How to add query filter to the Nearest Neighbors algorithm?

Question

I have Nearest Neighbors model, built with sklearn sklearn.neighbors.NearestNeighbors, which I use to make content based recommendations.

Sometimes I need to retrieve recommendations based on query conditions. For example, I want to get items within concrete location.

How can I add location filter to Nearest Neighbors query?

Answer 1

I solved this problem with IDSelectorBatch. I have a dictionary where fields that I use in conditions are mapped to item Ids. Ids from this dictionary are used in 'IDSelectorBatch' to make queries work as queries with conditions. Example with location field:

1.Create IDSelectorBatch with required ids:

def _get_location_filter(self, request_locations):
    try:
        same_location_ids = []
        # use locations from request
        for loc in request_locations:
            same_location_ids.extend(self._locationsToId[loc])

        return faiss.IDSelectorBatch(list(same_location_ids))
    except KeyError:
        return faiss.IDSelectorAll()

1.Use it to search in faiss index:

sel = self._get_location_filter(request['locations'])

params = (
    faiss.SearchParametersIVF(sel=sel)
)

indices = self._classifier.predict(features, params=params, k=k)

return list(indices[0][:])