League Slime Volleyball Updates

[zxzzz0]

Hello DI-engine community,

Good work for the framework. Open a thread to track PR 23.

Background

Today is Sep 17, 2021. 1.5 months ago, on Aug 2, 2021 this issue was raised up trying to add a new env into di-zoo. Initial version of PR 23 was created as early as Aug 12, 2021. However there might be some issues happened. Since then, every time people checked this branch they didn't get too much useful information in the past 5 weeks.

Plan

We truly believe that adding a new env into di-zoo is an easy effort thanks to the good design of di-engine. In the future, let's post latest bugs/problems here to bring more transparency in this PR and iterate faster.

Best
zxzzz0

screenshot

[zxzzz0]

@PaParaZz1 Would you take a look at this next week?

[PaParaZz1]

We will continue to update this PR23 since 10.1.

[zxzzz0]

We were checking the code for doing mutation. However here it looks like the mutation is a no-op. Generally speaking, in mutation we would like the agent to be reset to one of the following randomly:
(1) A pre-trained supervised model
(2) Past historical model
(3) Pure random model

[PaParaZz1]

random uniformly from all the possible model?

[zxzzz0]

It will be a smart selection from past models. (2) and (3) are optional, which can be ignored for now.
But (1) is needed and the code there seems to be no-op. Why?

[PaParaZz1]

We will continue to league training in this issue #90.

[zxzzz0]

@PaParaZz1 Would you have time to take a look at it this week? #90 Can simply copy and paste the code from league demo.