Hi Kevin
As I wrote in my original mail, BLAS has no routine for this operation,
you need at least 2 BLAS (or numpy) operatios and I'd like to avoid the
drawback of those solutions because this is a very performance critical
and memory bound operation (X can be very large).
The undertaking o
As nice as numba is, it’s a very heavy dependency that many high profile libraries want to avoid.BestChristian Am 18.08.2025 um 14:25 schrieb Kevin Sheppard :This is a good place for numba if performance and memory use are a concern.Good NumPy125 ms ± 2.42 ms per loop (mean ± std. dev. of 7 runs,
This is a good place for numba if performance and memory use are a concern.
Good NumPy
125 ms ± 2.42 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
Naive numba (memory efficient)
386 ms ± 1.83 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
Parallel numba (memory efficient, cpu bo
If W is just a vector, i.e. my main use case, than what you write is
what I wanted to imply and actually how I implemented it in several cases.
The point is that X.T * weights[None, :] still doubles memory (same size
as X) and this solution doesn't exploit the symmetry.
Best
Christian
On 17.08
Why not:
XTWX = (X.T * weights[None, :]) @ X
if `weights` is a vector? Allocating `diag(weights)` is the big memory and
CPU hog at least for my generally long and skinny X matrices.
-Kevin
On Fri, Aug 15, 2025 at 10:08 AM Christian Lorentzen via NumPy-Discussion <
numpy-discussion@python.org>
Well, what I ment to say was "Perhaps just the treatment of complex numbers is
not available".
Best,
Andrey
___
NumPy-Discussion mailing list -- numpy-discussion@python.org
To unsubscribe send an email to numpy-discussion-le...@python.org
https://mail.p
Hi Andrey,
I don't think this kind of formatting belongs in numpy: there is no
obvious correct way to display numbers in tabular form, and this just
invites a hard-to-maintain jungle of keyword arguments. For instance, I
find lines in tables generally unnecessary, so they need to be optional.
But
Perhaps just the treatment of complex numbers.
Best,
Andrey
___
NumPy-Discussion mailing list -- numpy-discussion@python.org
To unsubscribe send an email to numpy-discussion-le...@python.org
https://mail.python.org/mailman3//lists/numpy-discussion.python
I see that the examples were misformatted, here how they were intended to look
like.
Best,
Andrey
___
NumPy-Discussion mailing list -- numpy-discussion@python.org
To unsubscribe send an email to numpy-discussion-le...@python.org
https://mail.python.
How much of this functionality is available in the tabulate package? _JM
__
John Mark Agosta LinkedIn:
https://www.linkedin.com/in/john-mark-agosta/
johnmark.ago...@gmail.comFind me at https://medium.com/@johnmark54
On Tue, Aug 5, 2025 at 6:55 AM Андрей Рыба
10 matches
Mail list logo
