Tianshou/tianshou/data/buffer.py
2020-09-14 15:59:32 +08:00

497 lines
18 KiB
Python

import torch
import numpy as np
from numbers import Number
from typing import Any, Dict, List, Tuple, Union, Optional
from tianshou.data import Batch, SegmentTree, to_numpy
from tianshou.data.batch import _create_value
class ReplayBuffer:
""":class:`~tianshou.data.ReplayBuffer` stores data generated from \
interaction between the policy and environment.
The current implementation of Tianshou typically use 7 reserved keys in
:class:`~tianshou.data.Batch`:
* ``obs`` the observation of step :math:`t` ;
* ``act`` the action of step :math:`t` ;
* ``rew`` the reward of step :math:`t` ;
* ``done`` the done flag of step :math:`t` ;
* ``obs_next`` the observation of step :math:`t+1` ;
* ``info`` the info of step :math:`t` (in ``gym.Env``, the ``env.step()`` \
function returns 4 arguments, and the last one is ``info``);
* ``policy`` the data computed by policy in step :math:`t`;
The following code snippet illustrates its usage:
::
>>> import pickle, numpy as np
>>> from tianshou.data import ReplayBuffer
>>> buf = ReplayBuffer(size=20)
>>> for i in range(3):
... buf.add(obs=i, act=i, rew=i, done=i, obs_next=i + 1, info={})
>>> buf.obs
# since we set size = 20, len(buf.obs) == 20.
array([0., 1., 2., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0.])
>>> # but there are only three valid items, so len(buf) == 3.
>>> len(buf)
3
>>> pickle.dump(buf, open('buf.pkl', 'wb')) # save to file "buf.pkl"
>>> buf2 = ReplayBuffer(size=10)
>>> for i in range(15):
... buf2.add(obs=i, act=i, rew=i, done=i, obs_next=i + 1, info={})
>>> len(buf2)
10
>>> buf2.obs
# since its size = 10, it only stores the last 10 steps' result.
array([10., 11., 12., 13., 14., 5., 6., 7., 8., 9.])
>>> # move buf2's result into buf (meanwhile keep it chronologically)
>>> buf.update(buf2)
array([ 0., 1., 2., 5., 6., 7., 8., 9., 10., 11., 12., 13., 14.,
0., 0., 0., 0., 0., 0., 0.])
>>> # get a random sample from buffer
>>> # the batch_data is equal to buf[incide].
>>> batch_data, indice = buf.sample(batch_size=4)
>>> batch_data.obs == buf[indice].obs
array([ True, True, True, True])
>>> len(buf)
13
>>> buf = pickle.load(open('buf.pkl', 'rb')) # load from "buf.pkl"
>>> len(buf)
3
:class:`~tianshou.data.ReplayBuffer` also supports frame_stack sampling
(typically for RNN usage, see issue#19), ignoring storing the next
observation (save memory in atari tasks), and multi-modal observation (see
issue#38):
::
>>> buf = ReplayBuffer(size=9, stack_num=4, ignore_obs_next=True)
>>> for i in range(16):
... done = i % 5 == 0
... buf.add(obs={'id': i}, act=i, rew=i, done=done,
... obs_next={'id': i + 1})
>>> print(buf) # you can see obs_next is not saved in buf
ReplayBuffer(
act: array([ 9., 10., 11., 12., 13., 14., 15., 7., 8.]),
done: array([0., 1., 0., 0., 0., 0., 1., 0., 0.]),
info: Batch(),
obs: Batch(
id: array([ 9., 10., 11., 12., 13., 14., 15., 7., 8.]),
),
policy: Batch(),
rew: array([ 9., 10., 11., 12., 13., 14., 15., 7., 8.]),
)
>>> index = np.arange(len(buf))
>>> print(buf.get(index, 'obs').id)
[[ 7. 7. 8. 9.]
[ 7. 8. 9. 10.]
[11. 11. 11. 11.]
[11. 11. 11. 12.]
[11. 11. 12. 13.]
[11. 12. 13. 14.]
[12. 13. 14. 15.]
[ 7. 7. 7. 7.]
[ 7. 7. 7. 8.]]
>>> # here is another way to get the stacked data
>>> # (stack only for obs and obs_next)
>>> abs(buf.get(index, 'obs')['id'] - buf[index].obs.id).sum().sum()
0.0
>>> # we can get obs_next through __getitem__, even if it doesn't exist
>>> print(buf[:].obs_next.id)
[[ 7. 8. 9. 10.]
[ 7. 8. 9. 10.]
[11. 11. 11. 12.]
[11. 11. 12. 13.]
[11. 12. 13. 14.]
[12. 13. 14. 15.]
[12. 13. 14. 15.]
[ 7. 7. 7. 8.]
[ 7. 7. 8. 9.]]
:param int size: the size of replay buffer.
:param int stack_num: the frame-stack sampling argument, should be greater
than or equal to 1, defaults to 1 (no stacking).
:param bool ignore_obs_next: whether to store obs_next, defaults to False.
:param bool save_only_last_obs: only save the last obs/obs_next when it has
a shape of (timestep, ...) because of temporal stacking, defaults to
False.
:param bool sample_avail: the parameter indicating sampling only available
index when using frame-stack sampling method, defaults to False.
This feature is not supported in Prioritized Replay Buffer currently.
"""
def __init__(
self,
size: int,
stack_num: int = 1,
ignore_obs_next: bool = False,
save_only_last_obs: bool = False,
sample_avail: bool = False,
) -> None:
super().__init__()
self._maxsize = size
self._indices = np.arange(size)
self.stack_num = stack_num
self._avail = sample_avail and stack_num > 1
self._avail_index: List[int] = []
self._save_s_ = not ignore_obs_next
self._last_obs = save_only_last_obs
self._index = 0
self._size = 0
self._meta: Batch = Batch()
self.reset()
def __len__(self) -> int:
"""Return len(self)."""
return self._size
def __repr__(self) -> str:
"""Return str(self)."""
return self.__class__.__name__ + self._meta.__repr__()[5:]
def __getattr__(self, key: str) -> Any:
"""Return self.key."""
try:
return self._meta[key]
except KeyError as e:
raise AttributeError from e
def __setstate__(self, state: Dict[str, Any]) -> None:
"""Unpickling interface.
We need it because pickling buffer does not work out-of-the-box
("buffer.__getattr__" is customized).
"""
self.__dict__.update(state)
def _add_to_buffer(self, name: str, inst: Any) -> None:
try:
value = self._meta.__dict__[name]
except KeyError:
self._meta.__dict__[name] = _create_value(inst, self._maxsize)
value = self._meta.__dict__[name]
if isinstance(inst, (torch.Tensor, np.ndarray)):
if inst.shape != value.shape[1:]:
raise ValueError(
"Cannot add data to a buffer with different shape with key"
f" {name}, expect {value.shape[1:]}, given {inst.shape}."
)
try:
value[self._index] = inst
except KeyError:
for key in set(inst.keys()).difference(value.__dict__.keys()):
value.__dict__[key] = _create_value(inst[key], self._maxsize)
value[self._index] = inst
@property
def stack_num(self) -> int:
return self._stack
@stack_num.setter
def stack_num(self, num: int) -> None:
assert num > 0, "stack_num should greater than 0"
self._stack = num
def update(self, buffer: "ReplayBuffer") -> None:
"""Move the data from the given buffer to self."""
if len(buffer) == 0:
return
i = begin = buffer._index % len(buffer)
stack_num_orig = buffer.stack_num
buffer.stack_num = 1
while True:
self.add(**buffer[i]) # type: ignore
i = (i + 1) % len(buffer)
if i == begin:
break
buffer.stack_num = stack_num_orig
def add(
self,
obs: Any,
act: Any,
rew: Union[Number, np.number, np.ndarray],
done: Union[Number, np.number, np.bool_],
obs_next: Any = None,
info: Optional[Union[dict, Batch]] = {},
policy: Optional[Union[dict, Batch]] = {},
**kwargs: Any,
) -> None:
"""Add a batch of data into replay buffer."""
assert isinstance(
info, (dict, Batch)
), "You should return a dict in the last argument of env.step()."
if self._last_obs:
obs = obs[-1]
self._add_to_buffer("obs", obs)
self._add_to_buffer("act", act)
self._add_to_buffer("rew", rew)
self._add_to_buffer("done", done)
if self._save_s_:
if obs_next is None:
obs_next = Batch()
elif self._last_obs:
obs_next = obs_next[-1]
self._add_to_buffer("obs_next", obs_next)
self._add_to_buffer("info", info)
self._add_to_buffer("policy", policy)
# maintain available index for frame-stack sampling
if self._avail:
# update current frame
avail = sum(self.done[i] for i in range(
self._index - self.stack_num + 1, self._index)) == 0
if self._size < self.stack_num - 1:
avail = False
if avail and self._index not in self._avail_index:
self._avail_index.append(self._index)
elif not avail and self._index in self._avail_index:
self._avail_index.remove(self._index)
# remove the later available frame because of broken storage
t = (self._index + self.stack_num - 1) % self._maxsize
if t in self._avail_index:
self._avail_index.remove(t)
if self._maxsize > 0:
self._size = min(self._size + 1, self._maxsize)
self._index = (self._index + 1) % self._maxsize
else:
self._size = self._index = self._index + 1
def reset(self) -> None:
"""Clear all the data in replay buffer."""
self._index = 0
self._size = 0
self._avail_index = []
def sample(self, batch_size: int) -> Tuple[Batch, np.ndarray]:
"""Get a random sample from buffer with size equal to batch_size.
Return all the data in the buffer if batch_size is 0.
:return: Sample data and its corresponding index inside the buffer.
"""
if batch_size > 0:
_all = self._avail_index if self._avail else self._size
indice = np.random.choice(_all, batch_size)
else:
if self._avail:
indice = np.array(self._avail_index)
else:
indice = np.concatenate([
np.arange(self._index, self._size),
np.arange(0, self._index),
])
assert len(indice) > 0, "No available indice can be sampled."
return self[indice], indice
def get(
self,
indice: Union[slice, int, np.integer, np.ndarray],
key: str,
stack_num: Optional[int] = None,
) -> Union[Batch, np.ndarray]:
"""Return the stacked result.
E.g. [s_{t-3}, s_{t-2}, s_{t-1}, s_t], where s is self.key, t is the
indice. The stack_num (here equals to 4) is given from buffer
initialization procedure.
"""
if stack_num is None:
stack_num = self.stack_num
if stack_num == 1: # the most often case
if key != "obs_next" or self._save_s_:
val = self._meta.__dict__[key]
try:
return val[indice]
except IndexError as e:
if not (isinstance(val, Batch) and val.is_empty()):
raise e # val != Batch()
return Batch()
indice = self._indices[:self._size][indice]
done = self._meta.__dict__["done"]
if key == "obs_next" and not self._save_s_:
indice += 1 - done[indice].astype(np.int)
indice[indice == self._size] = 0
key = "obs"
val = self._meta.__dict__[key]
try:
if stack_num == 1:
return val[indice]
stack: List[Any] = []
for _ in range(stack_num):
stack = [val[indice]] + stack
pre_indice = np.asarray(indice - 1)
pre_indice[pre_indice == -1] = self._size - 1
indice = np.asarray(
pre_indice + done[pre_indice].astype(np.int))
indice[indice == self._size] = 0
if isinstance(val, Batch):
return Batch.stack(stack, axis=indice.ndim)
else:
return np.stack(stack, axis=indice.ndim)
except IndexError as e:
if not (isinstance(val, Batch) and val.is_empty()):
raise e # val != Batch()
return Batch()
def __getitem__(
self, index: Union[slice, int, np.integer, np.ndarray]
) -> Batch:
"""Return a data batch: self[index].
If stack_num is larger than 1, return the stacked obs and obs_next with
shape (batch, len, ...).
"""
return Batch(
obs=self.get(index, "obs"),
act=self.act[index],
rew=self.rew[index],
done=self.done[index],
obs_next=self.get(index, "obs_next"),
info=self.get(index, "info"),
policy=self.get(index, "policy"),
)
class ListReplayBuffer(ReplayBuffer):
"""List-based replay buffer.
The function of :class:`~tianshou.data.ListReplayBuffer` is almost the same
as :class:`~tianshou.data.ReplayBuffer`. The only difference is that
:class:`~tianshou.data.ListReplayBuffer` is based on list. Therefore,
it does not support advanced indexing, which means you cannot sample a
batch of data out of it. It is typically used for storing data.
.. seealso::
Please refer to :class:`~tianshou.data.ReplayBuffer` for more detailed
explanation.
"""
def __init__(self, **kwargs: Any) -> None:
super().__init__(size=0, ignore_obs_next=False, **kwargs)
def sample(self, batch_size: int) -> Tuple[Batch, np.ndarray]:
raise NotImplementedError("ListReplayBuffer cannot be sampled!")
def _add_to_buffer(
self, name: str, inst: Union[dict, Batch, np.ndarray, float, int, bool]
) -> None:
if self._meta.__dict__.get(name) is None:
self._meta.__dict__[name] = []
self._meta.__dict__[name].append(inst)
def reset(self) -> None:
self._index = self._size = 0
for k in list(self._meta.__dict__.keys()):
if isinstance(self._meta.__dict__[k], list):
self._meta.__dict__[k] = []
class PrioritizedReplayBuffer(ReplayBuffer):
"""Implementation of Prioritized Experience Replay. arXiv:1511.05952.
:param float alpha: the prioritization exponent.
:param float beta: the importance sample soft coefficient.
.. seealso::
Please refer to :class:`~tianshou.data.ReplayBuffer` for more detailed
explanation.
"""
def __init__(
self, size: int, alpha: float, beta: float, **kwargs: Any
) -> None:
super().__init__(size, **kwargs)
assert alpha > 0.0 and beta >= 0.0
self._alpha, self._beta = alpha, beta
self._max_prio = self._min_prio = 1.0
# save weight directly in this class instead of self._meta
self.weight = SegmentTree(size)
self.__eps = np.finfo(np.float32).eps.item()
def add(
self,
obs: Any,
act: Any,
rew: Union[Number, np.number, np.ndarray],
done: Union[Number, np.number, np.bool_],
obs_next: Any = None,
info: Optional[Union[dict, Batch]] = {},
policy: Optional[Union[dict, Batch]] = {},
weight: Optional[Union[Number, np.number]] = None,
**kwargs: Any,
) -> None:
"""Add a batch of data into replay buffer."""
if weight is None:
weight = self._max_prio
else:
weight = np.abs(weight)
self._max_prio = max(self._max_prio, weight)
self._min_prio = min(self._min_prio, weight)
self.weight[self._index] = weight ** self._alpha
super().add(obs, act, rew, done, obs_next, info, policy, **kwargs)
def sample(self, batch_size: int) -> Tuple[Batch, np.ndarray]:
"""Get a random sample from buffer with priority probability.
Return all the data in the buffer if batch_size is 0.
:return: Sample data and its corresponding index inside the buffer.
The "weight" in the returned Batch is the weight on loss function
to de-bias the sampling process (some transition tuples are sampled
more often so their losses are weighted less).
"""
assert self._size > 0, "Cannot sample a buffer with 0 size!"
if batch_size == 0:
indice = np.concatenate([
np.arange(self._index, self._size),
np.arange(0, self._index),
])
else:
scalar = np.random.rand(batch_size) * self.weight.reduce()
indice = self.weight.get_prefix_sum_idx(scalar)
batch = self[indice]
# important sampling weight calculation
# original formula: ((p_j/p_sum*N)**(-beta))/((p_min/p_sum*N)**(-beta))
# simplified formula: (p_j/p_min)**(-beta)
batch.weight = (batch.weight / self._min_prio) ** (-self._beta)
return batch, indice
def update_weight(
self,
indice: Union[np.ndarray],
new_weight: Union[np.ndarray, torch.Tensor]
) -> None:
"""Update priority weight by indice in this buffer.
:param np.ndarray indice: indice you want to update weight.
:param np.ndarray new_weight: new priority weight you want to update.
"""
weight = np.abs(to_numpy(new_weight)) + self.__eps
self.weight[indice] = weight ** self._alpha
self._max_prio = max(self._max_prio, weight.max())
self._min_prio = min(self._min_prio, weight.min())
def __getitem__(
self, index: Union[slice, int, np.integer, np.ndarray]
) -> Batch:
return Batch(
obs=self.get(index, "obs"),
act=self.act[index],
rew=self.rew[index],
done=self.done[index],
obs_next=self.get(index, "obs_next"),
info=self.get(index, "info"),
policy=self.get(index, "policy"),
weight=self.weight[index],
)