BUG: Possible bug when using winsorize on pandas data instead of numpy data

Describe your issue.

When scipy.stats.mstats.winsorize is used with a nan_policy of omit on a numpy array it behaves as expected. However, when used on Pandas data such as a series, all NaNs are converted to the maximum possible value. I believe this may be a bug, but could possibly be an enhancement if pandas data is never meant to be used. An example can be seen in the image below:

Reproducing Code Example

import numpy as np
import pandas as pd
from scipy.stats.mstats import winsorize

nans_10 = list(np.repeat(np.nan,10))
data = np.array(
	nans_10
	+ list(range(100))
	+ nans_10
)


print(f"Numpy array: {winsorize(data, (0.2, 0.2), nan_policy = 'omit')}")

print(f"Pandas series: {winsorize(pd.Series(data), (0.2, 0.2), nan_policy = 'omit')}")

Error message

No error/warning message is given. The behavior is just unexpected. The pandas version used was 1.4.2.

SciPy/NumPy/Python version information

1.8.0 1.21.6 sys.version_info(major=3, minor=10, micro=2, releaselevel=‘final’, serial=0)

[mdhaber]

The problem was that _contains_nan relies on the sum of the array to be nan if there are any NaNs. This is not true for Pandas types, which ignore NaNs by default. I’ll submit a PR.

[mdhaber]

<facepalm>