hongshaorou/Tianshou
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Multimodal obs (#38, #27, #25)

Browse Source
This commit is contained in:
Trinkle23897 2020-04-28 20:56:02 +08:00
parent 959955fa2a
commit 80d661907e
9 changed files with 212 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
README.md
View File

@ -5,9 +5,9 @@
---
[![PyPI](https://img.shields.io/pypi/v/tianshou)](https://pypi.org/project/tianshou/)
[![Documentation Status](https://readthedocs.org/projects/tianshou/badge/?version=latest)](https://tianshou.readthedocs.io)
[![Unittest](https://github.com/thu-ml/tianshou/workflows/Unittest/badge.svg?branch=master)](https://github.com/thu-ml/tianshou/actions)
[![codecov](https://img.shields.io/codecov/c/gh/thu-ml/tianshou)](https://codecov.io/gh/thu-ml/tianshou)
[![Documentation Status](https://readthedocs.org/projects/tianshou/badge/?version=latest)](https://tianshou.readthedocs.io)
[![GitHub issues](https://img.shields.io/github/issues/thu-ml/tianshou)](https://github.com/thu-ml/tianshou/issues)
[![GitHub stars](https://img.shields.io/github/stars/thu-ml/tianshou)](https://github.com/thu-ml/tianshou/stargazers)
[![GitHub forks](https://img.shields.io/github/forks/thu-ml/tianshou)](https://github.com/thu-ml/tianshou/network)
@ -40,7 +40,7 @@ In Chinese, Tianshou means the innate talent, not taught by others. Tianshou is
Tianshou is currently hosted on [PyPI](https://pypi.org/project/tianshou/). It requires Python >= 3.6. You can simply install Tianshou with the following command:
```bash
pip3 install tianshou -U
pip3 install tianshou
```
You can also install with the newest version through GitHub:
@ -49,6 +49,17 @@ You can also install with the newest version through GitHub:
pip3 install git+https://github.com/thu-ml/tianshou.git@master
```
If you use Anaconda or Miniconda, you can install Tianshou through the following command lines:
```bash
# create a new virtualenv and install pip, change the env name if you like
conda create -n myenv pip
# activate the environment
conda activate myenv
# install tianshou
pip install tianshou
```
After installation, open your python console and type
```python

12
docs/index.rst
View File

@ -30,13 +30,23 @@ Installation
Tianshou is currently hosted on `PyPI <https://pypi.org/project/tianshou/>`_. You can simply install Tianshou with the following command:
::
pip3 install tianshou -U
pip3 install tianshou
You can also install with the newest version through GitHub:
::
pip3 install git+https://github.com/thu-ml/tianshou.git@master
If you use Anaconda or Miniconda, you can install Tianshou through the following command lines:
::
# create a new virtualenv and install pip, change the env name if you like
conda create -n myenv pip
# activate the environment
conda activate myenv
# install tianshou
pip install tianshou
After installation, open your python console and type
::

21
test/base/env.py
View File

@ -3,15 +3,16 @@ import time
class MyTestEnv(gym.Env):
def __init__(self, size, sleep=0):
def __init__(self, size, sleep=0, dict_state=False):
self.size = size
self.sleep = sleep
self.dict_state = dict_state
self.reset()
def reset(self, state=0):
self.done = False
self.index = state
return self.index
return {'index': self.index} if self.dict_state else self.index
def step(self, action):
if self.done:
@ -20,11 +21,21 @@ class MyTestEnv(gym.Env):
time.sleep(self.sleep)
if self.index == self.size:
self.done = True
return self.index, 0, True, {}
if self.dict_state:
return {'index': self.index}, 0, True, {}
else:
return self.index, 0, True, {}
if action == 0:
self.index = max(self.index - 1, 0)
return self.index, 0, False, {}
if self.dict_state:
return {'index': self.index}, 0, False, {}
else:
return self.index, 0, False, {}
elif action == 1:
self.index += 1
self.done = self.index == self.size
return self.index, int(self.done), self.done, {'key': 1}
if self.dict_state:
return {'index': self.index}, int(self.done), self.done, \
{'key': 1}
else:
return self.index, int(self.done), self.done, {'key': 1}

2
test/base/test_batch.py
View File

@ -15,7 +15,7 @@ def test_batch():
with pytest.raises(IndexError):
batch[2]
batch.obs = np.arange(5)
for i, b in enumerate(batch.split(1, permute=False)):
for i, b in enumerate(batch.split(1, shuffle=False)):
assert b.obs == batch[i].obs
print(batch)

26
test/base/test_collector.py
View File

@ -2,7 +2,7 @@ import numpy as np
from torch.utils.tensorboard import SummaryWriter
from tianshou.policy import BasePolicy
from tianshou.env import SubprocVectorEnv
from tianshou.env import VectorEnv, SubprocVectorEnv
from tianshou.data import Collector, Batch, ReplayBuffer
if __name__ == '__main__':
@ -12,10 +12,13 @@ else: # pytest
class MyPolicy(BasePolicy):
def __init__(self):
def __init__(self, dict_state=False):
super().__init__()
self.dict_state = dict_state
def forward(self, batch, state=None):
if self.dict_state:
return Batch(act=np.ones(batch.obs['index'].shape[0]))
return Batch(act=np.ones(batch.obs.shape[0]))
def learn(self):
@ -75,5 +78,24 @@ def test_collector():
1, 2, 3, 1, 2, 3, 4, 1, 2, 3, 4, 5])
def test_collector_with_dict_state():
env = MyTestEnv(size=5, sleep=0, dict_state=True)
policy = MyPolicy(dict_state=True)
c0 = Collector(policy, env, ReplayBuffer(size=100))
c0.collect(n_step=3)
c0.collect(n_episode=3)
env_fns = [
lambda: MyTestEnv(size=2, sleep=0, dict_state=True),
lambda: MyTestEnv(size=3, sleep=0, dict_state=True),
lambda: MyTestEnv(size=4, sleep=0, dict_state=True),
lambda: MyTestEnv(size=5, sleep=0, dict_state=True),
]
envs = VectorEnv(env_fns)
c1 = Collector(policy, envs, ReplayBuffer(size=100))
c1.collect(n_step=10)
c1.collect(n_episode=[2, 1, 1, 2])
if __name__ == '__main__':
test_collector()
test_collector_with_dict_state()

63
tianshou/data/batch.py
View File

@ -1,4 +1,5 @@
import torch
import pprint
import numpy as np
@ -23,7 +24,7 @@ class Batch(object):
)
In short, you can define a :class:`Batch` with any key-value pair. The
current implementation of Tianshou typically use 6 keys in
current implementation of Tianshou typically use 6 reserved keys in
:class:`~tianshou.data.Batch`:
* ``obs`` the observation of step :math:`t` ;
@ -56,7 +57,7 @@ class Batch(object):
array([0, 11, 22, 0, 11, 22])
>>> # split whole data into multiple small batch
>>> for d in data.split(size=2, permute=False):
>>> for d in data.split(size=2, shuffle=False):
... print(d.obs, d.rew)
[ 0 11] [6 6]
[22 0] [6 6]
@ -65,24 +66,56 @@ class Batch(object):
def __init__(self, **kwargs):
super().__init__()
self.__dict__.update(kwargs)
self._meta = {}
for k, v in kwargs.items():
if (isinstance(v, list) or isinstance(v, np.ndarray)) \
and len(v) > 0 and isinstance(v[0], dict) and k != 'info':
self._meta[k] = list(v[0].keys())
for k_ in v[0].keys():
k__ = '_' + k + '@' + k_
self.__dict__[k__] = np.array([
v[i][k_] for i in range(len(v))
])
elif isinstance(v, dict):
self._meta[k] = list(v.keys())
for k_ in v.keys():
k__ = '_' + k + '@' + k_
self.__dict__[k__] = v[k_]
else:
self.__dict__[k] = kwargs[k]
def __getitem__(self, index):
"""Return self[index]."""
if isinstance(index, str):
return self.__getattr__(index)
b = Batch()
for k in self.__dict__.keys():
if self.__dict__[k] is not None:
if k != '_meta' and self.__dict__[k] is not None:
b.__dict__.update(**{k: self.__dict__[k][index]})
b._meta = self._meta
return b
def __getattr__(self, key):
"""Return self.key"""
if key not in self._meta.keys():
if key not in self.__dict__.keys():
raise AttributeError(key)
return self.__dict__[key]
d = {}
for k_ in self._meta[key]:
k__ = '_' + key + '@' + k_
d[k_] = self.__dict__[k__]
return d
def __repr__(self):
"""Return str(self)."""
s = self.__class__.__name__ + '(\n'
flag = False
for k in sorted(self.__dict__.keys()):
if k[0] != '_' and self.__dict__[k] is not None:
for k in sorted(list(self.__dict__.keys()) + list(self._meta.keys())):
if k[0] != '_' and (self.__dict__.get(k, None) is not None or
k in self._meta.keys()):
rpl = '\n' + ' ' * (6 + len(k))
obj = str(self.__dict__[k]).replace('\n', rpl)
obj = pprint.pformat(self.__getattr__(k)).replace('\n', rpl)
s += f' {k}: {obj},\n'
flag = True
if flag:
@ -91,10 +124,18 @@ class Batch(object):
s = self.__class__.__name__ + '()\n'
return s
def keys(self):
"""Return self.keys()."""
return sorted([i for i in self.__dict__.keys() if i[0] != '_'] +
list(self._meta.keys()))
def append(self, batch):
"""Append a :class:`~tianshou.data.Batch` object to current batch."""
assert isinstance(batch, Batch), 'Only append Batch is allowed!'
for k in batch.__dict__.keys():
if k == '_meta':
self._meta.update(batch.__dict__[k])
continue
if batch.__dict__[k] is None:
continue
if not hasattr(self, k) or self.__dict__[k] is None:
@ -117,22 +158,22 @@ class Batch(object):
"""Return len(self)."""
return min([
len(self.__dict__[k]) for k in self.__dict__.keys()
if self.__dict__[k] is not None])
if k != '_meta' and self.__dict__[k] is not None])
def split(self, size=None, permute=True):
def split(self, size=None, shuffle=True):
"""Split whole data into multiple small batch.
:param int size: if it is ``None``, it does not split the data batch;
otherwise it will divide the data batch with the given size.
Default to ``None``.
:param bool permute: randomly shuffle the entire data batch if it is
:param bool shuffle: randomly shuffle the entire data batch if it is
``True``, otherwise remain in the same. Default to ``True``.
"""
length = len(self)
if size is None:
size = length
temp = 0
if permute:
if shuffle:
index = np.random.permutation(length)
else:
index = np.arange(length)

119
tianshou/data/buffer.py
View File

@ -1,3 +1,4 @@
import pprint
import numpy as np
from tianshou.data.batch import Batch
@ -92,6 +93,7 @@ class ReplayBuffer(object):
self._maxsize = size
self._stack = stack_num
self._save_s_ = not ignore_obs_next
self._meta = {}
self.reset()
def __len__(self):
@ -102,10 +104,11 @@ class ReplayBuffer(object):
"""Return str(self)."""
s = self.__class__.__name__ + '(\n'
flag = False
for k in self.__dict__.keys():
if k[0] != '_' and self.__dict__[k] is not None:
for k in sorted(list(self.__dict__.keys()) + list(self._meta.keys())):
if k[0] != '_' and (self.__dict__.get(k, None) is not None or
k in self._meta.keys()):
rpl = '\n' + ' ' * (6 + len(k))
obj = str(self.__dict__[k]).replace('\n', rpl)
obj = pprint.pformat(self.__getattr__(k)).replace('\n', rpl)
s += f' {k}: {obj},\n'
flag = True
if flag:
@ -114,23 +117,51 @@ class ReplayBuffer(object):
s = self.__class__.__name__ + '()\n'
return s
def __getattr__(self, key):
"""Return self.key"""
if key not in self._meta.keys():
if key not in self.__dict__.keys():
raise AttributeError(key)
return self.__dict__[key]
d = {}
for k_ in self._meta[key]:
k__ = '_' + key + '@' + k_
d[k_] = self.__dict__[k__]
return d
def _add_to_buffer(self, name, inst):
if inst is None:
if getattr(self, name, None) is None:
self.__dict__[name] = None
return
if name in self._meta.keys():
for k in inst.keys():
self._add_to_buffer('_' + name + '@' + k, inst[k])
return
if self.__dict__.get(name, None) is None:
if isinstance(inst, np.ndarray):
self.__dict__[name] = np.zeros([self._maxsize, *inst.shape])
elif isinstance(inst, dict):
self.__dict__[name] = np.array(
[{} for _ in range(self._maxsize)])
if name == 'info':
self.__dict__[name] = np.array(
[{} for _ in range(self._maxsize)])
else:
if self._meta.get(name, None) is None:
self._meta[name] = [
'_' + name + '@' + k for k in inst.keys()]
for k in inst.keys():
k_ = '_' + name + '@' + k
self._add_to_buffer(k_, inst[k])
else: # assume `inst` is a number
self.__dict__[name] = np.zeros([self._maxsize])
if isinstance(inst, np.ndarray) and \
self.__dict__[name].shape[1:] != inst.shape:
self.__dict__[name] = np.zeros([self._maxsize, *inst.shape])
self.__dict__[name][self._index] = inst
raise ValueError(
"Cannot add data to a buffer with different shape, "
f"key: {name}, expect shape: {self.__dict__[name].shape[1:]}, "
f"given shape: {inst.shape}.")
if name not in self._meta.keys():
self.__dict__[name][self._index] = inst
def update(self, buffer):
"""Move the data from the given buffer to self."""
@ -144,7 +175,8 @@ class ReplayBuffer(object):
if i == begin:
break
def add(self, obs, act, rew, done, obs_next=None, info={}, weight=None):
def add(self, obs, act, rew, done, obs_next=None, info={}, policy={},
**kwargs):
"""Add a batch of data into replay buffer."""
assert isinstance(info, dict), \
'You should return a dict in the last argument of env.step().'
@ -155,6 +187,7 @@ class ReplayBuffer(object):
if self._save_s_:
self._add_to_buffer('obs_next', obs_next)
self._add_to_buffer('info', info)
self._add_to_buffer('policy', policy)
if self._maxsize > 0:
self._size = min(self._size + 1, self._maxsize)
self._index = (self._index + 1) % self._maxsize
@ -180,11 +213,13 @@ class ReplayBuffer(object):
])
return self[indice], indice
def get(self, indice, key):
def get(self, indice, key, stack_num=None):
"""Return the stacked result, e.g. [s_{t-3}, s_{t-2}, s_{t-1}, s_t],
where s is self.key, t is indice. The stack_num (here equals to 4) is
given from buffer initialization procedure.
"""
if stack_num is None:
stack_num = self._stack
if not isinstance(indice, np.ndarray):
if np.isscalar(indice):
indice = np.array(indice)
@ -200,18 +235,37 @@ class ReplayBuffer(object):
indice += 1 - self.done[indice].astype(np.int)
indice[indice == self._size] = 0
key = 'obs'
if self._stack == 0:
if stack_num == 0:
self.done[last_index] = last_done
return self.__dict__[key][indice]
stack = []
for i in range(self._stack):
stack = [self.__dict__[key][indice]] + stack
if key in self._meta:
return {k.split('@')[-1]: self.__dict__[k][indice]
for k in self._meta[key]}
else:
return self.__dict__[key][indice]
if key in self._meta:
many_keys = self._meta[key]
stack = {k.split('@')[-1]: [] for k in self._meta[key]}
else:
stack = []
many_keys = None
for i in range(stack_num):
if many_keys is not None:
for k_ in many_keys:
k = k_.split('@')[-1]
stack[k] = [self.__dict__[k_][indice]] + stack[k]
else:
stack = [self.__dict__[key][indice]] + stack
pre_indice = indice - 1
pre_indice[pre_indice == -1] = self._size - 1
indice = pre_indice + self.done[pre_indice].astype(np.int)
indice[indice == self._size] = 0
self.done[last_index] = last_done
return np.stack(stack, axis=1)
if many_keys is not None:
for k in stack:
stack[k] = np.stack(stack[k], axis=1)
else:
stack = np.stack(stack, axis=1)
return stack
def __getitem__(self, index):
"""Return a data batch: self[index]. If stack_num is set to be > 0,
@ -223,7 +277,8 @@ class ReplayBuffer(object):
rew=self.rew[index],
done=self.done[index],
obs_next=self.get(index, 'obs_next'),
info=self.info[index]
info=self.info[index],
policy=self.get(index, 'policy'),
)
@ -234,7 +289,7 @@ class ListReplayBuffer(ReplayBuffer):
.. seealso::
Please refer to :class:`~tianshou.data.ListReplayBuffer` for more
Please refer to :class:`~tianshou.data.ReplayBuffer` for more
detailed explanation.
"""
@ -256,7 +311,13 @@ class ListReplayBuffer(ReplayBuffer):
class PrioritizedReplayBuffer(ReplayBuffer):
"""docstring for PrioritizedReplayBuffer"""
"""Prioritized replay buffer implementation.
.. seealso::
Please refer to :class:`~tianshou.data.ReplayBuffer` for more
detailed explanation.
"""
def __init__(self, size, alpha: float, beta: float,
mode: str = 'weight', **kwargs):
@ -270,17 +331,18 @@ class PrioritizedReplayBuffer(ReplayBuffer):
self._amortization_freq = 50
self._amortization_counter = 0
def add(self, obs, act, rew, done, obs_next=0, info={}, weight=1.0):
def add(self, obs, act, rew, done, obs_next=0, info={}, policy={},
weight=1.0):
"""Add a batch of data into replay buffer."""
self._weight_sum += np.abs(weight)**self._alpha - \
self.weight[self._index]
# we have to sacrifice some convenience for speed :(
self._add_to_buffer('weight', np.abs(weight)**self._alpha)
super().add(obs, act, rew, done, obs_next, info)
self._add_to_buffer('weight', np.abs(weight) ** self._alpha)
super().add(obs, act, rew, done, obs_next, info, policy)
self._check_weight_sum()
def sample(self, batch_size: int = 0, importance_sample: bool = True):
""" Get a random sample from buffer with priority probability. \
"""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.
@ -290,7 +352,8 @@ class PrioritizedReplayBuffer(ReplayBuffer):
# will cause weight update conflict
indice = np.random.choice(
self._size, batch_size,
p=(self.weight/self.weight.sum())[:self._size], replace=False)
p=(self.weight / self.weight.sum())[:self._size],
replace=False)
# self._weight_sum is not work for the accuracy issue
# p=(self.weight/self._weight_sum)[:self._size], replace=False)
elif batch_size == 0:
@ -305,8 +368,9 @@ class PrioritizedReplayBuffer(ReplayBuffer):
batch = self[indice]
if importance_sample:
impt_weight = Batch(
impt_weight=1/np.power(
self._size*(batch.weight/self._weight_sum), self._beta))
impt_weight=1 / np.power(
self._size * (batch.weight / self._weight_sum),
self._beta))
batch.append(impt_weight)
self._check_weight_sum()
return batch, indice
@ -316,7 +380,7 @@ class PrioritizedReplayBuffer(ReplayBuffer):
super().reset()
def update_weight(self, indice, new_weight: np.ndarray):
"""update priority weight by indice in this buffer
"""Update priority weight by indice in this buffer.
:param indice: indice you want to update weight
:param new_weight: new priority weight you wangt to update
@ -333,7 +397,8 @@ class PrioritizedReplayBuffer(ReplayBuffer):
done=self.done[index],
obs_next=self.get(index, 'obs_next'),
info=self.info[index],
weight=self.weight[index]
weight=self.weight[index],
policy=self.get(index, 'policy'),
)
def _check_weight_sum(self):

2
tianshou/policy/modelfree/a2c.py
View File

@ -54,7 +54,7 @@ class A2CPolicy(PGPolicy):
batch, None, gamma=self._gamma, gae_lambda=self._lambda)
v_ = []
with torch.no_grad():
for b in batch.split(self._batch, permute=False):
for b in batch.split(self._batch, shuffle=False):
v_.append(self.critic(b.obs_next).detach().cpu().numpy())
v_ = np.concatenate(v_, axis=0)
return self.compute_episodic_return(

4
tianshou/policy/modelfree/ppo.py
View File

@ -74,7 +74,7 @@ class PPOPolicy(PGPolicy):
batch, None, gamma=self._gamma, gae_lambda=self._lambda)
v_ = []
with torch.no_grad():
for b in batch.split(self._batch, permute=False):
for b in batch.split(self._batch, shuffle=False):
v_.append(self.critic(b.obs_next))
v_ = torch.cat(v_, dim=0).cpu().numpy()
return self.compute_episodic_return(
@ -111,7 +111,7 @@ class PPOPolicy(PGPolicy):
v = []
old_log_prob = []
with torch.no_grad():
for b in batch.split(batch_size, permute=False):
for b in batch.split(batch_size, shuffle=False):
v.append(self.critic(b.obs))
old_log_prob.append(self(b).dist.log_prob(
torch.tensor(b.act, device=v[0].device)))