Feature request: dims keyword for norm

[wesselb]

In some cases, it would be really handy to be able to do this:

julia> a = randn(5, 100);

julia> norm(a, dims=1)  # Doesn't work
ERROR: MethodError: no method matching norm(::Matrix{Float64}; dims=2)

julia> sqrt.(sum(abs2, a, dims=1))  # Intended result (in this case, with `p=2`)
1×100 Matrix{Float64}:
 1.26534  0.763155  2.55044  2.38645  1.69696  …  1.71206  2.61463  2.81637  1.82003  1.69302

As a more advanced example, for a tensor a of shape (m, n, k), norm(a, dims=(2, 3)) would compute norm(a[i, :, :]) for all i and return something of shape (m, 1, 1).

[martinholters]

julia> mapslices(norm, a; dims=1)
1×100 Matrix{Float64}:
 1.89751  3.06913  2.61767  3.51301  …  1.91144  0.557424  1.73838

Or mapslices(norm, a; dims=(2,3)) for the second case.

Not sure we need to add this to norm directly?

[mcabbott]

Duplicate of JuliaLang/LinearAlgebra.jl#697, BTW.