Package org.apache.sysds.runtime.compress.estim

Class AComEst

  • Direct Known Subclasses:
    ComEstExact, ComEstSample
    public abstract class AComEst
extends Object
    Main abstract class for estimating size of compressions on columns.

    • Method Detail

      • computeCompressedSizeInfos

        public final CompressedSizeInfo computeCompressedSizeInfos​(int k)
        Multi threaded version of extracting compression size info
        Parameters:
        k - The concurrency degree.
        Returns:
        The Compression Size info of each Column compressed isolated.

      • getColGroupInfo

        public final CompressedSizeInfoColGroup getColGroupInfo​(IColIndex colIndexes)
        Method for extracting Compressed Size Info of specified columns, together in a single ColGroup
        Parameters:
        colIndexes - The columns to group together inside a ColGroup
        Returns:
        The CompressedSizeInformation associated with the selected ColGroups.

      • getColGroupInfo

        public abstract CompressedSizeInfoColGroup getColGroupInfo​(IColIndex colIndexes,
                                                           int estimate,
                                                           int nrUniqueUpperBound)
        A method to extract the Compressed Size Info for a given list of columns, This method further limits the estimated number of unique values, since in some cases the estimated number of uniques is estimated higher than the number estimated in sub groups of the given colIndexes.
        Parameters:
        colIndexes - The columns to extract compression information from
        estimate - An estimate of number of unique elements in these columns
        nrUniqueUpperBound - The upper bound of unique elements allowed in the estimate, can be calculated from the number of unique elements estimated in sub columns multiplied together. This is flexible in the sense that if the sample is small then this unique can be manually edited like in CoCodeCostMatrixMult.
        Returns:
        The CompressedSizeInfoColGroup for the given column indexes.

      • getDeltaColGroupInfo

        public final CompressedSizeInfoColGroup getDeltaColGroupInfo​(IColIndex colIndexes)
        Method for extracting info of specified columns as delta encodings (delta from previous rows values)
        Parameters:
        colIndexes - The columns to group together inside a ColGroup
        Returns:
        The CompressedSizeInformation assuming delta encoding of the column.

      • getDeltaColGroupInfo

        public abstract CompressedSizeInfoColGroup getDeltaColGroupInfo​(IColIndex colIndexes,
                                                                int estimate,
                                                                int nrUniqueUpperBound)
        A method to extract the Compressed Size Info for a given list of columns, This method further limits the estimated number of unique values, since in some cases the estimated number of uniques is estimated higher than the number estimated in sub groups of the given colIndexes. The Difference for this method is that it extract the values as delta values from the matrix block input.
        Parameters:
        colIndexes - The columns to extract compression information from
        estimate - An estimate of number of unique delta elements in these columns
        nrUniqueUpperBound - The upper bound of unique elements allowed in the estimate, can be calculated from the number of unique elements estimated in sub columns multiplied together. This is flexible in the sense that if the sample is small then this unique can be manually edited like in CoCodeCostMatrixMult.
        Returns:
        The CompressedSizeInfoColGroup for the given column indexes.

      • combine

        public final CompressedSizeInfoColGroup combine​(CompressedSizeInfoColGroup g1,
                                                CompressedSizeInfoColGroup g2)
        combine two analyzed column groups together. without materializing the dictionaries of either side. if the number of distinct elements in both sides multiplied is larger than Integer, return null. If either side was constructed without analysis then fall back to default materialization of double arrays. O
        Parameters:
        g1 - First group
        g2 - Second group
        Returns:
        A combined compressed size estimation for the group.

      • combine

        public final CompressedSizeInfoColGroup combine​(IColIndex combinedColumns,
                                                CompressedSizeInfoColGroup g1,
                                                CompressedSizeInfoColGroup g2)
        Combine two analyzed column groups together. without materializing the dictionaries of either side. if the number of distinct elements in both sides multiplied is larger than Integer, return null. If either side was constructed without analysis then fall back to default materialization of double arrays.
        Parameters:
        combinedColumns - The combined column indexes.
        g1 - First group
        g2 - Second group
        Returns:
        A combined compressed size estimation for the columns specified using the combining algorithm

      • clearNNZ

        public void clearNNZ()
        Clear the pointer to the materialized list of nnz in columns