tejank10 / AlphaGo.jl

AlphaGo Zero implementation using Flux.jl
https://fluxml.ai/experiments/go
Other
73 stars 14 forks source link

Interface #2

Open MikeInnes opened 6 years ago

MikeInnes commented 6 years ago

Right now this package relies pretty heavily on scripts in src which take command line arguments. Ideally, these should all become normal function calls. Using the package might look like this:

using AlphaGo

model = AlphaGo.Model(...) # should download trained weights by default
# can also call gpu(model) here

AlphaGo.train!(model, ...) # optionally train

AlphaGo.play(model, model) # play the model against itself in the terminal/WebIO UI

AlphaGo.play(model) # play against the user

This is just an example, it might not be exactly what makes sense, but the key thing is being able to get going really quickly in the repl.

Separately, I wonder if this package should be called AlphaZero, since in future it might have chess etc.

vijendransridhar commented 5 years ago

I would like to contribute for chess, how do I take it forward?