onehotbatch with CuArray

[CarloLucibello]

Not a great solution but this used to error before (when scalar indexing is deactivated) due to the following operation:

function _onehotbatch(data, labels)
      indices = UInt32[something(_findval(i, labels), 0) for i in data]
      ...

so I guess going to cpu and back to gpu is strictly better.

[ToucheSir]

I can't think of a great way of doing this either. If anyone does, we can test it out in OneHotArrays.jl first.

[YichengDWu]

What about this?

 onehotbatch(data::CuArray,labels)= cu(labels).==reshape(data, 1,size(data)...)

soure: https://discourse.julialang.org/t/all-the-ways-to-do-one-hot-encoding/64807/9

[ToucheSir]

That defeats the purpose of having a one-hot array type in the first place. If you can find a way to preserve the outer wrapper and use less memory than a dense matrix, we would be very interested.

[YichengDWu]

I will leave it to the smart ones