DataFrame input for SequentialFeatureSelector doesn't work for backward feature selection

[daniel-sch]

When modifying Example 12 from the documentation by replacing forward=True with forward=False, an error gets raised. In a quick test, only forward=True, floating=False seems to work, all other combinations raise errors when the input is a DataFrame.

The error was TypeError: unhashable type: 'slice' before upgrading pandas to 0.24.1, now it is TypeError: '(slice(None, None, None), (0, 1, 2))' is an invalid key.

I think this was the underlying issue in #486, but this was closed, so I'm opening a new issue.

I'm using mlxtend 0.13.0, please let me know if you need any other information.

[daniel-sch]

I think I just localized the error and I will create a PR shortly.