lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Robert Muir (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SOLR-2754) create Solr similarity factories for new ranking algorithms
Date Fri, 16 Sep 2011 23:56:08 GMT

    [ https://issues.apache.org/jira/browse/SOLR-2754?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13106924#comment-13106924
] 

Robert Muir commented on SOLR-2754:
-----------------------------------

{quote}
I can understand the reasoning behind this idea. On the other hand, for some models, the parameter
has a value that's optimal in a wide range of cases. In such cases, I think it we could make
the life of the user easier by falling back to this value. (Actually, that's why LMJelinekMercerSimilarity
does not have a default constructor; there is no single parameter value that is kind-of-optimal
in all cases).
{quote}

Well, we can do both: we can provide these basic parameters as default values to be friendly,
but at the same time in the test or example xml configurations that use these, our examples
can have the parameters set. Even in the JelinekMercer case, our example can also be set to
0.7, because thats the default for long queries and you typically don't use this smoothing
for short queries (you would usually use Dirichlet instead), at least that was my reasoning
with the default.

{quote}
Yes, I know, it'd be a bit difficult to support that... maybe if all Similarities and models
had a constructor with a map as a parameter? I'm not sure we want that, though.
{quote}

Yeah, I think we want to have hard type-safe apis for the sims themselves, and part of my
line of thinking is the case of "I'm going to plug in a custom normalization into DFR" is
a pretty expert case for a Solr user at this moment, if you are that expert you could also
write a 3 LOC sim factory that sets up your sim with your custom normalization method.


> create Solr similarity factories for new ranking algorithms
> -----------------------------------------------------------
>
>                 Key: SOLR-2754
>                 URL: https://issues.apache.org/jira/browse/SOLR-2754
>             Project: Solr
>          Issue Type: New Feature
>    Affects Versions: 4.0
>            Reporter: Robert Muir
>            Assignee: Robert Muir
>         Attachments: SOLR-2754.patch
>
>
> To make it easy to use some of the new ranking algorithms, we should add factories to
solr:
> * for parametric models like LM and BM25 so that parameters can be set from schema.xml
> * for framework models like DFR and IB, so that different basic models/normalizations/lambdas
can be chosen

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org


Mime
View raw message