randnlike(A), randlike(A) similar to pytorch

[isentropic]

Not sure if this is new thing or not but the only relevant disc. so far is here
Why don't we have randnlike(A), randlike(A) https://pytorch.org/docs/stable/generated/torch.randn_like.html
Very convenient for ML.

rand(size(A)) doesn't cut it as it doesn't propagate types. The equivalent would be, rand(eltype(A), size(A)) which is a little verbose

[oscardssmith]

our idiom for that would be rand!(similar(A)) which is only 2 characters longer

[isentropic]

Then wouldn't it be useful just to introduce?

randlike(a) = rand!(similar(a))
randnlike(a) = randn!(similar(a))

rand!(similar(A)) is just not discoverable through repl, but such construct is widely used in ML particularly in neural networks

[oscardssmith]

imo this would be a bad function because it doesn't compose well. lots of ml applications will want this functionality but for different random distributions, so it's better to use rand! which allows user defined distributions than make a new name per distribution for this composition.

There also isn't any reason this has to be in Base. New names like this can always be exported by a package.