Diffenence against Rlax

Question

Diffenence against Rlax

JiahaoYao opened this issue 4 years ago · comments

Jimmy Yao commented 4 years ago

Hi, there are currently two lib for RL using Jax, Dopamine and Rlax. Might I ask what is the difference? Thanks!

Jimmy Yao commented 4 years ago

Thanks!

Pablo Samuel Castro · Answer 1 · Mon Oct 19 2020 21:16:04 GMT+0800 (China Standard Time)

they are similar libraries but different and independent implementations. i don't know too many details about relax, but from what i understand it is mostly meant to be a collection of building blocks which you can use to construct agents. on the other hand, in dopamine we provide monolithic agents with tunable parameters that can modify their behaviour (for example turning prioritized replay on or off in the rainbow agent). hope this helps,

…

-psc

On Sat, Oct 17, 2020 at 11:30 PM Jimmy Yao ***@***.***> wrote: Hi, there are currently two lib for RL using Jax, Dopamine and Rlax. Might I ask what is the difference? Thanks! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#155>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AE3CCMNGUUROGTI6NGY3K4LSLJOMBANCNFSM4SUZJR4Q> .

8bitmp3 · Answer 2 · Mon Oct 19 2020 21:39:02 GMT+0800 (China Standard Time)

Thanks @psc-g and great episode on talkRL https://www.talkrl.com/episodes/pablo-samuel-castro 👍

... knowledge on distributional RL, bisimulation, the Dopamine RL Framework, TF-Agents, and much more!

I understand that, unlike Dopamine (Google Research), RLax has DeepMind-built dependencies like Haiku (similar to Sonnet in TensorFlow) and Optax.

Do you know if both libraries are meant for fast prototyping? The Dopamine example on CartPole with DQN and C51 appears to have less imports from the get-go.

Pablo Samuel Castro · Answer 3 · Mon Oct 19 2020 23:10:41 GMT+0800 (China Standard Time)

i can say that we built dopamine for fast prototyping, but i can't speak for relax. you can ask the people on that repo :)

…

On Mon, Oct 19, 2020 at 9:39 AM 8bitmp3 ***@***.***> wrote: Thanks @psc-g <https://github.com/psc-g> and great episode on talkRL https://www.talkrl.com/episodes/pablo-samuel-castro 👍 ... knowledge on distributional RL, bisimulation, the Dopamine RL Framework, TF-Agents, and much more! I understand that, unlike Dopamine (Google Research), RLax has DeepMind-built dependencies like Haiku (similar to Sonnet in TensorFlow) and Optax. Do you know if both libraries are meant for fast prototyping? The Dopamine example on CartPole with DQN and C51 appears to have less imports from the get-go. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#155 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AE3CCMPVSUBKNTIA6353PU3SLQ6QTANCNFSM4SUZJR4Q> .