Class StandardFacetsAccumulator

  • Direct Known Subclasses:
    AdaptiveFacetsAccumulator, SamplingAccumulator

    public class StandardFacetsAccumulator
    extends FacetsAccumulator
    Standard implementation for FacetsAccumulator, utilizing partitions to save on memory.

    Why partitions? Because if there are say 100M categories out of which only top K are required, we must first compute value for all 100M categories (going over all documents) and only then could we select top K. This is made easier on memory by working in partitions of distinct categories: Once a values for a partition are found, we take the top K for that partition and work on the next partition, them merge the top K of both, and so forth, thereby computing top K with RAM needs for the size of a single partition rather than for the size of all the 100M categories.

    Decision on partitions size is done at indexing time, and the facet information for each partition is maintained separately.

    Implementation detail: Since facets information of each partition is maintained in a separate "category list", we can be more efficient at search time, because only the facet info for a single partition need to be read while processing that partition.

    WARNING: This API is experimental and might change in incompatible ways in the next release.
    • Field Detail

      • partitionSize

        protected int partitionSize
      • maxPartitions

        protected int maxPartitions
      • isUsingComplements

        protected boolean isUsingComplements
    • Method Detail

      • actualDocsToAccumulate

        protected ScoredDocIDs actualDocsToAccumulate​(ScoredDocIDs docids)
                                               throws IOException
        Set the actual set of documents over which accumulation should take place.

        Allows to override the set of documents to accumulate for. Invoked just before actual accumulating starts. From this point that set of documents remains unmodified. Default implementation just returns the input unchanged.

        Parameters:
        docids - candidate documents to accumulate for
        Returns:
        actual documents to accumulate for
        Throws:
        IOException
      • shouldComplement

        protected boolean shouldComplement​(ScoredDocIDs docids)
        Check if it is worth to use complements
      • getTotalCountsFactor

        protected double getTotalCountsFactor()
        Expert: factor by which counts should be multiplied when initializing the count arrays from total counts. Default implementation for this returns 1, which is a no op.
        Returns:
        a factor by which total counts should be multiplied
      • getCategoryListMap

        protected HashMap<CategoryListIterator,​Aggregator> getCategoryListMap​(FacetArrays facetArrays,
                                                                                    int partition)
                                                                             throws IOException
        Create an Aggregator and a CategoryListIterator for each and every FacetRequest. Generating a map, matching each categoryListIterator to its matching aggregator.

        If two CategoryListIterators are served by the same aggregator, a single aggregator is returned for both. NOTE: If a given category list iterator is needed with two different aggregators (e.g counting and association) - an exception is thrown as this functionality is not supported at this time.

        Throws:
        IOException