Flux Chain function

Specific Domains Machine Learning
1

Why does the Chain function not warn if the output of one layer is not compatible with the imput of the next?

2

Chain can take arbitrary functions. Given that, how would you check the inputs/outputs of the following are compatible?

f1(x) = sum(x, dims=1)
function f2(x)
  ndims(x) == 1 && error("too few dims")
  return x .+ 1
end

c = Chain(f1, f2)

Without actually calling c, there is no way to know whether you’ll get a size mismatch. Thankfully, Flux provides a way to do this and get inferred output sizes without having to run a full forward pass of your model. See outputsize.