-
### Scenario
I'd like to reuse the replay_buffer from an earlier run, but on a slightly different model. Is there anyway to do this correctly? I've tried copying just the replay_buffer.pkl file to …
-
Recently while training I started getting the following errors.
```
2021-01-11 16:34:00,823.ERROR worker.py:980 -- Possible unhandled error from worker: ray::ReplayBuffer.get_batch() (pid=31490…
-
While attempting to train with resnet, I'm always getting ridiculously high loss rates. Eventually they go so high, the update_weights process goes out of bounds and crashes. What factors can be con…
-
I saw your muzero-general repo. It had an open_spiel game wrapper. Can you please add it here too?
Keras is kinda slow.
Also all games result in a draw. So no new model is selected.
-
I've performed several tests (different games) using the initial configuration, and i always get mean values (Total reward/mean values [plot 2 / tictactoe]) alternating around 0. So there seems to be …
-
## Question or maybe Enhancement
I'm missing a feature to scale the gradient for backward pass (as e.g. used in MuZero) ... something like
tensor * scale + stop_gradient(tensor) * (1 - scale)
I'm…
-
Hello,
I just tried the colab notebook (copied from https://github.com/werner-duvaud/muzero-general/blob/master/notebook.ipynb). After 13 minutes it still got a low score (see below). Other framew…
-
Dear Mr. Duvaud,
Upon cloning your repository to ".../Documents/AI", adding swig-4.0.2 (https://sourceforge.net/projects/swig/files/swigwin/swigwin-4.0.2/swigwin-4.0.2.zip/download?use_mirror=pho…
-
Mr. Varty,
Would be willing to provide scripts or ideas for self-play, console-IO play, and/or 2D games?
Tom Lever
-
May be change
game_history.action_history.append( numpy.random.choice(self.game.legal_actions()) )
?
nnnet updated
4 years ago