[
https://issues.apache.org/jira/browse/LUCENE-7974?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16180245#comment-16180245
]
Steve Rowe edited comment on LUCENE-7974 at 9/26/17 4:53 AM:
-------------------------------------------------------------
bq. I'm wondering whether the use of getMinDelta could be replaced with
Math.nextUp/Math.nextDown?
These do different things, and I'm not sure how to express one in terms of the
other. Suggestions welcome :).
{{getMinDelta}} calculates a fudge factor from the distance exponent reduced by
(at most) 23, the number of bits in a float mantissa. This is necessary when
the result of subtracting/adding the distance in a single dimension has an
exponent that differs significantly from that of the distance value. Without
this fudge factor (i.e. only subtracting/adding the distance), cells and values
can be inappropriately judged as outside the search radius.
By contrast, {{Math.nextUp}}/{{Math.nextDown}} produce adjacent values (i.e.
the equivalent of incrementing/decrementing the mantissa value by one).
was (Author: steve_rowe):
bq. I'm wondering whether the use of getMinDelta could be replaced with
Math.nextUp/Math.nextDown?
These do different things, and I'm not sure how to express one in terms of the
other. Suggestions welcome :).
{{getMinDelta}} calculates a fudge factor from the distance exponent reduced by
(at most) 23, the number of bits in a float mantissa. This is necessary when
the result of subtracting/adding the distance in a single dimension has an
exponent that differs significantly from the distance value. Without this
fudge factor (i.e. only subtracting/adding the distance), cells and values can
be inappropriately judged as outside the search radius.
By contrast, {{Math.nextUp}}/{{Math.nextDown}} produce adjacent values (i.e.
the equivalent of incrementing/decrementing the mantissa value by one).
> Add N-dimensional FloatPoint K-nearest-neighbor implementation
> --------------------------------------------------------------
>
> Key: LUCENE-7974
> URL: https://issues.apache.org/jira/browse/LUCENE-7974
> Project: Lucene - Core
> Issue Type: New Feature
> Components: modules/sandbox
> Reporter: Steve Rowe
> Assignee: Steve Rowe
> Priority: Minor
> Attachments: LUCENE-7974.patch
>
>
> From LUCENE-7069:
> {quote}
> KD trees (used by Lucene's new dimensional points) excel at finding "nearest
> neighbors" to a given query point ... I think we should add this to Lucene's
> sandbox
> [...]
> It could also be generalized to more than 2 dimensions, but for now I'm
> making the class package private in sandbox for just the geo2d (lat/lon) use
> case.
> {quote}
> This issue is to generalize {{LatLonPoint.nearest()}} to more than 2
> dimensions.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
