Implementation of Deep Reinforcement Learning Algorithms in Swift for TensorFlow

[D. Suresh]

Hello all, I just learnt how to use swift for Tensorflow. I have implemented various Deep RL algorithms using TensorFlow. 

I am trying to implement some RL algorithms in swift for Tensorflow. 

Can anyone help where should I head to, or any other word of advice.

Thanks in advance.

[Steven Prichard]

Do you have a public repo to share? Maybe looking at the code could help get a clearer picture. Of the Deep RL algorithms you have implemented, are they implemented in Python? If so, you could leverage the interoperability of Swift for Tensorflow and call your classes/functions from Swift.

My next question would be, what is it you want to get out of this? Are you attempting to re-implement your algorithms in Swift? Or just leverage your existing implementation, but from Swift?

Cheers,

Steven Prichard

--
You received this message because you are subscribed to the Google Groups "Swift for TensorFlow" group.
To unsubscribe from this group and stop receiving emails from it, send an email to swift+un...@tensorflow.org.

[D. Suresh]

Hello Steven, sorry I don't have any public repo. But i want to re-implement algos such as deep Q network, vanilla policy gradient etc in swift.

Also I want to implement them in such way that they come use for others swift - tensorflow users as well :).

Steven Prichard

To unsubscribe from this group and stop receiving emails from it, send an email to sw...@tensorflow.org.

[D. Suresh]

Hello all, can anyone give me some advice here?

Thanks in advance.

[Brett Koonce]

There’s some swift reinforcement learning models in https://github.com/tensorflow/swift-models (Gym directory) you should get working on your machine.  After that, you could port some other models/agents from https://github.com/openai/gym over, and then make a pull request!  

-Brett

P.S. OpenAI has a nice collection of rl papers here: https://spinningup.openai.com/en/latest/index.html

To unsubscribe from this group and stop receiving emails from it, send an email to swift+un...@tensorflow.org.

[Anthony Platanios]

FYI, I'm also working on a general RL library for Swift, along with a Gym equivalent including Gym Retro support. It's very much work-in-progress but you can find the code in the "gym" branch of this repository: https://github.com/eaplatanios/retro-swift/tree/gym 

-Anthony