What you are saying seems to be right. Can you provide an minimal example to be certain of the behavior.

or just submit a fix

I'm assuming it's just removing the lines that raise the exception and adding a test? I should be able to submit a PR in the next day or two.

The issue is clarifying class documentation of RobustScaler as you can only transform() on sparse input after fitting on dense input. Perhaps it makes sense to add a test confirming that exceptions are raised appropriately.

(Note a typo here) http://scikit-learn.org/stable/modules/preprocessing.html#scaling-sparse-data

RobustScaler cannot be fited to sparse inputs, but you can use the transform method on sparse inputs.

See also #4125 (comment) - RobustScaler.fit() on sparse input hasn't been implemented.

The issue is clarifying class documentation of RobustScaler as you can only transform() on sparse input after fitting on dense input

A PR would be more than welcome.

(Note a typo here) http://scikit-learn.org/stable/modules/preprocessing.html#scaling-sparse-data

@naoyak can you clarify what you mean by typo ?