DQN playing QbertNoFrameskip-v4 from https://github.com/sgoodfriend/rl-algo-impls/tree/e8bc541d8b5e67bb4d3f2075282463fb61f5f2c6
f540c0e
import gym | |
import torch as th | |
import torch.nn as nn | |
from gym.spaces import Discrete | |
from typing import Optional, Sequence, Type | |
from shared.module.feature_extractor import FeatureExtractor | |
from shared.module.module import mlp | |
class QNetwork(nn.Module): | |
def __init__( | |
self, | |
observation_space: gym.Space, | |
action_space: gym.Space, | |
hidden_sizes: Sequence[int] = [], | |
activation: Type[nn.Module] = nn.ReLU, # Used by stable-baselines3 | |
cnn_feature_dim: int = 512, | |
cnn_style: str = "nature", | |
cnn_layers_init_orthogonal: Optional[bool] = None, | |
) -> None: | |
super().__init__() | |
assert isinstance(action_space, Discrete) | |
self._feature_extractor = FeatureExtractor( | |
observation_space, | |
activation, | |
cnn_feature_dim=cnn_feature_dim, | |
cnn_style=cnn_style, | |
cnn_layers_init_orthogonal=cnn_layers_init_orthogonal, | |
) | |
layer_sizes = ( | |
(self._feature_extractor.out_dim,) + tuple(hidden_sizes) + (action_space.n,) | |
) | |
self._fc = mlp(layer_sizes, activation) | |
def forward(self, obs: th.Tensor) -> th.Tensor: | |
x = self._feature_extractor(obs) | |
return self._fc(x) | |