rl
rl copied to clipboard
pytorch
A modular, primitive-first, python-first PyTorch library for Reinforcement Learning.
## Description When setting a ParallelEnv in a remote data collector and not embedding the env maker in an EnvCreator, an error is raised. This PR solves this issue.
This issue is a list of contributions requests from the community. ## How to use this list If you're willing to contribute to the library, have a look at the...
### 🚀 The feature, motivation and pitch I am totally unable to create a `EnvBase` subclass, where the `*_spec` attribute have complex shapes. For example, I have a state with...
Loss functions accept non-tensordict data thanks to the tensordict.dispatch decorator. However, we do not provide an overloaded forward, which could be useful to let users know about the typical signature...
We project on using @tensorclass to represent losses. The advantage of tensorclass for losses instead of tensordict is that it will help us use all the features of tensordict while...
## Motivation In many scenarios we need to perform a step only on a subset of batched envs. This includes collecting a complete trajectory for many envs when they end...
Value function mixer using a GNN from https://arxiv.org/abs/2205.13005 missing: - [ ] docs - [ ] tests implemented by @acciorocketships
## Motivation It would be great to have an MCTS and Alphazero implementation, including other model-based RL for benchmarking and comparison. ## Solution I can write a loss function of...
## Motivation When creating a custom environment, deciding the correct TensorSpec (TS) to use is not straightforward. Some TS are never used, some look very similar to each other and...
## Description In this PR, I propose a script to run all our benchmarks before the release. cc @matteobettini @albertbou92 @BY571 @giadefa TODO: - [ ] make sure Wandb logging...