[ 
https://issues.apache.org/jira/browse/LUCENE-5536?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13939372#comment-13939372
 ] 

Shai Erera commented on LUCENE-5536:
------------------------------------

Don't know that we should because in practice nothing is rolled up. Because 
DimConfig.hierarchical=false, and rollup() does {{if (ft.hierarchical && 
ft.multiValued == false) {}}, it means we're just wasting a fraction of a CPU 
today... but there's no bug per se. I will backport if it's important, but I 
didn't even plan to mention that in CHANGES :).

> TaxonomyFacetSumInt/FloatAssociations should not rollup()
> ---------------------------------------------------------
>
>                 Key: LUCENE-5536
>                 URL: https://issues.apache.org/jira/browse/LUCENE-5536
>             Project: Lucene - Core
>          Issue Type: Bug
>          Components: modules/facet
>            Reporter: Shai Erera
>         Attachments: LUCENE-5536.patch
>
>
> Stumbled upon this by accident when I reviewed the code. The previous 
> associations impl never rolled-up. The assumption is that association values 
> are given to exact categories and have no hierarchical meaning. For instance 
> if a document is associated with two categories: {{Category/CS/Algo}} and 
> {{Category/CS/DataStructure}} with weights {{0.95}} and {{0.43}} 
> respectively, it is not associated with {{Category/CS}} with weight {{1.38}}! 
> :)
> If the app wants to association values to apply to parents in the hierarchy 
> as well, it needs to explicitly specify that (as in passing the hierarchy 
> categories with their own association value).
> I will fix the bug and also make sure the app cannot trip it by accidentally 
> specifying hierarchical on these categories, or that if it does (cause e.g. 
> it indexes the categories for both counting and assoc values) then we don't 
> apply the association to all the categories in the hierarchy.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to