[SciPy-Dev] SciPy-Dev Digest, Vol 193, Issue 12

[Robert Kern]

On Sat, Nov 16, 2019 at 12:07 AM Warren Weckesser <
warren.weckesser at gmail.com> wrote:

> On 11/15/19, Robert Kern <robert.kern at gmail.com> wrote:
>
> > I still don't really like that part of the test. I think that I would
> > prefer that all distributions would be treated as "shape_only". Did it
> help
> > us solve a bug once?
>
> The original bug report (first reported in 2011):
>
>     https://github.com/scipy/scipy/issues/2069
>
> For some distributions, rvs() returned samples with the correct shape,
> but the random variates were not properly generated.
>
> Checking just the shape is a crude test; how does one verify that the
> parameter values were used correctly to generate the random variates?
>

It looks like this particular bug manifested itself in a particular way
that would be easy to recognize. It might be worth checking if we see that
pattern instead (i.e. assume it's good unless it tests bad rather than
assume it's bad unless if it precisely reproduces these numbers computed in
a different way). I think that would cover the "shape_only" set, too.

-- 
Robert Kern
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20191116/49e83884/attachment.html>