Michael Abbott
Michael Abbott
Can we farm more of this out to Functors / Optimisers? Instead of building an IdSet by hand, let Functors cache things. Then this will inherit its understanding of Adjoint...
Functors uses a cache which should detect such sharing. It's a little smarter than just using `objectid`, so as not to catch immutable objects which are accidentally `===`. ```julia julia>...
If revived this should probably follow https://github.com/FluxML/Zygote.jl/pull/1415
On latest Flux, using new-style training with `setup`, something like `dec = Dense(transpose(encoder.weight))` should just work. It will see through the `transpose` and notice that the same array appears twice....
How easy would it be to make this friendlier? ```julia julia> using Flux, Metal julia> x = rand(Float32, 3) |> Flux.gpu ┌ Info: The CUDA functionality is being called but...
We should think more about the right signatures. Matching the categorical functions in `Flux.Losses` is one place to start, and suggests: ``` accuracy(yhat::AbstractMatrix, y::OneHotMatrix) # like crossentropy accuracy(yhat::AbstractMatrix, y::AbstractVector{Bool}) #...
The error is because `conv` is getting a mix of CuArray and Array input: ``` [5] conv_im2col!( y::SubArray{Float32, 5, CUDA.CuArray{Float32, 5, CUDA.Mem.DeviceBuffer}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, UnitRange{Int64}, Base.Slice{Base.OneTo{Int64}}}, false}, x::SubArray{Float32, 5,...
Yes that looks right to me. But I didn't run it, I hope it works!
Either this or #2139 is going to be a fun rebase, sadly. Are there any other PRs similarly affected? That (and git blame) are the objections to fixing white-space extensively,...
Welcome, and glad you persisted! I made these examples recently. The goals I suppose were: * Have *something* you can copy & run off the readme, previously there was nothing....