Jonathan Laurent

Now I am getting this: ``` MethodError: no method matching bytes(::Int64) Closest candidates are: bytes(::KnetArray{T,N} where N) where T at /home/jonathan/.julia/packages/Knet/bTNMd/src/conv.jl:505 Stacktrace: [1] conv4(::KnetArray{Float32,4}, ::KnetArray{Float32,4}; handle::Ptr{Nothing}, alpha::Int64, o::Base.Iterators.Pairs{Symbol,Tuple{Int64,Int64},Tuple{Symbol},NamedTuple{(:padding,),Tuple{Tuple{Int64,Int64}}}}) at /home/jonathan/.julia/packages/Knet/bTNMd/src/conv.jl:41...

Ok, so now I am getting: ``` cudnnConvolutionForward: 7: CUDNN_STATUS_MAPPING_ERROR Stacktrace: [1] error(::String) at ./error.jl:33 [2] macro expansion at /home/jonathan-laurent/.julia/packages/Knet/bTNMd/src/gpu.jl:34 [inlined] [3] conv4(::KnetArray{Float32,4}, ::KnetArray{Float32,4}; handle::Ptr{Nothing}, alpha::Int64, o::Base.Iterators.Pairs{Symbol,Tuple{Int64,Int64},Tuple{Symbol},NamedTuple{(:padding,),Tuple{Tuple{Int64,Int64}}}}) at /home/jonathan-laurent/.julia/packages/Knet/bTNMd/src/conv.jl:41 [4]...

I am not using multiple GPUs indeed. In fact, in the program that exhibits the bug above, all calls to the GPU happen within a single thread. Note that this...

It depends on the size of the network you are using. With the default hyperparameters that we set for many games, training on CPU would be very slow (up to...

The only way the MCTS implementation prevents taking an invalid action is by setting a qvalue of -oo. Isn't it possible for `gumbel_explore` to take an invalid actions if `num_considered_actions=4`...

AlphaZero.jl does not have multiple workers explore the same MCTS tree and therefore there is no need for a virtual loss. Instead, many games are simulated in parallel during self-play...

This is true and as a matter of fact, matches would be quite slow with the current implementation. This is on my TODO list though.

> will it ever be implemented? Doing so is not my priority right now but I would be happy to welcome contributions here. Note that I am working on a...

Previous versions will still be accessible from git or the package manager. But the new version will break compatibility with existing code indeed.

See https://github.com/jonathan-laurent/AlphaZero.jl/tree/master/redesign.