rllib.js - Reinforcement learning library with JavaScript, Three.js, Tensorflow.js

[Игорь Токарев]

Hello, guys! It seems, like people want make RL with JS) I have been started rllib.js few times ago, and there are many to work.

At now, there DQN and PPO implementations in progress. After implementing it will be improve.

In plan:

- build envs that will works in browser in exec mode and node.js-gpu in train.

- implement all another algorithms: evolution strategies, AC, DDPG and many new.

If you want to contribute and develop lib, you are welcome! Write me! Let's make cool RL stuff)

https://github.com/polyzer/rllib.js

[Jason Mayes]

Super cool! Excited to see where this goes with time. I'm starting to see a more folk ask about RL in JS these days. Someone reached out recently about wanting to convert TF-Agents to TFJS along with support for VR environments etc too. Seems like some overlap with you too.

[Игорь Токарев]

Hello, Jason!

Yes, implementing all popular agents and many envs in progress)

So, rllib.js - was conceived as an alternative for tf-agents for JS))

At this moment I have implemented PPO. Now there would be one more cool 3d env.

Then in roadmap:

 -- improve api;

 -- add Node.js for training/saving models;

 -- adding DoubleDQN, 

 -- DDPG, A2C.

General tasks:

 -- using gym-like envs api;

 -- fast calculations of advanced RL algorithms.

 -- adding simple ways to use both browser and node.js for training/exec.

 -- MORE and MORE cool envs in: physics simulations, finance, 3d/2d, VR.

And, if I could help with implementing tf-agents, that would be cool!

среда, 2 сентября 2020 г. в 01:53:45 UTC+5, Jason Mayes:

[Jason Mayes]

PS you should totally join this group of folk on Discord who are interested in similar goals:  https://discord.gg/3KbgmcW (see other thread on this forum https://groups.google.com/a/tensorflow.org/g/tfjs/c/9AfY5RkZ40U  ) - I think there is some good overlap here and may be room to collaborate.

--

•  Jason Mayes   •  Senior Developer Advocate, TensorFlow.js   •  Google   •   Follow my discoveries or contribute your own

This email may be confidential and privileged. If you received this communication by mistake, please don't forward it to anyone else, please erase all copies and attachments, and please let me know that it has gone to the wrong person.