TrieRangeQuery - thanks for the tip.
-----Original Message----- From: Michael McCandless <luc...@mikemccandless.com> Reply-To: java-user@lucene.apache.org To: java-user@lucene.apache.org Subject: Re: Does Lucene fail fast on boolean queries? Date: Thu, 21 May 2009 11:39:23 -0400 On Thu, May 21, 2009 at 10:58 AM, Joel Halbert <j...@su3analytics.com> wrote: > Thx. We're not relying on the internal implementation, but I was > wondering with respect to how efficient it is with respect to doing a > boolean AND query. > > i.e. does clause precedence effect the efficiency of the query - so is X > && Y faster than Y && X if there are fewer hits for X. From how you > describe it it is equally efficient, either way. Lucene has a simple heuristic (in ConjunctionScorer): it first calls next() on each of X and Y, then it looks at that first docID for each and re-orders the clauses so that whichever of X or Y had the largest first docID is the one that "drives" the intersection. But that heuristic could easily be wrong if it just so happened that a rare term appears in an early document. A simple improvement would be to add an approxCount() to Query or Scorer/DocIdSet and then order based on that. But we haven't done so yet... > In particular we are trying to work out whether a particular numerical > RangeQuery that needs to be AND'd with a TermQuery is fastest as: > > BooleanQuery(RangeQuery && TermQuery) > > or as a TermQuery which then has it's results filtered by processing > each in turn. If the RangeQuery will hit lots of terms and/or docs, you're likely better off building the filter; else, the query. Test both and let the machine tell you :) But: orthogonal to that decision, you should look @ TrieRangeQuery (in 2.9-dev, contrib/queries) which does numeric filtering typically with far fewer terms needing to be visited. Mike --------------------------------------------------------------------- To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org For additional commands, e-mail: java-user-h...@lucene.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org For additional commands, e-mail: java-user-h...@lucene.apache.org
