Use vector of indexes

[jariji]

I wanted to reduce into a 2-vector by using the indexes stored in ixs. Currently it seems to ignore ixs. What do you think about this idea?

using TensorCast
vs = [[10,20,30,40,50,60] .+ rand(6) for _ in 1:10]
ixs = [1,2,1,2,1,2]
@reduce v[ixs[j]] := mean(i) vs[i][j]

[mcabbott]

This should probably be an error. Right now it does not notice the indexing on the left, and treats it like v[j]:

julia> v ≈ @reduce v2[j] := mean(i) vs[i][j]
true

julia> @pretty @reduce v[ixs[j]] := mean(i) vs[i][j]
begin
    @boundscheck vs isa Tuple || (ndims(vs) == 1 || throw(ArgumentError("expected a vector or tuple vs[i]")))
    local dragonfly = transmute(stack(vs), Val((2, 1)))
    v = dropdims(mean(dragonfly, dims = 1), dims = 1)
end

The package used to support things like @cast _[i] := first(vs)[ixs[i]], indexing on the right. But this was complicated & fragile, and I removed it at some point, in favour of using Tullio for such things.

Indexing on the left is messier, as it's not entirely clear whether you want points with the same ixs[j] to overwrite or accumulate? Tullio's current behaviour is to overwrite, but that depends on what order it iterates in. I think it should probably be made an error, like https://github.com/mcabbott/Tullio.jl/issues/136 :

julia> @tullio v[ixs[j]] := vs[i][j] / 10
2-element Vector{Float64}:
 50.51324364045704
 60.28909109465455

[jariji]

Indexing on the left is messier, as it's not entirely clear whether you want points with the same ixs[j] to overwrite or accumulate

If I understand correctly, overwriting is a special case of aggregation using last, so maybe there is a way for the user to specify what they want.

[mcabbott]

I agree this could be given meanings. However I'm hoping not to add too many more features to this package! For now at least it shouldn't silently do the wrong thing.