Faster sum over axis
liuzhenqi77 opened this issue · comments
Description
Hi, currently I'm facing a performance issue where I need to sum over the first axis of a float64 coo matrix with shape like (14371, 91, 109, 91) and density around 0.01. I'm doing it with the native coo_matrix.sum(axis=0). I was wondering is there any way to optimize this for better speed?
Numba does not work in this case (No implementation found for signature sum(COOType[float64, int64, 4], axis=Literal[int](0))). I'm also not sure how to apply the custom solutions online on the sparse.coo type.
Thanks!
Well, unfortunately there isn't a faster way currently.