Support sub query raw scores in Hybrid Search Response #1369

owaiskazi19 · 2025-06-03T21:46:31Z

Description

Support sub query raw scores in Hybrid Search Response through getFetchSubPhases extension point.
Also, added a small improvement for RRF processors while working on this.

Flag to enable sub query scores while creating the pipeline

{
  "description": "Post processor for hybrid search",
  "phase_results_processors": [
    {
      "normalization-processor": {
        "normalization": {
          "technique": "z_score"
        },
        "sub-query-scores": true,
        "combination": {
          "technique": "arithmetic_mean",
          "parameters": {
            "weights": [
              0.3,
              0.4,
              0.3
            ]
          }
        }
      }
    }
  ]
}

Response

"hits": {
        "total": {
            "value": 2,
            "relation": "eq"
        },
        "max_score": 0.016393442,
        "hits": [
            {
                "_index": "my-nlp-index1",
                "_id": "3",
                "_score": 0.016393442,
                "_source": {
                    "passage_text": "zoo keeper here",
                    "id": "s3"
                },
                "fields": {
                    "hybridization_sub_query_scores": [
                        0.0,
                        0.0,
                        0.26156494
                    ]
                }
            },
            {
                "_index": "my-nlp-index1",
                "_id": "1",
                "_score": 0.016393442,
                "_source": {
                    "passage_text": "hello hi what",
                    "id": "s1"
                },
                "fields": {
                    "hybridization_sub_query_scores": [
                        0.13076457,
                        0.0,
                        0.0
                    ]
                }
            }
        ]
    }
}

Related Issues

Resolves #1294, #1180 & #1419

Check List

New functionality includes testing.
New functionality has been documented.
API changes companion pull request created.
Commits are signed per the DCO using --signoff.
Public documentation issue/PR created.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

owaiskazi19 · 2025-06-03T21:50:54Z

...main/java/org/opensearch/neuralsearch/processor/normalization/RRFNormalizationTechnique.java

    ) {
-        for (ScoreDoc scoreDoc : topDocs.scoreDocs) {
-            float normalizedScore = calculateNormalizedScore(Arrays.asList(topDocs.scoreDocs).indexOf(scoreDoc));


This was causing O(N2), changed it to index based reference

src/main/java/org/opensearch/neuralsearch/processor/NormalizationProcessorWorkflow.java

vibrantvarun

Great job @owaiskazi19 on creating the logic for capturing subquery scores.

vibrantvarun · 2025-07-08T02:50:08Z

src/main/java/org/opensearch/neuralsearch/common/MinClusterVersionUtil.java

@@ -58,6 +59,10 @@ public static boolean isClusterOnOrAfterMinReqVersionForStatCategoryFiltering()
        return NeuralSearchClusterUtil.instance().getClusterMinVersion().onOrAfter(MINIMAL_SUPPORTED_VERSION_STATS_CATEGORY_FILTERING);
    }

+    public static boolean isClusterOnOrAfterMinReqVersionForSubQuerySupport() {


We need one BWC test to validate this condition does not break anything.

vibrantvarun · 2025-07-08T02:52:46Z

src/main/java/org/opensearch/neuralsearch/processor/HybridScoreRegistry.java

+import java.util.Map;
+
+/**
+ * Registry to store hybrid score for each search context


As per my understanding, there is only one searchcontext per search. Therefore, can we reframe this javadoc in a specific way like "Registry to store hybrid score for each document" ?

In the later part, I see we are using searchPhaseContext. SearchPhaseContext can be multiple like after the QP , then after the FP as well.

vibrantvarun · 2025-07-08T02:57:36Z

src/main/java/org/opensearch/neuralsearch/processor/HybridizationFetchSubPhase.java

+    @Override
+    public FetchSubPhaseProcessor getProcessor(FetchContext fetchContext) throws IOException {
+        // Check if inner hits are present
+        boolean hasInnerHits = fetchContext.innerHits() != null && !fetchContext.innerHits().getInnerHits().isEmpty();


nit: Do we really need to check !fetchContext.innerHits().getInnerHits().isEmpty() as well?

I am assuming if fetchContext.innerHits is true then it should work?

fetchContext.innerHits() returns an InnerHitsContext object — which may be non-null, but contain no inner hits. Then hasInnerHits will be true even if there are no actual inner hits to process

vibrantvarun · 2025-07-08T02:59:55Z

src/main/java/org/opensearch/neuralsearch/processor/HybridizationFetchSubPhase.java

+
+                    if (subqueryScores != null) {
+                        // Add it as a field rather than modifying _source
+                        List<Object> hybridScores = new ArrayList<>(subqueryScores.length);


vibrantvarun · 2025-07-08T03:02:23Z

src/main/java/org/opensearch/neuralsearch/processor/NormalizationProcessorWorkflow.java

+            // Check for all the conditions
+            boolean shouldAddHybridScores = subqueryScores != null
+                && documentFields.containsKey(SUB_QUERY_SCORES_NAME) == false
+                && isClusterOnOrAfterMinReqVersionForSubQuerySupport()


Shouldn't this whole check should be wrapped under

if (isClusterOnOrAfterMinReqVersionForSubQuerySupport() && hasInnerHits ==false )

Isn't it the same thing checking condtion1 && condition2 and then condition3 && condition4 or all together like condtion1 && condition2 && condtion3 && condtion4 ?

vibrantvarun · 2025-07-08T03:04:07Z

src/main/java/org/opensearch/neuralsearch/processor/NormalizationProcessorWorkflow.java

+
+            // Check for all the conditions
+            boolean shouldAddHybridScores = subqueryScores != null
+                && documentFields.containsKey(SUB_QUERY_SCORES_NAME) == false


Could you add a BWC test when out of 3 nodes 1 node is running on newer version and other 2 are running older version. Then request for hybridization scores.

.../java/org/opensearch/neuralsearch/processor/normalization/L2ScoreNormalizationTechnique.java

src/main/java/org/opensearch/neuralsearch/processor/normalization/ScoreNormalizer.java

martin-gaievski

Added few comments after recently added commits

src/main/java/org/opensearch/neuralsearch/processor/HybridScoreRegistry.java

.../java/org/opensearch/neuralsearch/processor/normalization/L2ScoreNormalizationTechnique.java

src/main/java/org/opensearch/neuralsearch/processor/HybridizationFetchSubPhase.java

.../java/org/opensearch/neuralsearch/processor/normalization/L2ScoreNormalizationTechnique.java

src/main/java/org/opensearch/neuralsearch/processor/HybridScoreRegistry.java

codecov · 2025-07-18T17:21:35Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 0.00%. Comparing base (8c5fb57) to head (2ca0411).
Report is 2 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff              @@
##               main   #1369       +/-   ##
============================================
- Coverage     79.95%       0   -79.96%     
============================================
  Files           160       0      -160     
  Lines          8584       0     -8584     
  Branches       1396       0     -1396     
============================================
- Hits           6863       0     -6863     
+ Misses         1186       0     -1186     
+ Partials        535       0      -535

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

owaiskazi19 · 2025-07-18T21:46:03Z

#1416 is failing most of the times

heemin32 · 2025-07-18T22:14:57Z

I can help with retrying only failed tests.

owaiskazi19 · 2025-07-21T20:03:47Z

Will run benchmarking again after #1457 is merged

Signed-off-by: Owais <[email protected]>

…hits are enabled Signed-off-by: Owais <[email protected]>

Signed-off-by: Owais <[email protected]>

github-actions bot added the enhancement label Jun 3, 2025

owaiskazi19 commented Jun 3, 2025

View reviewed changes

owaiskazi19 force-pushed the hybrid-scores branch 3 times, most recently from 07b39e7 to 2eb0d1f Compare June 11, 2025 08:34

owaiskazi19 force-pushed the hybrid-scores branch 3 times, most recently from bc183c6 to 70ec7ec Compare June 23, 2025 17:05

ryanbogan reviewed Jun 27, 2025

View reviewed changes

src/main/java/org/opensearch/neuralsearch/processor/NormalizationProcessorWorkflow.java Outdated Show resolved Hide resolved

owaiskazi19 force-pushed the hybrid-scores branch 5 times, most recently from 63a54ff to 09664ab Compare July 1, 2025 22:03

owaiskazi19 changed the title ~~[DRAFT] Support sub query raw scores in Hybrid Search Response~~ Support sub query raw scores in Hybrid Search Response Jul 2, 2025

owaiskazi19 marked this pull request as ready for review July 2, 2025 20:26

owaiskazi19 requested review from heemin32, navneet1v, VijayanB, vamshin, jmazanec15, naveentatikonda, junqiu-lei, martin-gaievski, sean-zheng-amazon, model-collapse, zane-neo, vibrantvarun, zhichao-aws and yuye-aws as code owners July 2, 2025 20:26

vibrantvarun reviewed Jul 8, 2025

View reviewed changes

martin-gaievski reviewed Jul 8, 2025

View reviewed changes

owaiskazi19 force-pushed the hybrid-scores branch 4 times, most recently from d7173ec to 2ca0411 Compare July 17, 2025 22:08

owaiskazi19 closed this Jul 18, 2025

owaiskazi19 reopened this Jul 18, 2025

owaiskazi19 closed this Jul 18, 2025

owaiskazi19 reopened this Jul 18, 2025

owaiskazi19 force-pushed the hybrid-scores branch from 2ca0411 to 1e303f1 Compare July 18, 2025 19:22

owaiskazi19 added 13 commits July 21, 2025 13:30

Initial sub query scores in Hybrid Search Response

83f758d

Signed-off-by: Owais <[email protected]>

Fixed single shard case

ae5603a

Signed-off-by: Owais <[email protected]>

Added more tests and fixed BWC test

71deb58

Signed-off-by: Owais <[email protected]>

Added logic to disable sub query scores when sort, explain and inner …

97afe95

…hits are enabled Signed-off-by: Owais <[email protected]>

Refactored sub query scores flag

92f3193

Signed-off-by: Owais <[email protected]>

Added more tests for the new flag

57bf8a5

Signed-off-by: Owais <[email protected]>

Fix multiple shards scenario and added more tests

101e640

Signed-off-by: Owais <[email protected]>

Fixed tests

9639df5

Signed-off-by: Owais <[email protected]>

Added more integ tests

0d6d2be

Signed-off-by: Owais <[email protected]>

Addressed PR comments

d77d88b

Signed-off-by: Owais <[email protected]>

Fixed integ tests

68f5596

Signed-off-by: Owais <[email protected]>

Added BWC test and upgraded bwc version

a68bfb7

Signed-off-by: Owais <[email protected]>

Fixed failed semanticIT bwc test

101c629

Signed-off-by: Owais <[email protected]>

owaiskazi19 force-pushed the hybrid-scores branch from 1e303f1 to 101c629 Compare July 21, 2025 20:31

Fixed integ test

f1a6b7c

Signed-off-by: Owais <[email protected]>

Support sub query raw scores in Hybrid Search Response #1369

Are you sure you want to change the base?

Support sub query raw scores in Hybrid Search Response #1369

Conversation

owaiskazi19 commented Jun 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Related Issues

Check List

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

vibrantvarun left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

martin-gaievski left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

codecov bot commented Jul 18, 2025

Codecov Report

Uh oh!

owaiskazi19 commented Jul 18, 2025

Uh oh!

heemin32 commented Jul 18, 2025

Uh oh!

owaiskazi19 commented Jul 21, 2025

Uh oh!

Uh oh!

owaiskazi19 commented Jun 3, 2025 •

edited

Loading