FluxML / model-zoo

Please do not feed the models
https://fluxml.ai/
Other
911 stars 333 forks source link

GSoC Proposal #28

Closed tejank10 closed 6 years ago

tejank10 commented 6 years ago

I am interested in contributing to the model zoo of FluxML as part of GSoC'18. Below are the models which I want to implement over the course of summer:

@MikeInnes Does it sound like a good project to work on?

MikeInnes commented 6 years ago

Sounds good – nice and ambitious :)

One thing that would be really interesting is applying AlphaGo or other RL to other games, e.g. chess, pong or tetris. We could possible then get a fun play-against-the-AI demo going via FluxJS.

tejank10 commented 6 years ago

Thanks :). Playing games against AI made using Flux should be really fun! I was wondering how should I go about prioritizing the models mentioned above? If we want to deploy the human vs AI game using FluxJS, perhaps I shall give more priority to the RL models.

MikeInnes commented 6 years ago

Yeah, let's lean towards the RL stuff for now, and we can branch out if it makes sense.

Worth checking out Chess.jl, haven't looked at it closely but it might be useful as a starting point. Or it should be pretty simple to build a pong simulator or similar.

tejank10 commented 6 years ago

Thanks for Chess.jl! I'll add the pong simulator as a stretch goal. Perhaps it could be incorporated into Reinforce.jl environment. I wanted to know whether explaining the architectures I wish to implement in the implementation part of the project proposal would do?

EDIT: Got the answer in the Slack forum, Cheers!

tejank10 commented 6 years ago

@MikeInnes Here is the initial draft of my proposal: GSoC Proposal. I would be grateful if you could provide your valuable inputs over it. Thanks! Looking forward to your comments. :)

MikeInnes commented 6 years ago

GSoC is in progress now, so closing these issues.

xysun commented 5 years ago

Is there any further update regarding this? I noticed in README it mentioned RL algorithms but couldn't find them.

tejank10 commented 5 years ago

The RL algorithms mentioned are made as a separate repo, https://github.com/tejank10/Flux-baselines