[
https://issues.apache.org/jira/browse/LUCENE-7229?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15246490#comment-15246490
]
Robert Muir commented on LUCENE-7229:
-------------------------------------
Well it fails the tests which have nothing to do with BKD or anything else.
Also, it seems to return CROSSES very often when it should not (the cause of
the test failures).
This causes bad performance, for example with the "real world polygons" above,
performance is 0.8 QPS vs 8.6 QPS with this patch.
> Improve Polygon.relate
> ----------------------
>
> Key: LUCENE-7229
> URL: https://issues.apache.org/jira/browse/LUCENE-7229
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Robert Muir
> Attachments: LUCENE-7229.patch, LUCENE-7229.patch
>
>
> This method is currently quite slow and in many cases does more work than is
> required. The speed actually directly impacts queries (tree traversal) and
> bounds grid size to something tiny making it less effective.
> I think we should replace it line intersections based on orientation methods
> described here http://www.cs.berkeley.edu/~jrs/meshpapers/robnotes.pdf and
> https://www.cs.cmu.edu/~quake/robust.html
> For one, a naive implementation is considerably faster than the method today:
> both because it reduces the cost of BKD tree traversals and also because it
> makes grid construction cheaper. This means we can increase its level of
> detail with similar or lower startup cost. Now its more like a Mario Brothers
> 2 picture of your polygon instead of Space Invaders.
> Synthetic polygons from luceneUtil
> ||vertices||old QPS||new QPS||old startup cost||new startup cost||
> |50|20.4|21.7|1ms|1ms|
> |500|11.2|14.4|5ms|4ms|
> |1000|7.4|10.0|9ms|8ms|
> Real polygons (33 london districts:
> http://data.london.gov.uk/2011-boundary-files)
> ||vertices||old QPS||new QPS||old startup cost||new startup cost||
> |avg 5.6k|4.9|8.6|94ms|85ms|
> But I also like using this method because its possible to extend it to remove
> floating point error completely in the future with techniques described in
> those links. This may be necessary if we want to do smarter things (e.g. not
> linear time).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
