gsmiller commented on code in PR #13568: URL: https://github.com/apache/lucene/pull/13568#discussion_r1693220989
########## lucene/sandbox/src/java/org/apache/lucene/sandbox/facet/ordinals/CandidateSetOrdinalIterator.java: ########## @@ -0,0 +1,48 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +package org.apache.lucene.sandbox.facet.ordinals; + +import java.io.IOException; +import org.apache.lucene.internal.hppc.IntHashSet; + +/** + * {@link OrdinalIterator} that filters out ordinals from delegate if they are not in the candidate Review Comment: Right, that makes sense. This is a little tricky right? Let's say I want to do "long value" faceting on a numeric field, I have a matchset with ~10,000 distinct values in the field, I specifically care about 20 of them (that I know ahead-of-time) and then I want the top-10 from those 20. So in that case, what's the ideal access pattern? It seems like I'd want to do 20 map lookups for the values I care about and then use a heap-sort (or heck, maybe some other sort would be better?) to get my top-10 right? The user can always do that directly with the recorder API and handle sorting on their own, but this abstraction nudges them towards an inefficient chain of iterators doesn't it? With the iterator pattern, they would end up iterating all 10,000 distinct values in the map just to filter down to the 20 they care about (then the top-n iterator logic is fine). If `k` is the number of values we care about and `n` is the number of distinct values seen during accumulation, then we're tur ning an `O(k)` problem into `O(n)`. If `k << n` then this is kind of silly. What about this as an idea (maybe terrible... just brainstorming here)? What if we, (1) added a `bool contains(int ord)` method to `FacetRecorder`, and (2) changed this candidate iterator to take in a `FacetRecorder` during initialization instead of wrapping another iterator, then (3) it could "lead" the iteration from the candidate set provided and use `contains` to filter it down (rather than the current pattern that starts with all ordinals seen by the recorder and filters down using the provided set)? This is of course assuming the common case is the recorder observes many more ordinals than provided candidates. Or maybe we could support both patterns somehow? Maybe the iterator could be smart about this like we are in other places ("leading" with the candidate or the recorder ordinals depending on heuristics)? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
