Feature Request: pd.MultiIndex.from_frame(). Complement to pd.MultiIndex.to_frame().

[ms7463]

The MultiIndex.to_frame function is great for working with multi-indexes as a meta-dataframe. I find myself using this paradigm very often.

df = create_multiindex_df()
meta = df.columns.to_frame(index=False)
meta = filter_multiindex()
df.reindex(columns=convert_df_to_multiindex(df))

Having the convert_df_to_multiindex as a pd.MultiIndex method would be extremely complimentary to pd.MultiIndex.to_frame.

A simplified one-line implementation provided below, not including accounting for some corner case behaviors and discerning between series/frames vs Index/Multiindex.

pd.MultiIndex.from_tuples(list(df.values), names=df.columns)

[jreback]

can u show a complete example

[ms7463]

Here is a more complete implementation:

# pd.Index

@classmethod
def from_frame(cls, df):
    if not isinstance(df.squeeze(), pd.Series):
        raise ValueError('DataFrame must be be single column')
    return cls.from_series(df.squeeze())


@classmethod
def from_series(cls, s):
    return cls(s, name=s.name)



# pd.MultiIndex

@classmethod
def from_frame(cls, df, squeeze=True):
    """
    :param df
    :param squeeze
        Squeeze single level multiindex to be a regular index
    """
    # just let column level names be the tuple of the meta df columns since they're not required to be strings
    # columns = ['.'.join(col) for col in list(df)]  
    columns = list(df)
    mi = cls.from_tuples(list(df.values), names=columns)
    if squeeze:
        if len(mi.levels) == 1:
            return mi.levels[0][mi.labels[0]]
        else:
            return mi
    else:
        return mi

[ms7463]

Please see this comment for a justification of this feature.

Please see this comment for a demonstration of the new vs current way to do specific tasks.