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).
69 lines
2.1 KiB
Python
69 lines
2.1 KiB
Python
import torch
|
|
import numpy as np
|
|
|
|
from tianshou.utils import MovAvg
|
|
from tianshou.exploration import GaussianNoise, OUNoise
|
|
from tianshou.utils.net.common import Net
|
|
from tianshou.utils.net.discrete import DQN
|
|
from tianshou.utils.net.continuous import RecurrentActorProb, RecurrentCritic
|
|
|
|
|
|
def test_noise():
|
|
noise = GaussianNoise()
|
|
size = (3, 4, 5)
|
|
assert np.allclose(noise(size).shape, size)
|
|
noise = OUNoise()
|
|
noise.reset()
|
|
assert np.allclose(noise(size).shape, size)
|
|
|
|
|
|
def test_moving_average():
|
|
stat = MovAvg(10)
|
|
assert np.allclose(stat.get(), 0)
|
|
assert np.allclose(stat.mean(), 0)
|
|
assert np.allclose(stat.std() ** 2, 0)
|
|
stat.add(torch.tensor([1]))
|
|
stat.add(np.array([2]))
|
|
stat.add([3, 4])
|
|
stat.add(5.)
|
|
assert np.allclose(stat.get(), 3)
|
|
assert np.allclose(stat.mean(), 3)
|
|
assert np.allclose(stat.std() ** 2, 2)
|
|
|
|
|
|
def test_net():
|
|
# here test the networks that does not appear in the other script
|
|
bsz = 64
|
|
# common net
|
|
state_shape = (10, 2)
|
|
action_shape = (5, )
|
|
data = torch.rand([bsz, *state_shape])
|
|
expect_output_shape = [bsz, *action_shape]
|
|
net = Net(3, state_shape, action_shape, norm_layer=torch.nn.LayerNorm)
|
|
assert list(net(data)[0].shape) == expect_output_shape
|
|
net = Net(3, state_shape, action_shape, dueling=(2, 2))
|
|
assert list(net(data)[0].shape) == expect_output_shape
|
|
# recurrent actor/critic
|
|
data = data.flatten(1)
|
|
net = RecurrentActorProb(3, state_shape, action_shape)
|
|
mu, sigma = net(data)[0]
|
|
assert mu.shape == sigma.shape
|
|
assert list(mu.shape) == [bsz, 5]
|
|
net = RecurrentCritic(3, state_shape, action_shape)
|
|
data = torch.rand([bsz, 8, np.prod(state_shape)])
|
|
act = torch.rand(expect_output_shape)
|
|
assert list(net(data, act).shape) == [bsz, 1]
|
|
# DQN
|
|
state_shape = (4, 84, 84)
|
|
action_shape = (6, )
|
|
data = np.random.rand(bsz, *state_shape)
|
|
expect_output_shape = [bsz, *action_shape]
|
|
net = DQN(*state_shape, action_shape)
|
|
assert list(net(data)[0].shape) == expect_output_shape
|
|
|
|
|
|
if __name__ == '__main__':
|
|
test_noise()
|
|
test_moving_average()
|
|
test_net()
|