Value error in histplot with binwidth smaller than half the data range

[jhncls]

sns.histplot([1, 2, 3], binwidth=7) crashes with

"ValueError: bins must be positive, when an integer"

It is related to #2721, which is marked as solved, but the error also happens in the dev version.

The cause seems to be line 136 in counting.py with bins = int(round((stop - start) / binwidth)), setting it to 0 for small ranges. Changing this to bins = max(1, int(round((stop - start) / binwidth))) would probably solve it adequately.

(The code also has problems when binwidth=0 (division by zero) or negative (this makes bins negative, which causes numpy to protest)).

By the way, binwidth is ignored when discrete=True. As an example, sns.histplot(titanic[titanic['who'] == 'woman'], x='age', binwidth=5) has some bins with 5 and others with 6 ages. Of course, people who really care can provide their own custom bin edges.

[mwaskom]

Should probably just reject unless 0 < binwidth <= binrange. Not sure any other parametrization makes sense.

By the way, binwidth is ignored when discrete=True.

Yeah, the parameter is documented as

If True, default to binwidth=1 ...

Maybe "default to" implies that you can override it and should be "set binwidth=1

[jhncls]

Should probably just reject unless 0 < binwidth <= binrange. Not sure any other parametrization makes sense.

When the data isn't known in advance, in some situations only 0 or 1 data points will pop up. Without setting binwidth, now an empty plot is returned for 0 data points, and a symmetric bin of 1 wide for only one data point. With binwidth set, the code could behave similarly, using that binwidth. A plot showing something a bit reasonable could be friendlier than an error message.

If True, default to binwidth=1 ...

Well, my comment was a suggestion to also allow other integer binwidths for discrete data, shifting the edges by a half. (But maybe this would complicate things too much. Next, people will be asking for discrete units measured fractions, e.g. 1/10ths.)