COSMIT Avoid writing out vectorizable operations in sparsefuncs_fast#10615
COSMIT Avoid writing out vectorizable operations in sparsefuncs_fast#10615rth merged 2 commits intoscikit-learn:masterfrom
Conversation
|
I think 1 review is fine for this if it gets a green tick. |
jnothman
left a comment
There was a problem hiding this comment.
I don't understand how yet, but apparently there are test failures due to numerical instability :(
jnothman
left a comment
There was a problem hiding this comment.
No, due to missed parentheses, not numerical instability.
jnothman
left a comment
There was a problem hiding this comment.
This one should be trivial to review
rth
left a comment
There was a problem hiding this comment.
Actually I'm wondering: is writing a for loop in cython for a sum of two numpy arrays is strictly equivalent from the performance perspective to adding them via the numpy API?
Numpy may have a little more overhead in checking inputs, but is often more efficient in the calculation of vectorised operations. |
Right, due to BLAS I imagine.
LGTM, merging. Thanks @jnothman ! |
Someone got a bit cython crazy.