You are viewing a plain text version of this content. The canonical link for it is here.

Posted to solr-user@lucene.apache.org by ilayaraja <il...@gmail.com> on 2018/05/02 05:23:40 UTC

Re: Learning to Rank (LTR) with grouping

*
"Top K shouldn't start from the "start" parameter, if it does, it is a bug.
"***

1. I clearly see that LTR do re-rank based on the start parameter.
2. When reRankDocs=24, pageSize=24, I still get the second page of results
re-ranked by ltr plugin when I query with start=24.


Alessandro Benedetti wrote
> Are you using SolrCloud or any distributed search ?
> 
> If you are using just a single Solr instance, LTR should have no problem
> with pagination.
> The re-rank involves the top K and then you paginate.
> So if a document from the original score page 1 ends up in page 3, you
> will
> see it at page three.
> have you verified that : "Say, if an item (Y) from second page is moved to
> first page after 
> re-ranking, while an item (X) from first page is moved away from the first 
> page.  ?" 
> Top K shouldn't start from the "start" parameter, if it does, it is a bug.
> 
> The situation change a little with distributed search where you can
> experiment this behaviour : 
> 
> *Pagination*
> Let’s explore the scenario on a single Solr node and on a sharded
> architecture.
> 
> SINGLE SOLR NODE
> 
> reRankDocs=15
> rows=10
> This means each page is composed by 10 results.
> What happens when we hit the page 2 ?
> The first 5 documents in the search results will have been rescored and
> affected by the reranking.
> The latter 5 documents will preserve the original score and original
> ranking.
> 
> e.g.
> Doc 11 – score= 1.2
> Doc 12 – score= 1.1
> Doc 13 – score= 1.0
> Doc 14 – score= 0.9
> Doc 15 – score= 0.8
> Doc 16 – score= 5.7
> Doc 17 – score= 5.6
> Doc 18 – score= 5.5
> Doc 19 – score= 4.6
> Doc 20 – score= 2.4
> This means that score(15) could be < score(16), but document 15 and 16 are
> still in the expected order.
> The reason is that the top 15 documents are rescored and reranked and the
> rest is left unchanged.
> 
> *SHARDED ARCHITECTURE*
> 
> reRankDocs=15
> rows=10
> Shards number=2
> When looking for the page 2, Solr will trigger queries to she shards to
> collect 2 pages per shard :
> Shard1 : 10 ReRanked docs (page1) + 5 ReRanked docs + 5 OriginalScored
> docs
> (page2)
> Shard2 : 10 ReRanked docs (page1) + 5 ReRanked docs + 5 OriginalScored
> docs
> (page2)
> 
> The the results will be merged, and possibly, original scored search
> results
> can top up reranked docs.
> A possible solution could be to normalise the scores to prevent any
> possibility that a reranked result is surpassed by original scored ones.
> 
> Note: The problem is going to happen after you reach rows * page >
> reRankDocs. In situations when reRankDocs is quite high , the problem will
> occur only in deep paging.
> 
> 
> 
> -----
> ---------------
> Alessandro Benedetti
> Search Consultant, R&D Software Engineer, Director
> Sease Ltd. - www.sease.io
> --
> Sent from: http://lucene.472066.n3.nabble.com/Solr-User-f472068.html





-----
--Ilay
--
Sent from: http://lucene.472066.n3.nabble.com/Solr-User-f472068.html

Re: Learning to Rank (LTR) with grouping

Posted by ilayaraja <il...@gmail.com>.

Also, would like to understand what are the ways to optimize for performance
at search time with LTR. Queries with terms (that fetch more results) lead
to very high latency with re-rank query even for reRankDocs=24. 

Is there best practices to reduce the latency?

Can fv cache help?
	<cache name="QUERY_DOC_FV"
      		class="solr.search.LRUCache"
      		size="4096"
      		initialSize="2048"
      		autowarmCount="4096"
      		regenerator="solr.search.NoOpRegenerator" />

Should I increase the cache size?





-----
--Ilay
--
Sent from: http://lucene.472066.n3.nabble.com/Solr-User-f472068.html

Re: Learning to Rank (LTR) with grouping

Posted by Diego Ceccarelli <di...@gmail.com>.

Thanks ilayaraja,

I updated the PR today integrating your and Alan's comments. Now it works
also in distributed mode. Please let me know what do you think :)

Cheers
Diego

On Wed, May 2, 2018, 17:46 ilayaraja <il...@gmail.com> wrote:

> Figured out that offset is used as part of the grouping patch which I
> applied
> (SOLR-8776) :
> solr/core/src/java/org/apache/solr/handler/component/QueryComponent.java
> +      if (query instanceof AbstractReRankQuery){
> +        topNGroups = cmd.getOffset() +
> ((AbstractReRankQuery)query).getReRankDocs();
> +      } else {
> +        topNGroups = cmd.getOffset() + cmd.getLen();
>
>
>
>
>
>
> -----
> --Ilay
> --
> Sent from: http://lucene.472066.n3.nabble.com/Solr-User-f472068.html
>

Re: Learning to Rank (LTR) with grouping

Posted by ilayaraja <il...@gmail.com>.

Figured out that offset is used as part of the grouping patch which I applied
(SOLR-8776) :
solr/core/src/java/org/apache/solr/handler/component/QueryComponent.java
+      if (query instanceof AbstractReRankQuery){
+        topNGroups = cmd.getOffset() +
((AbstractReRankQuery)query).getReRankDocs();
+      } else {
+        topNGroups = cmd.getOffset() + cmd.getLen();






-----
--Ilay
--
Sent from: http://lucene.472066.n3.nabble.com/Solr-User-f472068.html