gyx icon indicating copy to clipboard operation
gyx copied to clipboard

Reinforcement Learning environment for Elixir

Results 29 gyx issues
Sort by recently updated
recently updated
newest added

The `Env.step/1` callback utilizes `Gyx.Core.Spaces.contains?` for checking if the action is valid. Make a simple test on a dummy environment giving a valid and invalid action. Note: The environment process...

enhancement
help wanted

At this moment, for developing purposes, each module is started by the main application as a single process. This is cool, as tests can be done right away, but having...

enhancement
help wanted
high priority

Microsoft has released [TextWorld](https://towardsdatascience.com/whats-new-in-deep-learning-research-microsoft-s-textworld-is-the-openai-gym-of-language-learning-e726d64eb8a3), a tool for creating text-based environments for training conversational gents. Environments can be created with `tw-make` giving `.ulx` files. A TextWorld environment must be registered as...

enhancement
help wanted

Implement a module with [`Env`](https://github.com/doctorcorral/gyx/blob/master/lib/core/env.ex) behaviour similar to [`Gym.Environment`](https://github.com/doctorcorral/gyx/blob/master/lib/Gym/environment.ex) based on [Pyrlang](https://github.com/Pyrlang/Pyrlang) project for interacting with Python [Gym](https://gym.openai.com/).

There is alreay one [Buffer Memory implementation based on ETS](https://github.com/doctorcorral/gyx/blob/master/lib/experience/replay_buffer_ets.ex). This implementation uses the `Gyx.Core.ReplayMemory` behaviour. The issue with this first approach, is that not Erlang based systems might have...

The goal is to have a simple mix tooling for executing a training on a given environment, with any given learning agent. It would be possible to define the number...

The multiple agent-environment copies must give their experiences to a centralized [Replay Buffer](https://github.com/doctorcorral/gyx/blob/master/lib/experience/replay_buffer_ets.ex)

enhancement

Currently, `Trainer.Single.trainer/2` just runs a number of episodes, making agent interact with the environment and getting experiences stored in `%Trainer.Single.experiences`. These experiences must be fed to agent learning method.

enhancement

It must be straightforward to create Gyx releases with [Distillery](https://github.com/bitwalker/distillery).