API/DEPR: disallow casting nan-float to bool

[jbrockmendel]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

ser = pd.Series([np.nan])

>>> ser.astype(bool)
0    True

This is correct inasmuch as bool(np.nan) is True, but I think it's a bit surprising. Analogous to how we deprecated and later disallowed .astype("int64") when NaNs are present, should we deprecate allowing this for bool?

xref #43018

[jbrockmendel]

Also Series(index=[0], dtype=bool)[0] is True, which is correct-but-surpising.

[agnes-sharan]

Hi, I'm new to the repo and just stumbled upon this issue. According to python docs, there are explicit values that are considered False

https://docs.python.org/2/library/stdtypes.html#truth-value-testing
https://stackoverflow.com/questions/15686318/why-do-not-a-number-values-equal-true-when-cast-as-boolean-in-python-numpy

Unfortunately, since any object can be tested for truth value, the same reasoning as disallowing .astype("int64") may not be applicable here.

[rhshadrach]

Don't disagree that OP is surprising, but I'd say it's NumPy's fault 😆. I rather like the simplicity of .astype(bool) being roughly [bool(e) for e in ser], but would not be strongly opposed to raising.