[ https://issues.apache.org/jira/browse/LUCENE-6422?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
David Smiley resolved LUCENE-6422. ---------------------------------- Resolution: Fixed BTW I did a little more benchmarking. For point data there's a nice 10% improvement or so. Smaller index size too. For non-point data I used pruning and a distErrPct of 0.025 (default). The recursive prune of RPT on legacy quad (does more pruning then packed quad) amounted to a 25% smaller index than PackedQuad, and the index & search were practically the same. I can only surmise PackedQuad required less memory while indexing, of course. I _suspect_ that if RPT's recursive prune were to do its thing on PackedQuad, PackedQuad would have a better showing. I'll leave that benchmark to another day. My experience (& benchmarks) have shown that there is a high correlation between index size and search, and that pruning helps a ton. Some day I may write a better recursive prune that doesn't require buffering all tokens. Thanks again for contributing this Nick! > Add PackedQuadPrefixTree > ------------------------ > > Key: LUCENE-6422 > URL: https://issues.apache.org/jira/browse/LUCENE-6422 > Project: Lucene - Core > Issue Type: Improvement > Components: modules/spatial > Affects Versions: 5.x > Reporter: Nicholas Knize > Assignee: David Smiley > Fix For: 5.2 > > Attachments: LUCENE-6422-TRUNK.patch, LUCENE-6422.patch, > LUCENE-6422.patch, LUCENE-6422.patch, LUCENE-6422.patch, LUCENE-6422.patch, > LUCENE-6422_with_SPT_factory_and_benchmark.patch > > > This task introduces a PackedQuadPrefixTree that includes two things: > 1. A packed 8 byte representation for a QuadCell, including more efficient > implementations of the SPT API than the existing QuadPrefixTree or > GeoHashPrefixTree. > 2. An alternative implementation to RPT's "pruneLeafyBranches" that streams > the cells without buffering them all, which is way more memory efficient. > However pruning is limited to the target detail level, where it accomplishes > the most good. > Future improvements over this approach may include the generation of the > packed cells using an AutoPrefixAutomaton -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org