Enable None sentinel for columns

[vsriram11]

There is a None sentinel for transformers, but not columns. Having a None sentinel for the columns allows a set of transformers to act on all columns of the incoming dataframe.

None sentinel for columns is important, because sometimes we don't know the columns of the dataframe. For instance, if DataFrameMapper is embedded in a long sklearn Pipeline that may have added many columns to the original dataframe, it's hard to know exact column names in the dataframe that comes in (but we may want to apply some transfomrers to all columns).

This feature also bridges the gap to make DataFrameMapper a fully-fledged PandasFeatureUnion. - something that has been requested multiple times in Issue #62 , #64 , and #69 .

[ragrawal]

@vsriram11 this is now (release V2.2.0) possible using callable methods. You can do something like this:

class AllColumnSelector:
        def __call__(self, X:pd.DataFrame=None):
              return list(X.columns)

mapper = DataFrameMapper([
      (AllColumnSelector(), None)
])