[
https://issues.apache.org/jira/browse/LUCENE-8368?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16521755#comment-16521755
]
Robert Muir commented on LUCENE-8368:
-------------------------------------
On the OSM benchmark it can facet-by-london-borough at ~ 20M hits/sec on my
machine, better than I thought it would do.
There is some smelly stuff to figure out first though:
* the LongValues abstraction used here doesn't support SortedNumeric, that's a
problem for LatLonDV. It seems to also be an issue for the numeric range facet
classes here.
* the re-use of range stuff makes for less code, but we really shouldn't be
doing binary search, since our ranges are 100% dense.
* maybe not so great in the API to force construction of a polygon2D in every
query? But when i benchmarked with the boroughs polygons (33 polygons, 186,318
total vertices), this didn't seem to matter either.
> facet by polygon
> ----------------
>
> Key: LUCENE-8368
> URL: https://issues.apache.org/jira/browse/LUCENE-8368
> Project: Lucene - Core
> Issue Type: New Feature
> Components: modules/facet
> Reporter: Robert Muir
> Priority: Major
> Attachments: LUCENE-8368.patch
>
>
> This can give some nice flexibility if you are working with search results on
> a map. Of course if everything about your use-case is static, its better to
> compute this up-front and index string values, but its not always the case.
> Also it can be helpful if your polygons are changing often, since you don't
> have to reindex.
> Polygon2D already supports multipolygons, but today it only returns a boolean
> value. This patch adds a {{find}} method that returns the polygon that
> actually matched, or -1 if it doesn't match. {{contains}} is then just
> written as {{find >= 0}}.
> Then we can solve the problem with just some sugar over the existing range
> faceting, as each multipolygon is just a range of ids coming back from
> {{find}} that correspond with it. e.g. if you were faceting by country, you
> might have ~200 countries with 100,000 total polygons, and polygons
> 22,000-32,000 correspond to Canada or whatever.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
