How to filter columns of a matrix by columns of another matrix

Specific Domains Data
9

This is a bit inefficient for large matrices because it scales as the product of the number of columns of A and B. It might be better to construct a Set of B’s columns so that you can use fast hash-table lookup:

Bcols = Set(eachcol(B))
C = reduce(hcat, collect(c for c in eachcol(A) if c ∉ Bcols))

(I use collect here because of issue #31636: Julia already has an optimized reduce(hcat, x) method but only for the case where x is an array of arrays.) Alternatively, you could do

Bcols = Set(eachcol(B))
C = A[:, @views [i for i in axes(A,2) if A[:,i] ∉ Bcols]]

which might be faster than reduce?

As usual, if you want to benchmark these it is best to first put the code into a function:

function filtercols(A, B)
    Bcols = Set(eachcol(B))
    return A[:, @views [i for i in axes(A,2) if A[:,i] ∉ Bcols]]
end

(In general, you might want to step back and re-think the data structures you are using—matrices might not be the best choice depending on the operations you want to perform. e.g. you might want to store B as a Set of columns to begin with.)

5 Likes