Scaling factors for metafeatures should not be learned using test data

[bjschoenfeld]

In kND.py, the _scale method uses the test data (other) to learn the min and max values used to scale the metafeatures. It is typically not a good idea to use test data to learn any parameters.

auto-sklearn/autosklearn/metalearning/metalearning/kNearestDatasets/kND.py

Lines 165 to 166 in 1c6af59

	mins = pd.DataFrame(data=[mins, other]).min()
	maxs = pd.DataFrame(data=[maxs, other]).max()

Would it be a good idea to use sklearn's MinMaxScaler anyways?

[mfeurer]

It is typically not a good idea to use test data to learn any parameters.

I agree. Do you want to submit a pull request taking care of this?

Would it be a good idea to use sklearn's MinMaxScaler anyways?

If it can handle pandas, yes.