[
https://issues.apache.org/jira/browse/LUCENE-7862?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16625692#comment-16625692
]
Ignacio Vera commented on LUCENE-7862:
--------------------------------------
For reference:
I was doing a bit research about how to further improve performance of the BKD
tree and I come across this paper:
http://infolab.stanford.edu/~nsample/pubs/samplehaines.pdf
In point 6 they speak about a performance improvement by doing exactly what was
implemented ihere. They put a name to the approach, BOUNDS-OVERLAPS-BALL (BOB)
test.
> Should BKD cells store their min/max packed values?
> ---------------------------------------------------
>
> Key: LUCENE-7862
> URL: https://issues.apache.org/jira/browse/LUCENE-7862
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Adrien Grand
> Assignee: Ignacio Vera
> Priority: Minor
> Fix For: 7.5, master (8.0)
>
> Attachments: LUCENE-7862.patch, LUCENE-7862.patch, LUCENE-7862.patch
>
>
> The index of the BKD tree already allows to know lower and upper bounds of
> values in a given dimension. However the actual range of values might be more
> narrow than what the index tells us, especially if splitting on one dimension
> reduces the range of values in at least one other dimension. For instance
> this tends to be the case with range fields: since we enforce that lower
> bounds are less than upper bounds, splitting on one dimension will also
> affect the range of values in the other dimension.
> So I'm wondering whether we should store the actual range of values for each
> dimension in leaf blocks, this will hopefully allow to figure out that either
> none or all values match in a block without having to check them all.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
