stable-baselines3/stable_baselines3/common/buffers.py

384 lines
16 KiB
Python
Raw Normal View History

Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
import warnings
from typing import Generator, Optional, Union
2020-02-03 14:40:34 +00:00
2019-09-05 15:29:41 +00:00
import numpy as np
import torch as th
from gym import spaces
2019-09-05 15:29:41 +00:00
Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
try:
# Check memory used by replay buffer when possible
import psutil
except ImportError:
psutil = None
2020-05-05 13:02:35 +00:00
from stable_baselines3.common.preprocessing import get_action_dim, get_obs_shape
from stable_baselines3.common.type_aliases import ReplayBufferSamples, RolloutBufferSamples
from stable_baselines3.common.vec_env import VecNormalize
2020-02-03 14:40:34 +00:00
2019-09-18 13:35:17 +00:00
2019-09-19 09:43:15 +00:00
class BaseBuffer(object):
2019-11-22 12:33:12 +00:00
"""
Base class that represent a buffer (rollout or replay)
:param buffer_size: (int) Max number of element in the buffer
:param observation_space: (spaces.Space) Observation space
:param action_space: (spaces.Space) Action space
2020-02-03 14:40:34 +00:00
:param device: (Union[th.device, str]) PyTorch device
to which the values will be converted
2019-11-22 12:33:12 +00:00
:param n_envs: (int) Number of parallel environments
"""
def __init__(
self,
buffer_size: int,
observation_space: spaces.Space,
action_space: spaces.Space,
device: Union[th.device, str] = "cpu",
n_envs: int = 1,
):
2019-09-19 09:43:15 +00:00
super(BaseBuffer, self).__init__()
2019-09-05 15:29:41 +00:00
self.buffer_size = buffer_size
self.observation_space = observation_space
self.action_space = action_space
2020-03-20 10:20:57 +00:00
self.obs_shape = get_obs_shape(observation_space)
self.action_dim = get_action_dim(action_space)
2019-09-05 15:29:41 +00:00
self.pos = 0
self.full = False
self.device = device
2019-09-20 13:19:04 +00:00
self.n_envs = n_envs
@staticmethod
2020-02-03 14:40:34 +00:00
def swap_and_flatten(arr: np.ndarray) -> np.ndarray:
2019-09-20 13:19:04 +00:00
"""
Swap and then flatten axes 0 (buffer_size) and 1 (n_envs)
to convert shape from [n_steps, n_envs, ...] (when ... is the shape of the features)
to [n_steps * n_envs, ...] (which maintain the order)
2020-02-03 14:40:34 +00:00
:param arr: (np.ndarray)
:return: (np.ndarray)
2019-09-20 13:19:04 +00:00
"""
2020-02-03 14:40:34 +00:00
shape = arr.shape
2019-09-20 13:19:04 +00:00
if len(shape) < 3:
shape = shape + (1,)
2020-02-03 14:40:34 +00:00
return arr.swapaxes(0, 1).reshape(shape[0] * shape[1], *shape[2:])
2019-09-05 15:29:41 +00:00
2020-02-03 14:40:34 +00:00
def size(self) -> int:
2019-11-22 12:33:12 +00:00
"""
:return: (int) The current size of the buffer
"""
2019-09-05 15:29:41 +00:00
if self.full:
return self.buffer_size
return self.pos
2020-02-03 14:40:34 +00:00
def add(self, *args, **kwargs) -> None:
2019-11-22 12:33:12 +00:00
"""
Add elements to the buffer.
"""
2019-09-19 09:43:15 +00:00
raise NotImplementedError()
2020-02-11 15:40:44 +00:00
def extend(self, *args, **kwargs) -> None:
"""
Add a new batch of transitions to the buffer
"""
# Do a for loop along the batch axis
for data in zip(*args):
self.add(*data)
2020-02-03 14:40:34 +00:00
def reset(self) -> None:
2019-11-22 12:33:12 +00:00
"""
Reset the buffer.
"""
2019-09-19 09:43:15 +00:00
self.pos = 0
self.full = False
def sample(self, batch_size: int, env: Optional[VecNormalize] = None):
2019-11-22 12:33:12 +00:00
"""
:param batch_size: (int) Number of element to sample
2020-02-03 14:40:34 +00:00
:param env: (Optional[VecNormalize]) associated gym VecEnv
2019-11-22 12:33:12 +00:00
to normalize the observations/rewards when sampling
2020-03-10 15:43:10 +00:00
:return: (Union[RolloutBufferSamples, ReplayBufferSamples])
2019-11-22 12:33:12 +00:00
"""
2019-09-19 09:43:15 +00:00
upper_bound = self.buffer_size if self.full else self.pos
2020-02-03 14:40:34 +00:00
batch_inds = np.random.randint(0, upper_bound, size=batch_size)
2019-11-14 13:35:00 +00:00
return self._get_samples(batch_inds, env=env)
2019-09-19 09:43:15 +00:00
def _get_samples(self, batch_inds: np.ndarray, env: Optional[VecNormalize] = None):
2019-11-22 12:33:12 +00:00
"""
:param batch_inds: (th.Tensor)
2020-02-03 14:40:34 +00:00
:param env: (Optional[VecNormalize])
2020-03-10 15:43:10 +00:00
:return: (Union[RolloutBufferSamples, ReplayBufferSamples])
2019-11-22 12:33:12 +00:00
"""
2019-09-19 09:43:15 +00:00
raise NotImplementedError()
2020-02-03 14:42:14 +00:00
def to_torch(self, array: np.ndarray, copy: bool = True) -> th.Tensor:
2020-02-03 14:40:34 +00:00
"""
Convert a numpy array to a PyTorch tensor.
2020-02-03 14:42:14 +00:00
Note: it copies the data by default
2020-02-03 14:40:34 +00:00
:param array: (np.ndarray)
:param copy: (bool) Whether to copy or not the data
(may be useful to avoid changing things be reference)
:return: (th.Tensor)
"""
if copy:
return th.tensor(array).to(self.device)
return th.as_tensor(array).to(self.device)
2020-01-20 10:17:55 +00:00
@staticmethod
def _normalize_obs(obs: np.ndarray, env: Optional[VecNormalize] = None) -> np.ndarray:
2019-11-14 13:35:00 +00:00
if env is not None:
2020-02-03 14:40:34 +00:00
return env.normalize_obs(obs).astype(np.float32)
2019-11-14 13:35:00 +00:00
return obs
2020-03-12 10:12:10 +00:00
@staticmethod
def _normalize_reward(reward: np.ndarray, env: Optional[VecNormalize] = None) -> np.ndarray:
2019-11-14 13:35:00 +00:00
if env is not None:
2020-02-03 14:40:34 +00:00
return env.normalize_reward(reward).astype(np.float32)
2019-11-14 13:35:00 +00:00
return reward
2019-09-21 15:17:09 +00:00
2019-09-19 09:43:15 +00:00
class ReplayBuffer(BaseBuffer):
"""
2019-11-22 12:33:12 +00:00
Replay buffer used in off-policy algorithms like SAC/TD3.
:param buffer_size: (int) Max number of element in the buffer
:param observation_space: (spaces.Space) Observation space
:param action_space: (spaces.Space) Action space
2019-11-22 12:33:12 +00:00
:param device: (th.device)
:param n_envs: (int) Number of parallel environments
Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
:param optimize_memory_usage: (bool) Enable a memory efficient variant
of the replay buffer which reduces by almost a factor two the memory used,
at a cost of more complexity.
See https://github.com/DLR-RM/stable-baselines3/issues/37#issuecomment-637501195
and https://github.com/DLR-RM/stable-baselines3/pull/28#issuecomment-637559274
2019-09-19 09:43:15 +00:00
"""
def __init__(
self,
buffer_size: int,
observation_space: spaces.Space,
action_space: spaces.Space,
device: Union[th.device, str] = "cpu",
n_envs: int = 1,
optimize_memory_usage: bool = False,
):
super(ReplayBuffer, self).__init__(buffer_size, observation_space, action_space, device, n_envs=n_envs)
2019-09-19 09:43:15 +00:00
2020-02-03 14:40:34 +00:00
assert n_envs == 1, "Replay buffer only support single environment for now"
2019-09-05 15:29:41 +00:00
Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
# Check that the replay buffer can fit into the memory
if psutil is not None:
mem_available = psutil.virtual_memory().available
self.optimize_memory_usage = optimize_memory_usage
self.observations = np.zeros((self.buffer_size, self.n_envs) + self.obs_shape, dtype=observation_space.dtype)
Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
if optimize_memory_usage:
# `observations` contains also the next observation
self.next_observations = None
else:
self.next_observations = np.zeros((self.buffer_size, self.n_envs) + self.obs_shape, dtype=observation_space.dtype)
Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
self.actions = np.zeros((self.buffer_size, self.n_envs, self.action_dim), dtype=action_space.dtype)
2020-02-03 14:40:34 +00:00
self.rewards = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
self.dones = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
if psutil is not None:
total_memory_usage = self.observations.nbytes + self.actions.nbytes + self.rewards.nbytes + self.dones.nbytes
Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
if self.next_observations is not None:
total_memory_usage += self.next_observations.nbytes
if total_memory_usage > mem_available:
# Convert to GB
total_memory_usage /= 1e9
mem_available /= 1e9
warnings.warn(
"This system does not have apparently enough memory to store the complete "
f"replay buffer {total_memory_usage:.2f}GB > {mem_available:.2f}GB"
)
def add(self, obs: np.ndarray, next_obs: np.ndarray, action: np.ndarray, reward: np.ndarray, done: np.ndarray) -> None:
# Copy to avoid modification by reference
2020-02-03 14:40:34 +00:00
self.observations[self.pos] = np.array(obs).copy()
Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
if self.optimize_memory_usage:
self.observations[(self.pos + 1) % self.buffer_size] = np.array(next_obs).copy()
else:
self.next_observations[self.pos] = np.array(next_obs).copy()
2020-02-03 14:40:34 +00:00
self.actions[self.pos] = np.array(action).copy()
self.rewards[self.pos] = np.array(reward).copy()
self.dones[self.pos] = np.array(done).copy()
2019-09-05 15:29:41 +00:00
self.pos += 1
if self.pos == self.buffer_size:
self.full = True
self.pos = 0
def sample(self, batch_size: int, env: Optional[VecNormalize] = None) -> ReplayBufferSamples:
Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
"""
Sample elements from the replay buffer.
Custom sampling when using memory efficient variant,
as we should not sample the element with index `self.pos`
See https://github.com/DLR-RM/stable-baselines3/pull/28#issuecomment-637559274
:param batch_size: (int) Number of element to sample
:param env: (Optional[VecNormalize]) associated gym VecEnv
to normalize the observations/rewards when sampling
:return: (Union[RolloutBufferSamples, ReplayBufferSamples])
"""
if not self.optimize_memory_usage:
return super().sample(batch_size=batch_size, env=env)
# Do not sample the element with index `self.pos` as the transitions is invalid
# (we use only one array to store `obs` and `next_obs`)
if self.full:
batch_inds = (np.random.randint(1, self.buffer_size, size=batch_size) + self.pos) % self.buffer_size
else:
batch_inds = np.random.randint(0, self.pos, size=batch_size)
return self._get_samples(batch_inds, env=env)
def _get_samples(self, batch_inds: np.ndarray, env: Optional[VecNormalize] = None) -> ReplayBufferSamples:
Implement DQN (#28) * Created DQN template according to the paper. Next steps: - Create Policy - Complete Training - Debug * Changed Base Class * refactor save, to be consistence with overriding the excluded_save_params function. Do not try to exclude the parameters twice. * Added simple DQN policy * Finished learn and train function - missing correct loss computation * changed collect_rollouts to work with discrete space * moved discrete space collect_rollouts to dqn * basic dqn working * deleted SDE related code * added gradient clipping and moved greedy policy to policy * changed policy to implement target network and added soft update(in fact standart tau is 1 so hard update) * fixed policy setup * rebase target_update_intervall on _n_updates * adapted all tests all tests passing * Move to stable-baseline3 * Fixes for DQN * Fix tests + add CNNPolicy * Allow any optimizer for DQN * added some util functions to create a arbitrary linear schedule, fixed pickle problem with old exploration schedule * more documentation * changed buffer dtype * refactor and document * Added Sphinx Documentation Updated changelog.rst * removed custom collect_rollouts as it is no longer necessary * Implemented suggestions to clean code and documentation. * extracted some functions on tests to reduce duplicated code * added support for exploration_fraction * Fixed exploration_fraction * Added documentation * Fixed get_linear_fn -> proper progress scaling * Merged master * Added nature reference * Changed default parameters to https://www.nature.com/articles/nature14236/tables/1 * Fixed n_updates to be incremented correctly * Correct train_freq * Doc update * added special parameter for DQN in tests * different fix for test_discrete * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Update docs/modules/dqn.rst Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org> * Added RMSProp in optimizer_kwargs, as described in nature paper * Exploration fraction is inverse of 50.000.000 (total frames) / 1.000.000 (frames with linear schedule) according to nature paper * Changelog update for buffer dtype * standard exlude parameters should be always excluded to assure proper saving only if intentionally included by ``include`` parameter * slightly more iterations on test_discrete to pass the test * added param use_rms_prop instead of mutable default argument * forgot alpha * using huber loss, adam and learning rate 1e-4 * account for train_freq in update_target_network * Added memory check for both buffers * Doc updated for buffer allocation * Added psutil Requirement * Adapted test_identity.py * Fixes with new SB3 version * Fix for tensorboard name * Convert assert to warning and fix tests * Refactor off-policy algorithms * Fixes * test: remove next_obs in replay buffer * Update changelog * Fix tests and use tmp_path where possible * Fix sampling bug in buffer * Do not store next obs on episode termination * Fix replay buffer sampling * Update comment * moved epsilon from policy to model * Update predict method * Update atari wrappers to match SB2 * Minor edit in the buffers * Update changelog * Merge branch 'master' into dqn * Update DQN to new structure * Fix tests and remove hardcoded path * Fix for DQN * Disable memory efficient replay buffer by default * Fix docstring * Add tests for memory efficient buffer * Update changelog * Split collect rollout * Move target update outside `train()` for DQN * Update changelog * Update linear schedule doc * Cleanup DQN code * Minor edit * Update version and docker images Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
2020-06-29 09:16:54 +00:00
if self.optimize_memory_usage:
next_obs = self._normalize_obs(self.observations[(batch_inds + 1) % self.buffer_size, 0, :], env)
else:
next_obs = self._normalize_obs(self.next_observations[batch_inds, 0, :], env)
data = (
self._normalize_obs(self.observations[batch_inds, 0, :], env),
self.actions[batch_inds, 0, :],
next_obs,
self.dones[batch_inds],
self._normalize_reward(self.rewards[batch_inds], env),
)
2020-03-10 15:43:10 +00:00
return ReplayBufferSamples(*tuple(map(self.to_torch, data)))
2019-09-18 13:35:17 +00:00
2019-09-19 09:43:15 +00:00
class RolloutBuffer(BaseBuffer):
2019-11-22 12:33:12 +00:00
"""
Rollout buffer used in on-policy algorithms like A2C/PPO.
:param buffer_size: (int) Max number of element in the buffer
:param observation_space: (spaces.Space) Observation space
:param action_space: (spaces.Space) Action space
2019-11-22 12:33:12 +00:00
:param device: (th.device)
:param gae_lambda: (float) Factor for trade-off of bias vs variance for Generalized Advantage Estimator
Equivalent to classic advantage when set to 1.
:param gamma: (float) Discount factor
:param n_envs: (int) Number of parallel environments
"""
2020-03-12 10:12:10 +00:00
def __init__(
self,
buffer_size: int,
observation_space: spaces.Space,
action_space: spaces.Space,
device: Union[th.device, str] = "cpu",
gae_lambda: float = 1,
gamma: float = 0.99,
n_envs: int = 1,
):
super(RolloutBuffer, self).__init__(buffer_size, observation_space, action_space, device, n_envs=n_envs)
2019-09-24 12:53:03 +00:00
self.gae_lambda = gae_lambda
2019-09-18 13:35:17 +00:00
self.gamma = gamma
2019-09-24 12:53:03 +00:00
self.observations, self.actions, self.rewards, self.advantages = None, None, None, None
2019-09-20 13:19:04 +00:00
self.returns, self.dones, self.values, self.log_probs = None, None, None, None
self.generator_ready = False
self.reset()
2020-02-03 14:40:34 +00:00
def reset(self) -> None:
self.observations = np.zeros((self.buffer_size, self.n_envs) + self.obs_shape, dtype=np.float32)
2020-02-03 14:40:34 +00:00
self.actions = np.zeros((self.buffer_size, self.n_envs, self.action_dim), dtype=np.float32)
self.rewards = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
self.returns = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
self.dones = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
self.values = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
self.log_probs = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
self.advantages = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
2019-09-20 13:19:04 +00:00
self.generator_ready = False
super(RolloutBuffer, self).reset()
def compute_returns_and_advantage(self, last_value: th.Tensor, dones: np.ndarray) -> None:
2019-09-18 13:35:17 +00:00
"""
2019-11-22 12:33:12 +00:00
Post-processing step: compute the returns (sum of discounted rewards)
and GAE advantage.
2019-11-22 12:33:12 +00:00
Adapted from Stable-Baselines PPO2.
Uses Generalized Advantage Estimation (https://arxiv.org/abs/1506.02438)
to compute the advantage. To obtain vanilla advantage (A(s) = R - V(S))
where R is the discounted reward with value bootstrap,
set ``gae_lambda=1.0`` during initialization.
2019-10-29 11:36:40 +00:00
:param last_value: (th.Tensor)
2020-02-03 14:40:34 +00:00
:param dones: (np.ndarray)
2019-09-18 13:35:17 +00:00
"""
2020-02-03 14:40:34 +00:00
# convert to numpy
last_value = last_value.clone().cpu().numpy().flatten()
last_gae_lam = 0
for step in reversed(range(self.buffer_size)):
if step == self.buffer_size - 1:
next_non_terminal = 1.0 - dones
next_value = last_value
else:
next_non_terminal = 1.0 - self.dones[step + 1]
next_value = self.values[step + 1]
delta = self.rewards[step] + self.gamma * next_value * next_non_terminal - self.values[step]
last_gae_lam = delta + self.gamma * self.gae_lambda * next_non_terminal * last_gae_lam
self.advantages[step] = last_gae_lam
self.returns = self.advantages + self.values
2019-09-18 13:35:17 +00:00
def add(
self, obs: np.ndarray, action: np.ndarray, reward: np.ndarray, done: np.ndarray, value: th.Tensor, log_prob: th.Tensor
) -> None:
2019-11-22 12:33:12 +00:00
"""
:param obs: (np.ndarray) Observation
:param action: (np.ndarray) Action
:param reward: (np.ndarray)
:param done: (np.ndarray) End of episode signal.
:param value: (th.Tensor) estimated value of the current state
following the current policy.
:param log_prob: (th.Tensor) log probability of the action
following the current policy.
"""
if len(log_prob.shape) == 0:
# Reshape 0-d tensor to avoid error
log_prob = log_prob.reshape(-1, 1)
2019-10-25 10:00:37 +00:00
2020-02-03 14:40:34 +00:00
self.observations[self.pos] = np.array(obs).copy()
self.actions[self.pos] = np.array(action).copy()
self.rewards[self.pos] = np.array(reward).copy()
self.dones[self.pos] = np.array(done).copy()
self.values[self.pos] = value.clone().cpu().numpy().flatten()
self.log_probs[self.pos] = log_prob.clone().cpu().numpy()
2019-09-19 09:43:15 +00:00
self.pos += 1
if self.pos == self.buffer_size:
self.full = True
2019-09-18 13:35:17 +00:00
2020-02-11 16:33:22 +00:00
def get(self, batch_size: Optional[int] = None) -> Generator[RolloutBufferSamples, None, None]:
assert self.full, ""
2020-02-03 14:40:34 +00:00
indices = np.random.permutation(self.buffer_size * self.n_envs)
2019-09-20 13:19:04 +00:00
# Prepare the data
if not self.generator_ready:
for tensor in ["observations", "actions", "values", "log_probs", "advantages", "returns"]:
2019-09-20 13:19:04 +00:00
self.__dict__[tensor] = self.swap_and_flatten(self.__dict__[tensor])
self.generator_ready = True
2019-10-28 13:27:32 +00:00
# Return everything, don't create minibatches
if batch_size is None:
batch_size = self.buffer_size * self.n_envs
2019-09-19 09:43:15 +00:00
start_idx = 0
2019-09-20 13:19:04 +00:00
while start_idx < self.buffer_size * self.n_envs:
yield self._get_samples(indices[start_idx : start_idx + batch_size])
2019-09-19 09:43:15 +00:00
start_idx += batch_size
def _get_samples(self, batch_inds: np.ndarray, env: Optional[VecNormalize] = None) -> RolloutBufferSamples:
data = (
self.observations[batch_inds],
self.actions[batch_inds],
self.values[batch_inds].flatten(),
self.log_probs[batch_inds].flatten(),
self.advantages[batch_inds].flatten(),
self.returns[batch_inds].flatten(),
)
2020-03-10 15:43:10 +00:00
return RolloutBufferSamples(*tuple(map(self.to_torch, data)))