ENH: Accept pd.NA as a float

[mdruiter]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

In float[pyarrow] data, missings values are pd.NA. I can easily convert these frames to floats:

pd.Series([pd.NA], dtype='float[pyarrow]').astype('float')

0 NaN
dtype: float64

But in the pd.Series constructor, I cannot use pd.NAs:

pd.Series([pd.NA], dtype='float')

TypeError: float() argument must be a string or a real number, not 'NAType'

Is this by design?

Feature Description

It would be so much more convenient if pd.NA would 'just' be recognized as a np.nan in float context...

Alternative Solutions

Using .astype('float') is a workaround but makes code less generic.

Additional Context

I searched for issues like this or a design statement, but could not find anything about this specific thing, which surprised me as it feels like a common inconvenience.

[mdruiter]

I think I found a relevant statement after all: #32265 (comment)

[rhshadrach]

It would be so much more convenient if pd.NA would 'just' be recognized as a np.nan in float context...

This is the essentially the direction pandas plans to head in.

cc @phofl @jorisvandenbossche - I think this issue is likely duplicative of others we have on np.nan, pd.NA. Good to close?

[phofl]

Yes