-
### Description
I'm writing an [open source MuZero](https://github.com/wbrenton/OpenZero.git) for continuous action spaces with jax, haiku, mctx, and brax.
As training progresses the wall time pe…
-
When I try to generate audio using batch generation on long text, it drops some words/sentences in audio. like they are never there.
the code I am using is given below
```
def sentense_tokeni…
-
I've seen others reporting that the `muzero_policy` is slow and I've run into this problem myself so I wanted to add a bit more information. I'm not expecting a solution to this problem, but it might …
-
How could I use the [2020 humanoid_CMU torques, kp and damping values](https://github.com/deepmind/dm_control/blob/4e1a35595124742015ae0c7a829e099a5aa100f5/dm_control/locomotion/walkers/cmu_humanoid.p…
-
I tried to implement the cube solver. Everything works but the heuristic function is taking to long to create the heuristic database for a depth of 20. I was only able to generate a database with dept…
-
The current AtariWrapper by default has `terminate_on_life_loss` set to True. This goes against the recommendations of Revisiting the Arcade Learning Environment (https://arxiv.org/pdf/1709.06009.pdf)…
-
A person told me that the bottleneck can improve the 30% performance without losing precision. But Kata Go uses the nest bottleneck instead the bottleneck. Is the nest bottleneck significantly better …
-
Hello,
I've been having issues with doing self-play on GPU, and after about a week of experimentation I've realized that it is necessary to use this option if I want to train a model on my custom g…
-
- https://arxiv.org/abs/2111.00210
- NeurIPS 2021
強化学習は、多くのアプリケーションで大きな成功を収めています。
しかし、サンプル効率は依然として重要な課題であり、多くの手法では学習に数百万(あるいは数十億)の環境ステップを必要とします。
最近では、サンプル効率の良い画像ベースのRLアルゴリズムが大きく進歩しているが、Atariゲームベ…
e4exp updated
2 years ago
-
You can get a nice speed boost for pure Python MCTS by vectorizing the logic for running N tree searches in parallel. The basic idea is to store the data describing, e.g., parent and child pointers de…