The unroll parameters passed to each compute_mean call in MMD.compute should be augmented. If the user passes unroll as a tuple (unroll_x, unroll_y), then we should augment this tuple for each call to compute_mean
compute_mean(x, x, unroll=(unroll_x, unroll_x))
compute_mean(y, y, unroll=(unroll_y, unroll_y))
compute_mean(x, y, unroll=(unroll_x, unroll_y))
What value does this add?
This provides consistent handling with that of block_size.
What's the new feature?
The unroll parameters passed to each
compute_meancall inMMD.computeshould be augmented. If the user passesunrollas a tuple(unroll_x, unroll_y), then we should augment this tuple for each call tocompute_meancompute_mean(x, x, unroll=(unroll_x, unroll_x))compute_mean(y, y, unroll=(unroll_y, unroll_y))compute_mean(x, y, unroll=(unroll_x, unroll_y))What value does this add?
This provides consistent handling with that of
block_size.Is there an alternative you've considered?
No response
Additional context
No response