Acer

[mirraaj]

This is implementation of ACER algorithm using keras-rl.

Paper : https://arxiv.org/pdf/1611.01224.pdf

[mirraaj]

@matthiasplappert Current implementation supports single agent learning for ACER. Do you think we should add synchronous agent learning approach as implemented by OpenAI gym?

@RaphaelMeudec Let me know if your views as well.

[AloshkaD]

I guess we need more reviewers. I can volunteer to review and test.

[mirraaj]

@AloshkaD the current implementation supports simple toy examples. It will really be helpful if you volunteer to review this PR. You could also work to add support for Atari games in ACER. Let me know if you need any explanation in the code. Thanks