Dan Schult
Dan Schult
Great! Thanks for the pointer! (Of course we always have to take the proof of a complexity result to be part 1 of the process of getting a usable algorithm....
If you make the PR we can take a look and see not only what the trouble with floating point is, but also how the algorithm works generally, etc. So...
~~I have created a separate PR with a sparse version of `broadcast_shapes` that is needed for this PR to pass the 32-bit system tests. I didn't include it here to...
Resolved merge conflicts-- :)
Thanks very much @tgbrooks especially for pointing out the problems with asymmetric operators and the potential efficiency improvement when res0 == 0. And also for your all-in-one code-path example. The...
I added new commits to split broadcasting of columns into its own function in each case of `general` and `canonical`. The previous combined version is in my branch `broadcast_binopt_if_clauses`. The...
I have rebased on main and updated the separated broadcasting functions. The structure and issues are largely the same as I described in the previous post. I think the next...
This looks to me like a problem with `numpy.max` raising when in input array is empty. Or said another way, the `hb.from_data` function is not checking whether there are any...
OK... looks like I'll approve this and make a small PR to add the hash to the ignore file. :)
The failures are from test_base.py checking for NotImplementedError for DOK and BSR when axis is set. Need to update `not_supported` var. But maybe (if DIA support is added) we could...