Mava icon indicating copy to clipboard operation
Mava copied to clipboard

Published 20 hours ago • verified publisher icon instadeepai

Feature/make continuous ppo

Open siddarthsingh1 opened this issue 1 year ago • 1 comments

What?

Added continuous network option for IPPO.

Why?

Extending functionality for IPPO implementation.

How?

Added function to create IPPO network with a clipped Gaussian function instead of a categorical policy.

Extra

Closes issue #631 -~~The get_action(...) function needs an if statement to check the network type in use since the discrete actions are of type int64 and continuous actions are of type float32~~

  • ~~Need to make custom networks to clip distribution. Functionality is present in acme.tf but not acme.jax~~
  • NaN's generated in Multiwalker

siddarthsingh1 avatar Jul 19 '22 10:07 siddarthsingh1

@siddarthsingh1 please create and link an issue for this within the correct milestone (IPPO)

mmorris44 avatar Jul 19 '22 14:07 mmorris44

We are closing this PR as we are focussing on the JAX release. We will re-open it when needed again in the future.

DriesSmit avatar Sep 07 '22 08:09 DriesSmit