-
For any new games which I could add, is there any min/max range for the reward to be followed? For the gym Atari games, I am not aware of any clipping.
Specifically
`def step(self, action):
…
whyrv updated
4 years ago
-
The [MuZero ray code](https://github.com/werner-duvaud/muzero-general/) seems to want gym environments to be pickleable with the cloudpickle module, but currently one gets:
```
>>> import gym
>>>…
-
https://github.com/werner-duvaud/muzero-general/blob/091897759f2f6748f651c289405afe8a1dfa1101/trainer.py#L97
-
Problem: https://docs.google.com/presentation/d/1iYfbMrMZUn88cet36qe0h-2oHG9HpbFfEZHTVp4xlik/edit
Possible solution:
1. train the D with sparsity loss on the model + reconstruction loss where the …
-
https://github.com/werner-duvaud/muzero-general/blob/4d541626a2d1ace2e3bdf30d25d9a843e4cb613c/replay_buffer.py#L79
Shouldn't this be:
`position_probs = numpy.array(game_history.priorities[:-1]) / …
-
I am running your code with the game connect4. It is already doing more than 250k steps. but the reward value is declining and approaching the bottom.
-
https://github.com/werner-duvaud/muzero-general/blob/ecca75c8d5893048b0acc6c5897a504c6334b871/models.py#L137
Since you are predicting a distributions, the first reward should be scalar_to_support(0…
-
Do you have a plain to implemention muzero general?
-
There was a question on the Disqus regarding finding the `best_action`; I'd like to surface it here to get more attention:
```python
args1 = dotdict({'numMCTSSims': 50, 'cpuct':1.0})
mcts1 = MCTS…
-
In Appendix G of the muzero paper, they define the priority of a sample as p_i = | nu_i - z_i |, and write "nu is the search value and z the observed n-step return." (I'll use "nu" in place of ν for …