[Numpy-discussion] Converting np.sinc into a ufunc
ralf.gommers at gmail.com
Wed May 22 16:58:48 EDT 2019
On Wed, May 22, 2019 at 7:34 PM Nathan Goldbaum <nathan12343 at gmail.com>
> It might be worth using BigQuery to search the github repository public
> dataset for usages of np.sinc with keyword arguments.
We spent some effort at Quansight to try different approaches to this.
BigQuery turns out to be suboptimal, parsing code with ast.parse is more
robust. Chris Ostrouchov just released some code for this (blog post with
details to follow) and the results of running that code:
np.sinc has 35 usages. to put that in perspective, np.array has ~31,000,
np.dot ~2200, np.floor ~220, trace/inner/spacing/copyto are all similar to
> On Wed, May 22, 2019 at 1:05 PM Sebastian Berg <sebastian at sipsolutions.net>
>> Hi all,
>> there is an open PR (https://github.com/numpy/numpy/pull/12924) to
>> convert `np.sinc` into a ufunc. Since it should improve general
>> precision in `np.sinc`, I thought we could try to move that forward a
>> bit. We check whether this is worth it or not in the end.
Can you quantify the precision improvement (approximately)?
>> However, it would also change behaviour slightly since `np.sinc(x=arr)`
>> will not work, as ufuncs are positional arguments only (we could wrap
>> `sinc`, but that hides all the nice features).
This would give an exception, so at least it's easy to fix. As backwards
compat breaks go, this is a pretty minor one I think.
Otherwise, there should
>> be no change except additional features of ufuncs and the move to a C
I see this is one of the functions that uses asanyarray, so what about
impact on subclass behavior?
>> This is mostly to see if anyone is worried about possible slight API
>> change here.
>> All the Best,
>> NumPy-Discussion mailing list
>> NumPy-Discussion at python.org
> NumPy-Discussion mailing list
> NumPy-Discussion at python.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the NumPy-Discussion