ValueError: array must not contain infs or NaNs in seaborn.histplot(... kde=True)

[petsuter]

>>> import seaborn
>>> seaborn.__version__
'0.13.2'
>>> seaborn.histplot(x=range(5), weights=[0,0,3,0,0], kde=False)
<Axes: ylabel='Count'>
>>> seaborn.histplot(x=range(5), weights=[0,0,3,0,0], kde=True)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Python311\Lib\site-packages\seaborn\distributions.py", line 1416, in histplot
    p.plot_univariate_histogram(
  File "C:\Python311\Lib\site-packages\seaborn\distributions.py", line 447, in plot_univariate_histogram
    densities = self._compute_univariate_density(
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Python311\Lib\site-packages\seaborn\distributions.py", line 345, in _compute_univariate_density
    density, support = estimator(observations, weights=weights)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Python311\Lib\site-packages\seaborn\_statistics.py", line 193, in __call__
    return self._eval_univariate(x1, weights)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Python311\Lib\site-packages\seaborn\_statistics.py", line 154, in _eval_univariate
    kde = self._fit(x, weights)
          ^^^^^^^^^^^^^^^^^^^^^
  File "C:\Python311\Lib\site-packages\seaborn\_statistics.py", line 143, in _fit
    kde = gaussian_kde(fit_data, **fit_kws)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Python311\Lib\site-packages\scipy\stats\_kde.py", line 226, in __init__
    self.set_bandwidth(bw_method=bw_method)
  File "C:\Python311\Lib\site-packages\scipy\stats\_kde.py", line 574, in set_bandwidth
    self._compute_covariance()
  File "C:\Python311\Lib\site-packages\scipy\stats\_kde.py", line 586, in _compute_covariance
    self._data_cho_cov = linalg.cholesky(self._data_covariance,
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Python311\Lib\site-packages\scipy\linalg\_decomp_cholesky.py", line 101, in cholesky
    c, lower = _cholesky(a, lower=lower, overwrite_a=overwrite_a, clean=True,
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Python311\Lib\site-packages\scipy\linalg\_decomp_cholesky.py", line 18, in _cholesky
    a1 = asarray_chkfinite(a) if check_finite else asarray(a)
         ^^^^^^^^^^^^^^^^^^^^
  File "C:\Python311\Lib\site-packages\numpy\lib\function_base.py", line 630, in asarray_chkfinite
    raise ValueError(
ValueError: array must not contain infs or NaNs

[mwaskom]

This is happening because your data are effectively singular, with only one nonzero weight.

[petsuter]

Yes, but the data is what it is.... Is it unreasonable to expect seaborn to handle this (i.e. any data)?
Could seaborn log a warning and skip the kde in this case?

Should I use this:

def safe_kde_histplot(x,w):
    try:
        return seaborn.histplot(x=x, weights=w, kde=True)
    except:
        return seaborn.histplot(x=x, weights=w, kde=False)

Or what is the recommendation?

[mwaskom]

It's possible it could be handled here too; not sure. Annoying that scipy doesn't raise a consistent error that we could handle.

[petsuter]

Interesting.

• 13b3909 -> initial check
• _univariate_kdeplot check for singular data should compare variance to data magnitude #2186 -> use np.isclose to make check more robust
• Erroneous zero variance assertion leads to skipping of density estimation in kde/histplots #2294 -> use math.isclose to make check more robust
• kdeplot raising LinAlgError("singular matrix") instead of warning #2762 -> catch linalg.LinAlgError to make check more robust

Sounds like the intention is already to handle such cases, but it's not so easy to do reliably.
https://github.com/scipy/scipy/blob/v1.15.1/scipy/stats/_kde.py suggests maybe handling ValueError and LinAlgError would be sufficient?

The check seems to work as intended when scipy is not installed.

[mwaskom]

Yeah, there's a long history of singular data being passed into kdeplot being very annoying 😁