94bfb32cc1
1. add policy.eval() in all test scripts' "watch performance" 2. remove dict return support for collector preprocess_fn 3. add `__contains__` and `pop` in batch: `key in batch`, `batch.pop(key, deft)` 4. exact n_episode for a list of n_episode limitation and save fake data in cache_buffer when self.buffer is None (#184) 5. fix tensorboard logging: h-axis stands for env step instead of gradient step; add test results into tensorboard 6. add test_returns (both GAE and nstep) 7. change the type-checking order in batch.py and converter.py in order to meet the most often case first 8. fix shape inconsistency for torch.Tensor in replay buffer 9. remove `**kwargs` in ReplayBuffer 10. remove default value in batch.split() and add merge_last argument (#185) 11. improve nstep efficiency 12. add max_batchsize in onpolicy algorithms 13. potential bugfix for subproc.wait 14. fix RecurrentActorProb 15. improve the code-coverage (from 90% to 95%) and remove the dead code 16. fix some incorrect type annotation The above improvement also increases the training FPS: on my computer, the previous version is only ~1800 FPS and after that, it can reach ~2050 (faster than v0.2.4.post1).
100 lines
3.6 KiB
Python
100 lines
3.6 KiB
Python
import gym
|
|
import time
|
|
import random
|
|
import numpy as np
|
|
from gym.spaces import Discrete, MultiDiscrete, Box, Dict, Tuple
|
|
|
|
|
|
class MyTestEnv(gym.Env):
|
|
"""This is a "going right" task. The task is to go right ``size`` steps.
|
|
"""
|
|
|
|
def __init__(self, size, sleep=0, dict_state=False, recurse_state=False,
|
|
ma_rew=0, multidiscrete_action=False, random_sleep=False):
|
|
assert not (
|
|
dict_state and recurse_state), \
|
|
"dict_state and recurse_state cannot both be true"
|
|
self.size = size
|
|
self.sleep = sleep
|
|
self.random_sleep = random_sleep
|
|
self.dict_state = dict_state
|
|
self.recurse_state = recurse_state
|
|
self.ma_rew = ma_rew
|
|
self._md_action = multidiscrete_action
|
|
# how many steps this env has stepped
|
|
self.steps = 0
|
|
if dict_state:
|
|
self.observation_space = Dict(
|
|
{"index": Box(shape=(1, ), low=0, high=size - 1),
|
|
"rand": Box(shape=(1,), low=0, high=1, dtype=np.float64)})
|
|
elif recurse_state:
|
|
self.observation_space = Dict(
|
|
{"index": Box(shape=(1, ), low=0, high=size - 1),
|
|
"dict": Dict({
|
|
"tuple": Tuple((Discrete(2), Box(shape=(2,),
|
|
low=0, high=1, dtype=np.float64))),
|
|
"rand": Box(shape=(1, 2), low=0, high=1,
|
|
dtype=np.float64)})
|
|
})
|
|
else:
|
|
self.observation_space = Box(shape=(1, ), low=0, high=size - 1)
|
|
if multidiscrete_action:
|
|
self.action_space = MultiDiscrete([2, 2])
|
|
else:
|
|
self.action_space = Discrete(2)
|
|
self.done = False
|
|
self.index = 0
|
|
self.seed()
|
|
|
|
def seed(self, seed=0):
|
|
self.rng = np.random.RandomState(seed)
|
|
return [seed]
|
|
|
|
def reset(self, state=0):
|
|
self.done = False
|
|
self.index = state
|
|
return self._get_state()
|
|
|
|
def _get_reward(self):
|
|
"""Generate a non-scalar reward if ma_rew is True."""
|
|
x = int(self.done)
|
|
if self.ma_rew > 0:
|
|
return [x] * self.ma_rew
|
|
return x
|
|
|
|
def _get_state(self):
|
|
"""Generate state(observation) of MyTestEnv"""
|
|
if self.dict_state:
|
|
return {'index': np.array([self.index], dtype=np.float32),
|
|
'rand': self.rng.rand(1)}
|
|
elif self.recurse_state:
|
|
return {'index': np.array([self.index], dtype=np.float32),
|
|
'dict': {"tuple": (np.array([1],
|
|
dtype=np.int64), self.rng.rand(2)),
|
|
"rand": self.rng.rand(1, 2)}}
|
|
else:
|
|
return np.array([self.index], dtype=np.float32)
|
|
|
|
def step(self, action):
|
|
self.steps += 1
|
|
if self._md_action:
|
|
action = action[0]
|
|
if self.done:
|
|
raise ValueError('step after done !!!')
|
|
if self.sleep > 0:
|
|
sleep_time = random.random() if self.random_sleep else 1
|
|
sleep_time *= self.sleep
|
|
time.sleep(sleep_time)
|
|
if self.index == self.size:
|
|
self.done = True
|
|
return self._get_state(), self._get_reward(), self.done, {}
|
|
if action == 0:
|
|
self.index = max(self.index - 1, 0)
|
|
return self._get_state(), self._get_reward(), self.done, \
|
|
{'key': 1, 'env': self} if self.dict_state else {}
|
|
elif action == 1:
|
|
self.index += 1
|
|
self.done = self.index == self.size
|
|
return self._get_state(), self._get_reward(), \
|
|
self.done, {'key': 1, 'env': self}
|