnabla_context(self._config.gpu_id)): self._atom_p = value_distribution_builder("atom_p_train", self._env_info, self._config) self._atom_p_solver = value_distribution_solver_builder(self._env_info, self._config) self._target_atom_p = self._atom_p.deepcopy("target_atom_p_train") self._replay_buffer = replay_buffer_builder(self._env_info, self._config) self._evaluation_actor = _GreedyActionSelector(self._env_info, self._atom_p.shallowcopy().as_q_function()) self._exploration_actor = _GreedyActionSelector(self._env_info, self._atom_p.shallowcopy().as_q_function()) @eval_api def compute_eval_action(self, state, *, begin_of_episode=False, extra_info={}): with nn.context_scope(context.get_nnabla_context(self._config.gpu_id)): (action, _), _ = epsilon_greedy_action_selection( state, self._evaluation_action_selector, self._random_action_selector, epsilon=self._config.test_epsilon, begin_of_episode=begin_of_episode, ) return action def _before_training_start(self, env_or_buffer): # set context globally to ensure that the training runs on configured gpu context.set_nnabla_context(self._config.gpu_id) self._environment_explorer = self._setup_environment_explorer(env_or_buffer) self._model_trainer = self._setup_value_distribution_function_training(env_or_buffer) def _setup_environment_explorer(self, env_or_buffer): return None if self._is_buffer(env_or_buffer) else self._explorer_builder(self._env_info, self._config, self) def _setup_value_distribution_function_training(self, env_or_buffer): trainer_config = MT.q_value_trainers.CategoricalDQNQTrainerConfig( num_steps=self._config.num_steps, v_min=self._config.v_min, v_max=self._config.v_max, num_atoms=self._config.num_atoms, reduction_method=self._config.loss_reduction_method, unroll_steps=self._config.unroll_steps, burn_in_steps=self._config.burn_in_steps, reset_on_terminal=self._config.reset_rnn_on_terminal, ) model_trainer = MT.q_value_trainers.CategoricalDQNQTrainer( train_functions=self._atom_p, solvers={self._atom_p.scope_name: self._atom_p_solver}, target_function=self._target_atom_p, env_info=self._env_info, config=trainer_config, ) # NOTE: Copy initial parameters after setting up the training # Because the parameter is created after training graph construction sync_model(self._atom_p, self._target_atom_p) return model_trainer def _run_online_training_iteration(self, env): experiences = self._environment_explorer.step(env) self._replay_buffer.append_all(experiences) if self._config.start_timesteps < self.iteration_num: if self.iteration_num % self._config.learner_update_frequency == 0: self._categorical_dqn_training(self._replay_buffer) def _run_offline_training_iteration(self, buffer): self._categorical_dqn_training(buffer) def _categorical_dqn_training(self, replay_buffer): num_steps = self._config.num_steps + self._config.burn_in_steps + self._config.unroll_steps - 1 experiences_tuple, info = replay_buffer.sample(self._config.batch_size, num_steps=num_steps) if num_steps == 1: experiences_tuple = (experiences_tuple,) assert len(experiences_tuple) == num_steps batch = None for experiences in reversed(experiences_tuple): (s, a, r, non_terminal, s_next, rnn_states_dict, *_) = marshal_experiences(experiences) rnn_states = rnn_states_dict["rnn_states"] if "rnn_states" in rnn_states_dict else {} batch = TrainingBatch( batch_size=self._config.batch_size, s_current=s, a_current=a, gamma=self._config.gamma, reward=r, non_terminal=non_terminal, s_next=s_next, weight=info["weights"], next_step_batch=batch, rnn_states=rnn_states, ) self._model_trainer_state = self._model_trainer.train(batch) if self.iteration_num % self._config.target_update_frequency == 0: sync_model(self._atom_p, self._target_atom_p) td_errors = self._model_trainer_state["td_errors"] replay_buffer.update_priorities(td_errors) def _evaluation_action_selector(self, s, *, begin_of_episode=False): return self._evaluation_actor(s, begin_of_episode=begin_of_episode) def _exploration_action_selector(self, s, *, begin_of_episode=False): return self._exploration_actor(s, begin_of_episode=begin_of_episode) def _random_action_selector(self, s, *, begin_of_episode=False): action = self._env_info.action_space.sample() return np.asarray(action).reshape((1,)), {} def _models(self): models = {} models[self._atom_p.scope_name] = self._atom_p return models def _solvers(self): solvers = {} solvers[self._atom_p.scope_name] = self._atom_p_solver return solvers
[docs] @classmethod def is_supported_env(cls, env_or_env_info): env_info = ( EnvironmentInfo.from_env(env_or_env_info) if isinstance(env_or_env_info, gym.Env) else env_or_env_info ) return not env_info.is_continuous_action_env() and not env_info.is_tuple_action_env()
[docs] @classmethod def is_rnn_supported(self): return True
@property def latest_iteration_state(self): latest_iteration_state = super(CategoricalDQN, self).latest_iteration_state if hasattr(self, "_model_trainer_state"): latest_iteration_state["scalar"].update( {"cross_entropy_loss": float(self._model_trainer_state["cross_entropy_loss"])} ) latest_iteration_state["histogram"].update({"td_errors": self._model_trainer_state["td_errors"].flatten()}) return latest_iteration_state @property def trainers(self): return {"q_function": self._model_trainer}