As discussed in #12, wrapping sklearn estimators in a custom class would allow us to:
Track the data used for fitting, e.g. the number and names of targets and whether single-output targets are squeezed.
~Check fitted feature names against image feature names and raise appropriate warnings (currently we always get warnings when fitting with dataframes, but we can potentially suppress that when predicting to named DataArrays and Datasets).~ Moving this detail to #17.
Only override relevant methods, e.g. kneighbors for kNN estimators.
As discussed in #12, wrapping
sklearnestimators in a custom class would allow us to:kneighborsfor kNN estimators.