Mailing-List: contact dev-help@lucene.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@lucene.apache.org
Date: Fri, 7 Apr 2017 15:16:41 +0000 (UTC)
From: "Alessandro Benedetti (JIRA)" <jira@apache.org>
To: dev@lucene.apache.org
Message-ID: <JIRA.13062461.1491578175000.236169.1491578201520@Atlassian.JIRA>
In-Reply-To: <JIRA.13062461.1491578175000@Atlassian.JIRA>
References: <JIRA.13062461.1491578175000@Atlassian.JIRA> <JIRA.13062461.1491578175521@jira-lw-us.apache.org>
Subject: [jira] [Created] (SOLR-10448) [LTR] Feature vector caching
 separation
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Fri, 07 Apr 2017 15:16:46 -0000

Alessandro Benedetti created SOLR-10448:
-------------------------------------------

             Summary: [LTR] Feature vector caching separation
                 Key: SOLR-10448
                 URL: https://issues.apache.org/jira/browse/SOLR-10448
             Project: Solr
          Issue Type: Improvement
      Security Level: Public (Default Security Level. Issues are Public)
            Reporter: Alessandro Benedetti
            Priority: Minor


We have different type of features, and their nature is different :
SolrFeature -> can be query dependent / query independent
FieldValueFeature -> query independent
Value Feature -> query level
OriginalScoreFeature - > query dependent

Intuition
Let's see when the extracted value for a feature change :

Query independent ( document level features) -> indexing time, every commit may change the feature value and possibly invaldiate the cache entry

query dependent -> both request parameters and index can change the feature value

query level -> only request parameters can change the value, it is not affected by the index

Caching
let's see the benefits of having 3 separate caches ( so the feature vector components will be separated) :

Query independent feature Cache -> simply changing the request parameters will not invalidate the cache entry, only a commit will cause the autowarming

Query Dependent Feature Cache -> this will behave like the current cache, so changing params you will cache a different feature vector

Query level Cache ->the index is not involved at all

I have not yet investigated how caching internally works in the LTR plugin so i don't know how much effort would imply.


--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

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