org.apache.sysds.runtime.compress.colgroup

Class ColGroupFactory

java.lang.Object

org.apache.sysds.runtime.compress.colgroup.ColGroupFactory

public class ColGroupFactory
extends Object

Factory pattern for constructing ColGroups.

Constructor Summary

Constructors

Constructor and Description
ColGroupFactory()

Method Summary

Modifier and Type	Method and Description
static List<ColGroup>	assignColumns(int numCols, ColGroup[] colGroups, MatrixBlock rawBlock, CompressionSettings compSettings) Method for producing the final ColGroupList stored inside the CompressedMatrixBlock.
static ColGroup	compress(int[] colIndexes, int rlen, ABitmap ubm, ColGroup.CompressionType compType, CompressionSettings cs, MatrixBlock rawMatrixBlock) Method for compressing an ColGroup.
static ColGroup[]	compressColGroups(MatrixBlock in, HashMap<Integer,Double> compRatios, List<int[]> groups, CompressionSettings compSettings, int k) The actual compression method, that handles the logic of compressing multiple columns together.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

ColGroupFactory

public ColGroupFactory()

Method Detail

compressColGroups

public static ColGroup[] compressColGroups(MatrixBlock in,
                                           HashMap<Integer,Double> compRatios,
                                           List<int[]> groups,
                                           CompressionSettings compSettings,
                                           int k)

The actual compression method, that handles the logic of compressing multiple columns together. This method also have the responsibility of correcting any estimation errors previously made.

Parameters:

in - The input matrix, that could have been transposed if CompSettings was set to do that

compRatios - The previously computed Compression ratings of individual col indexes.

groups - The column groups to consider compressing together.

compSettings - The compression settings to construct the compression based on.

k - The degree of parallelism used.

Returns:

A Resulting array of ColGroups, containing the compressed information from the input matrix block.

compress

public static ColGroup compress(int[] colIndexes,
                                int rlen,
                                ABitmap ubm,
                                ColGroup.CompressionType compType,
                                CompressionSettings cs,
                                MatrixBlock rawMatrixBlock)

Method for compressing an ColGroup.

Parameters:

colIndexes - The Column indexes to compress

rlen - The number of rows in the columns

ubm - The Bitmap containing all the data needed for the compression (unless Uncompressed ColGroup)

compType - The CompressionType selected

cs - The compression Settings used for the given compression

rawMatrixBlock - The copy of the original input (maybe transposed) MatrixBlock

Returns:

A Compressed ColGroup

assignColumns

public static List<ColGroup> assignColumns(int numCols,
                                           ColGroup[] colGroups,
                                           MatrixBlock rawBlock,
                                           CompressionSettings compSettings)

Method for producing the final ColGroupList stored inside the CompressedMatrixBlock. TODO Redesign this method such that it does not utilize the null pointers to decide on which ColGroups should be incompressable. This is done by changing both this method and compressColGroup inside this class.

Parameters:

numCols - The number of columns in input matrix

colGroups - The colgroups made to assign

rawBlock - The (maybe transposed) original MatrixBlock

compSettings - The Compressionsettings used.

Returns:

return the final ColGroupList.