-
Notifications
You must be signed in to change notification settings - Fork 64
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Reciprocal Rank Fusion (RRF) normalization technique in hybrid query #874
Open
Johnsonisaacn
wants to merge
5
commits into
opensearch-project:feature/rrf-score-normalization-v2
Choose a base branch
from
Johnsonisaacn:RRF
base: feature/rrf-score-normalization-v2
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Open
Changes from all commits
Commits
Show all changes
5 commits
Select commit
Hold shift + click to select a range
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
39 changes: 39 additions & 0 deletions
39
src/main/java/org/opensearch/neuralsearch/processor/NormalizationExecuteDTO.java
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,39 @@ | ||
| /* | ||
| * Copyright OpenSearch Contributors | ||
| * SPDX-License-Identifier: Apache-2.0 | ||
| */ | ||
| package org.opensearch.neuralsearch.processor; | ||
|
|
||
| import lombok.AllArgsConstructor; | ||
| import lombok.Builder; | ||
| import lombok.Getter; | ||
| import lombok.NonNull; | ||
| import org.opensearch.common.Nullable; | ||
| import org.opensearch.neuralsearch.processor.combination.ScoreCombinationTechnique; | ||
| import org.opensearch.neuralsearch.processor.normalization.ScoreNormalizationTechnique; | ||
| import org.opensearch.search.fetch.FetchSearchResult; | ||
| import org.opensearch.search.query.QuerySearchResult; | ||
|
|
||
| import java.util.List; | ||
| import java.util.Optional; | ||
|
|
||
| /** | ||
| * DTO object to hold data required for score normalization passed to execute() function | ||
| * in NormalizationProcessorWorkflow. Field rankConstant | ||
| * used in RRF but will be null for other normalization techniques | ||
| */ | ||
| @AllArgsConstructor | ||
| @Builder | ||
| @Getter | ||
| public class NormalizationExecuteDTO { | ||
| @NonNull | ||
| private List<QuerySearchResult> querySearchResults; | ||
| @NonNull | ||
| private Optional<FetchSearchResult> fetchSearchResultOptional; | ||
| @NonNull | ||
| private ScoreNormalizationTechnique normalizationTechnique; | ||
| @NonNull | ||
| private ScoreCombinationTechnique combinationTechnique; | ||
| @Nullable | ||
| private int rankConstant; | ||
| } | ||
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
29 changes: 29 additions & 0 deletions
29
src/main/java/org/opensearch/neuralsearch/processor/NormalizeScoresDTO.java
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,29 @@ | ||
| /* | ||
| * Copyright OpenSearch Contributors | ||
| * SPDX-License-Identifier: Apache-2.0 | ||
| */ | ||
| package org.opensearch.neuralsearch.processor; | ||
|
|
||
| import lombok.AllArgsConstructor; | ||
| import lombok.Builder; | ||
| import lombok.Getter; | ||
| import lombok.NonNull; | ||
| import org.opensearch.common.Nullable; | ||
| import org.opensearch.neuralsearch.processor.normalization.ScoreNormalizationTechnique; | ||
|
|
||
| import java.util.List; | ||
|
|
||
| /** | ||
| * DTO object to hold data required for score normalization. | ||
| */ | ||
| @AllArgsConstructor | ||
| @Builder | ||
| @Getter | ||
| public class NormalizeScoresDTO { | ||
| @NonNull | ||
| private List<CompoundTopDocs> queryTopDocs; | ||
| @NonNull | ||
| private ScoreNormalizationTechnique normalizationTechnique; | ||
| @Nullable | ||
| private int rankConstant; | ||
| } |
149 changes: 149 additions & 0 deletions
149
src/main/java/org/opensearch/neuralsearch/processor/RRFProcessor.java
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,149 @@ | ||
| /* | ||
| * Copyright OpenSearch Contributors | ||
| * SPDX-License-Identifier: Apache-2.0 | ||
| */ | ||
| package org.opensearch.neuralsearch.processor; | ||
|
|
||
| import static org.opensearch.neuralsearch.search.util.HybridSearchResultFormatUtil.isHybridQueryStartStopElement; | ||
|
|
||
| import java.util.stream.Collectors; | ||
|
|
||
| import java.util.List; | ||
| import java.util.Objects; | ||
| import java.util.Optional; | ||
|
|
||
| import org.opensearch.neuralsearch.processor.combination.ScoreCombinationTechnique; | ||
| import org.opensearch.neuralsearch.processor.normalization.ScoreNormalizationTechnique; | ||
| import org.opensearch.search.fetch.FetchSearchResult; | ||
| import org.opensearch.search.query.QuerySearchResult; | ||
|
|
||
| import org.opensearch.action.search.QueryPhaseResultConsumer; | ||
| import org.opensearch.action.search.SearchPhaseContext; | ||
| import org.opensearch.action.search.SearchPhaseName; | ||
| import org.opensearch.action.search.SearchPhaseResults; | ||
| import org.opensearch.search.SearchPhaseResult; | ||
| import org.opensearch.search.internal.SearchContext; | ||
| import org.opensearch.search.pipeline.SearchPhaseResultsProcessor; | ||
|
|
||
| import lombok.AllArgsConstructor; | ||
| import lombok.extern.log4j.Log4j2; | ||
|
|
||
| /** | ||
| * Processor for implementing reciprocal rank fusion technique on post | ||
| * query search results. Updates query results with | ||
| * normalized and combined scores for next phase (typically it's FETCH) | ||
| * by using ranks from individual subqueries to calculate 'normalized' | ||
| * scores before combining results from subqueries into final results | ||
| */ | ||
| @Log4j2 | ||
| @AllArgsConstructor | ||
| public class RRFProcessor implements SearchPhaseResultsProcessor { | ||
| public static final String TYPE = "score-ranker-processor"; | ||
|
|
||
| private final String tag; | ||
| private final String description; | ||
| private final ScoreNormalizationTechnique normalizationTechnique; | ||
| private final ScoreCombinationTechnique combinationTechnique; | ||
| private final NormalizationProcessorWorkflow normalizationWorkflow; | ||
| private final int rankConstant; | ||
|
|
||
| /** | ||
| * Method abstracts functional aspect of score normalization and score combination. Exact methods for each processing stage | ||
| * are set as part of class constructor | ||
| * @param searchPhaseResult {@link SearchPhaseResults} DTO that has query search results. Results will be mutated as part of this method execution | ||
| * @param searchPhaseContext {@link SearchContext} | ||
| */ | ||
| @Override | ||
| public <Result extends SearchPhaseResult> void process( | ||
| final SearchPhaseResults<Result> searchPhaseResult, | ||
| final SearchPhaseContext searchPhaseContext | ||
| ) { | ||
| if (shouldSkipProcessor(searchPhaseResult)) { | ||
| log.debug("Query results are not compatible with RRF processor"); | ||
| return; | ||
| } | ||
| List<QuerySearchResult> querySearchResults = getQueryPhaseSearchResults(searchPhaseResult); | ||
| Optional<FetchSearchResult> fetchSearchResult = getFetchSearchResults(searchPhaseResult); | ||
|
|
||
| // make data transfer object to pass in, execute will get object with 4 or 5 fields, depending | ||
| // on coming from NormalizationProcessor or RRFProcessor | ||
| NormalizationExecuteDTO normalizationExecuteDTO = NormalizationExecuteDTO.builder() | ||
| .querySearchResults(querySearchResults) | ||
| .fetchSearchResultOptional(fetchSearchResult) | ||
| .normalizationTechnique(normalizationTechnique) | ||
| .combinationTechnique(combinationTechnique) | ||
| .rankConstant(rankConstant) | ||
| .build(); | ||
| normalizationWorkflow.execute(normalizationExecuteDTO); | ||
| } | ||
|
|
||
| @Override | ||
| public SearchPhaseName getBeforePhase() { | ||
| return SearchPhaseName.QUERY; | ||
| } | ||
|
|
||
| @Override | ||
| public SearchPhaseName getAfterPhase() { | ||
| return SearchPhaseName.FETCH; | ||
| } | ||
|
|
||
| @Override | ||
| public String getType() { | ||
| return TYPE; | ||
| } | ||
|
|
||
| @Override | ||
| public String getTag() { | ||
| return tag; | ||
| } | ||
|
|
||
| @Override | ||
| public String getDescription() { | ||
| return description; | ||
| } | ||
|
|
||
| @Override | ||
| public boolean isIgnoreFailure() { | ||
| return false; | ||
| } | ||
|
|
||
| private <Result extends SearchPhaseResult> boolean shouldSkipProcessor(SearchPhaseResults<Result> searchPhaseResult) { | ||
| if (Objects.isNull(searchPhaseResult) || !(searchPhaseResult instanceof QueryPhaseResultConsumer)) { | ||
| return true; | ||
| } | ||
|
|
||
| QueryPhaseResultConsumer queryPhaseResultConsumer = (QueryPhaseResultConsumer) searchPhaseResult; | ||
| return queryPhaseResultConsumer.getAtomicArray().asList().stream().filter(Objects::nonNull).noneMatch(this::isHybridQuery); | ||
| } | ||
|
|
||
| /** | ||
| * Return true if results are from hybrid query. | ||
| * @param searchPhaseResult | ||
| * @return true if results are from hybrid query | ||
| */ | ||
| private boolean isHybridQuery(final SearchPhaseResult searchPhaseResult) { | ||
| // check for delimiter at the end of the score docs. | ||
| return Objects.nonNull(searchPhaseResult.queryResult()) | ||
| && Objects.nonNull(searchPhaseResult.queryResult().topDocs()) | ||
| && Objects.nonNull(searchPhaseResult.queryResult().topDocs().topDocs.scoreDocs) | ||
| && searchPhaseResult.queryResult().topDocs().topDocs.scoreDocs.length > 0 | ||
| && isHybridQueryStartStopElement(searchPhaseResult.queryResult().topDocs().topDocs.scoreDocs[0]); | ||
| } | ||
|
|
||
| private <Result extends SearchPhaseResult> List<QuerySearchResult> getQueryPhaseSearchResults( | ||
| final SearchPhaseResults<Result> results | ||
| ) { | ||
| return results.getAtomicArray() | ||
| .asList() | ||
| .stream() | ||
| .map(result -> result == null ? null : result.queryResult()) | ||
| .collect(Collectors.toList()); | ||
| } | ||
|
|
||
| private <Result extends SearchPhaseResult> Optional<FetchSearchResult> getFetchSearchResults( | ||
| final SearchPhaseResults<Result> searchPhaseResults | ||
| ) { | ||
| Optional<Result> optionalFirstSearchPhaseResult = searchPhaseResults.getAtomicArray().asList().stream().findFirst(); | ||
| return optionalFirstSearchPhaseResult.map(SearchPhaseResult::fetchResult); | ||
| } | ||
| } |
35 changes: 35 additions & 0 deletions
35
.../java/org/opensearch/neuralsearch/processor/combination/RRFScoreCombinationTechnique.java
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,35 @@ | ||
| /* | ||
| * Copyright OpenSearch Contributors | ||
| * SPDX-License-Identifier: Apache-2.0 | ||
| */ | ||
| package org.opensearch.neuralsearch.processor.combination; | ||
|
|
||
| import lombok.ToString; | ||
| import lombok.extern.log4j.Log4j2; | ||
|
|
||
| import java.util.Map; | ||
|
|
||
| @Log4j2 | ||
| /** | ||
| * Abstracts combination of scores based on reciprocal rank fusion algorithm | ||
| */ | ||
| @ToString(onlyExplicitlyIncluded = true) | ||
|
|
||
| public class RRFScoreCombinationTechnique implements ScoreCombinationTechnique { | ||
| @ToString.Include | ||
| public static final String TECHNIQUE_NAME = "rrf"; | ||
|
|
||
| // Not currently using weights for RRF, no need to modify or verify these params | ||
| public RRFScoreCombinationTechnique(final Map<String, Object> params, final ScoreCombinationUtil combinationUtil) { | ||
| ; | ||
| } | ||
|
|
||
| @Override | ||
| public float combine(final float[] scores) { | ||
| float sumScores = 0.0f; | ||
| for (float score : scores) { | ||
| sumScores += score; | ||
| } | ||
| return sumScores; | ||
| } | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
the rank constant should be a parameter for RRF normalization technique class, not the high level field in this DTO. You can pass it as param to RRF normalization technique constructor while creating its instance in a factory class