On 6/10/2018 7:53 PM, Neil Girdhar wrote:
On Sun, Jun 10, 2018 at 5:41 PM Terry Reedy <tjreedy@udel.edu <mailto:tjreedy@udel.edu>> wrote:
On 6/10/2018 10:44 AM, Stephan Houben wrote:
> I would suggest that compatibility with a major Python library such as > SciPy is more important than compatibility > with other programming languages. > > I would go even further and argue that scipy.special.sindg and its > friends cosdg and tandg > can serve as the reference implementation for this proposal.
Or we could decide that we don't need to duplicate scipy.
Copying scipy's and numpy's interface makes vectorizing code a lot simpler. tensorflow also initially made the mistake of varying from numpy's interface only to then deprecate the variations and adopt the standard.
What I meant is to not add functions that already exist in scipy. Core devs do not need the burden of keeping up with whatever improvements are made to the scipy functions. -- Terry Jan Reedy