Class AbstractKnnVectorQuery

java.lang.Object
org.apache.lucene.search.Query
org.apache.lucene.search.AbstractKnnVectorQuery
Direct Known Subclasses:
KnnByteVectorQuery, KnnFloatVectorQuery

abstract class AbstractKnnVectorQuery extends Query
Uses KnnVectorsReader.search(java.lang.String, float[], org.apache.lucene.search.KnnCollector, org.apache.lucene.util.Bits) to perform nearest neighbour search.

This query also allows for performing a kNN search subject to a filter. In this case, it first executes the filter for each leaf, then chooses a strategy dynamically:

  • If the filter cost is less than k, just execute an exact search
  • Otherwise run a kNN search subject to the filter
  • If the kNN search visits too many vectors without completing, stop and run an exact search
  • Field Details

    • NO_RESULTS

      private static final TopDocs NO_RESULTS
    • field

      protected final String field
    • k

      protected final int k
    • filter

      protected final Query filter
  • Constructor Details

    • AbstractKnnVectorQuery

      public AbstractKnnVectorQuery(String field, int k, Query filter)
  • Method Details

    • rewrite

      public Query rewrite(IndexSearcher indexSearcher) throws IOException
      Description copied from class: Query
      Expert: called to re-write queries into primitive queries. For example, a PrefixQuery will be rewritten into a BooleanQuery that consists of TermQuerys.

      Callers are expected to call rewrite multiple times if necessary, until the rewritten query is the same as the original query.

      The rewrite process may be able to make use of IndexSearcher's executor and be executed in parallel if the executor is provided.

      Overrides:
      rewrite in class Query
      Throws:
      IOException
      See Also:
    • searchLeaf

      private TopDocs searchLeaf(LeafReaderContext ctx, Weight filterWeight, TimeLimitingKnnCollectorManager timeLimitingKnnCollectorManager) throws IOException
      Throws:
      IOException
    • getLeafResults

      private TopDocs getLeafResults(LeafReaderContext ctx, Weight filterWeight, TimeLimitingKnnCollectorManager timeLimitingKnnCollectorManager) throws IOException
      Throws:
      IOException
    • createBitSet

      private BitSet createBitSet(DocIdSetIterator iterator, Bits liveDocs, int maxDoc) throws IOException
      Throws:
      IOException
    • getKnnCollectorManager

      protected KnnCollectorManager getKnnCollectorManager(int k, IndexSearcher searcher)
    • approximateSearch

      protected abstract TopDocs approximateSearch(LeafReaderContext context, Bits acceptDocs, int visitedLimit, KnnCollectorManager knnCollectorManager) throws IOException
      Throws:
      IOException
    • createVectorScorer

      abstract VectorScorer createVectorScorer(LeafReaderContext context, FieldInfo fi) throws IOException
      Throws:
      IOException
    • exactSearch

      protected TopDocs exactSearch(LeafReaderContext context, DocIdSetIterator acceptIterator, QueryTimeout queryTimeout) throws IOException
      Throws:
      IOException
    • mergeLeafResults

      protected TopDocs mergeLeafResults(TopDocs[] perLeafResults)
      Merges all segment-level kNN results to get the index-level kNN results.

      The default implementation delegates to TopDocs.merge(int, TopDocs[]) to find the overall top k, which requires input results to be sorted.

      This method is useful for reading and / or modifying the final results as needed.

      Parameters:
      perLeafResults - array of segment-level kNN results.
      Returns:
      index-level kNN results (no constraint on their ordering).
    • createRewrittenQuery

      private Query createRewrittenQuery(IndexReader reader, TopDocs topK)
    • findSegmentStarts

      static int[] findSegmentStarts(List<LeafReaderContext> leaves, int[] docs)
    • visit

      public void visit(QueryVisitor visitor)
      Description copied from class: Query
      Recurse through the query tree, visiting any child queries.
      Specified by:
      visit in class Query
      Parameters:
      visitor - a QueryVisitor to be called by each query in the tree
    • equals

      public boolean equals(Object o)
      Description copied from class: Query
      Override and implement query instance equivalence properly in a subclass. This is required so that QueryCache works properly.

      Typically a query will be equal to another only if it's an instance of the same class and its document-filtering properties are identical to those of the other instance. Utility methods are provided for certain repetitive code.

      Specified by:
      equals in class Query
      See Also:
    • hashCode

      public int hashCode()
      Description copied from class: Query
      Override and implement query hash code properly in a subclass. This is required so that QueryCache works properly.
      Specified by:
      hashCode in class Query
      See Also:
    • getField

      public String getField()
      Returns:
      the knn vector field where the knn vector search happens.
    • getK

      public int getK()
      Returns:
      the max number of results the KnnVector search returns.
    • getFilter

      public Query getFilter()
      Returns:
      the filter that is executed before the KnnVector search happens. Only the results accepted by this filter are returned by the KnnVector search.