Inserting a column of nans / namespace.column_class

[MarcoGorelli]

Motivation: in seaborn they insert a column of 'nan's at one point, which in pandas is as simple as

df[new_column_name] = np.nan

How do we do that within the Standard? I'm currently doing something like:

df: DataFrame
namespace = df.__dataframe_namespace__()
df= df.insert(
    df.shape()[1],
    new_column_name,
    namespace.column_class.from_sequence([float('nan')] * len(df)), dtype="float64"),
)

, is that OK?

If so, it would require adding a namespace.column_class method, which would do something like:

>>> df: PandasDataFrame
>>> namespace = df.__dataframe_namespace__()
>>> namespace.column_class
PandasColumn
>>> df: PolarsDataFrame
>>> namespace = df.__dataframe_namespace__()
>>> namespace.column_class
PolarsColumn
>>> # etc...

[rgommers]

That looks a bit hard to understand. It suggests a change to insert may be useful, as well as a free function constructor, so it can be written as:

ns = df.__dataframe_namespace__()
col_nans = ns.from_sequence([float('nan')] * df.shape()[0]), dtype="float64"),
df= df.insert('new_nancol', col_nans, loc=df.shape()[1])

This highlights my the uneasiness I felt with the classmethod constructors: it enforces knowing the exact name of the class. There is a reason why constructors are functions in array libraries, I think it's strictly superior. Adding a function to retrieve that class name instead can't be the right solution.

[MarcoGorelli]

ns.from_sequence looks good too, thanks!