Class CompressionSettings
java.lang.Object
org.apache.sysds.runtime.compress.CompressionSettings
Compression Settings class, used as a bundle of parameters inside the Compression framework. See
CompressionSettingsBuilder for default non static parameters.
-
Field Summary
FieldsModifier and TypeFieldDescriptionfinal booleanShare DDC Dictionaries between ColGroups.static final intSize of the blocks used in a blocked bitmap representation.final doubleA Cocode parameter that differ in behavior based on compression method, in general it is a value that reflects aggressively likely coCoding is used.The selected method for column partitioning used in CoCoding compressed columnsThe cost computation type for the compressionThe sample type used for samplingfinal booleanIs a spark instructionfinal booleanTrue if lossy compression is enabledfinal intThe maximum number of columns CoCoded allowedfinal intThe maximum size of the sample extracted.final doubleThe minimum compression ratio to achieve.final intThe minimum size of the sample extracted.static intParallelization threshold for DDC compressionfinal booleanstatic booleanif the settings have been logged already.final doubleThe sampling ratio power to use when choosing sample size.final doubleThe sampling ratio used when choosing ColGroups.final double[]The sorting type used in sorting/joining offsets to create SDC groupsfinal intIf the seed is -1 then the system used system millisecond time and class hash for seeding.final booleanSorting of values by physical length helps by 10-20%, especially for serial, while slight performance decrease for parallel incl multi-threaded, hence not applied for distributed operations (also because compression time + garbage collection increases)booleanTranspose input matrix, to optimize access when extracting bitmaps.final StringBoolean specifying which transpose setting is used, can be auto, true or falsefinal EnumSet<AColGroup.CompressionType>Valid Compressions List, containing the ColGroup CompressionTypes that are allowed to be used for the compression Default is to always allow for Uncompromisable ColGroup. -
Method Summary
-
Field Details
-
PAR_DDC_THRESHOLD
public static int PAR_DDC_THRESHOLDParallelization threshold for DDC compression -
BITMAP_BLOCK_SZ
public static final int BITMAP_BLOCK_SZSize of the blocks used in a blocked bitmap representation. Note it is exactly Character.MAX_VALUE. This is not Character max value + 1 because it breaks the offsets in cases with fully dense values.- See Also:
-
sortTuplesByFrequency
public final boolean sortTuplesByFrequencySorting of values by physical length helps by 10-20%, especially for serial, while slight performance decrease for parallel incl multi-threaded, hence not applied for distributed operations (also because compression time + garbage collection increases) -
samplingRatio
public final double samplingRatioThe sampling ratio used when choosing ColGroups. Note that, default behavior is to use exact estimator if the number of elements is below 1000. DEPRECATED -
samplePower
public final double samplePowerThe sampling ratio power to use when choosing sample size. This is used in accordance to the function: sampleSize += nRows^samplePower; The value is bounded to be in the range of 0 to 1, 1 giving a sample size of everything, and 0 adding 1. -
transposeInput
Boolean specifying which transpose setting is used, can be auto, true or false -
seed
public final int seedIf the seed is -1 then the system used system millisecond time and class hash for seeding. -
lossy
public final boolean lossyTrue if lossy compression is enabled -
columnPartitioner
The selected method for column partitioning used in CoCoding compressed columns -
costComputationType
The cost computation type for the compression -
maxColGroupCoCode
public final int maxColGroupCoCodeThe maximum number of columns CoCoded allowed -
coCodePercentage
public final double coCodePercentageA Cocode parameter that differ in behavior based on compression method, in general it is a value that reflects aggressively likely coCoding is used. -
validCompressions
Valid Compressions List, containing the ColGroup CompressionTypes that are allowed to be used for the compression Default is to always allow for Uncompromisable ColGroup. -
minimumSampleSize
public final int minimumSampleSizeThe minimum size of the sample extracted. -
maxSampleSize
public final int maxSampleSizeThe maximum size of the sample extracted. -
estimationType
The sample type used for sampling -
transposed
public boolean transposedTranspose input matrix, to optimize access when extracting bitmaps. This setting is changed inside the script based on the transposeInput setting. This is intentionally left as a mutable value, since the transposition of the input matrix is decided in phase 3. -
minimumCompressionRatio
public final double minimumCompressionRatioThe minimum compression ratio to achieve. -
isInSparkInstruction
public final boolean isInSparkInstructionIs a spark instruction -
sdcSortType
The sorting type used in sorting/joining offsets to create SDC groups -
printedStatus
public static boolean printedStatusif the settings have been logged already. -
scaleFactors
public final double[] scaleFactors -
preferDeltaEncoding
public final boolean preferDeltaEncoding
-
-
Method Details