how can I select matrix values for those cols and rows for which an f(col,row)==true?

I have a huge matrix A (10k*10k) and I have a function that returns false/true for each combination of a row and a column number : f(row, col) = boolean. I need to do A(f(row,col)) = X;

How to do that quickly (I need to do this very often)?

I saw arrayfun and spfun but the element function does not allow for row and column arguments.

Answer 1

[Guillaume]

[rows, cols] = ndgrid(1:size(A, 1), 1:size(A, 2));

Then, if f can operate directly with pairs of matrices, simply:

A(f(rows, cols)) = X;

otherwise if f only works with scalar values:

A(arrayfun(@f, rows, cols)) = X;

[Guillaume]

Comment by Jasper van Casteren moved here:

thank you for this, but

[rows, cols] = ndgrid(1:size(A, 1), 1:size(A, 2));

takes more than a second, and allocates two huge matrices. Is there no more efficient way?

[Guillaume]

Depending on how f is implemented you may get away with:

A(f((1:size(A, 1))', 1:size(A, 2))

or

A(bsxfun(@f, (1:size(A, 1))', 1:size(A, 2))

but ultimately, it's a trade-off of memory vs speed. If a loop works better for you then use that.

It's also possible that modifying the way f works may provide some / lots of / no potential for improvement.