Pandas dtype: Float64 is not supported

[tzipperle]

Hi,

I have installed: python 3.8.6, pandas 1.2.1 and altair 4.1.0. In the pandas version 1.2.0 they introduced a new "experimental" data type for nullable floats. I know that this type is experimental but a proper handling for nullable data is really convenient.

When I use this new type with altair I get a type error:

df = pd.DataFrame({'a': [1, 2, 3, 4, 5], 'b': [1.1, 2.5, np.nan, 1, 3.3]})
df = df.convert_dtypes()  # convert b from float64 to Float64
chart = alt.Chart(df).mark_line().encode(
    x='a',
    y='b'
)

TypeError: Cannot interpret 'Float64Dtype()' as a data type

[jakevdp]

Thanks - it looks like this arises from the fact that np.issubdtype does not understand pandas new dtypes, which is unfortunate. I suspect that breaks a lot of people's code.

[jakevdp]

Indeed, it seems to have broken dask as well: dask/dask#7156