Package org.apache.sysds.runtime.transform.encode

Class ColumnEncoder

java.lang.Object

org.apache.sysds.runtime.transform.encode.ColumnEncoder

All Implemented Interfaces:

Externalizable, Serializable, Comparable<ColumnEncoder>, Encoder

Direct Known Subclasses:

ColumnEncoderBin, ColumnEncoderComposite, ColumnEncoderDummycode, ColumnEncoderFeatureHash, ColumnEncoderPassThrough, ColumnEncoderRecode, ColumnEncoderUDF

public abstract class ColumnEncoder
extends Object
implements Encoder, Comparable<ColumnEncoder>

Base class for all transform encoders providing both a row and block interface for decoding frames to matrices.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	ColumnEncoder.EncoderType

Field Summary

Fields

Modifier and Type	Field	Description
static int	APPLY_ROW_BLOCKS_PER_COLUMN
static int	BUILD_ROW_BLOCKS_PER_COLUMN

Method Summary

Modifier and Type	Method	Description
MatrixBlock	apply(CacheBlock in, MatrixBlock out, int outputCol)	Apply Functions are only used in Single Threaded or Multi-Threaded Dense context.
MatrixBlock	apply(CacheBlock in, MatrixBlock out, int outputCol, int rowStart, int blk)
void	build(CacheBlock in, double[] equiHeightMaxs)
void	build(CacheBlock in, Map<Integer,double[]> equiHeightMaxs)
void	buildPartial(FrameBlock in)	Partial build of internal data structures (e.g., in distributed spark operations).
int	compareTo(ColumnEncoder o)
List<DependencyTask<?>>	getApplyTasks(CacheBlock in, MatrixBlock out, int outputCol)
Callable<Object>	getBuildTask(CacheBlock in)
List<DependencyTask<?>>	getBuildTasks(CacheBlock in)
int	getColID()
MatrixBlock	getColMapping(FrameBlock meta)	Obtain the column mapping of encoded frames based on the passed meta data frame.
long	getEstMetaSize()
int	getEstNumDistincts()
Callable<Object>	getPartialBuildTask(CacheBlock in, int startRow, int blockSize, HashMap<Integer,Object> ret)
Callable<Object>	getPartialMergeBuildTask(HashMap<Integer,?> ret)
Set<Integer>	getSparseRowsWZeros()
boolean	isApplicable()	Indicates if this encoder is applicable, i.e, if there is a column to encode.
boolean	isApplicable(int colID)	Indicates if this encoder is applicable for the given column ID, i.e., if it is subject to this transformation.
void	mergeAt(ColumnEncoder other)	Merges another encoder, of a compatible type, in after a certain position.
void	prepareBuildPartial()	Allocates internal data structures for partial build.
void	readExternal(ObjectInput in)	Redirects the default java serialization via externalizable to our default hadoop writable serialization for efficient broadcast/rdd deserialization.
void	setColID(int colID)
void	setEstMetaSize(long estSize)
void	setEstNumDistincts(int numDistincts)
void	shiftCol(int columnOffset)
void	updateIndexRanges(long[] beginDims, long[] endDims, int colOffset)	Update index-ranges to after encoding.
void	writeExternal(ObjectOutput os)	Redirects the default java serialization via externalizable to our default hadoop writable serialization for efficient broadcast/rdd serialization.

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.apache.sysds.runtime.transform.encode.Encoder

allocateMetaData, build, getMetaData, initMetaData

Field Detail

APPLY_ROW_BLOCKS_PER_COLUMN

public static int APPLY_ROW_BLOCKS_PER_COLUMN

BUILD_ROW_BLOCKS_PER_COLUMN

public static int BUILD_ROW_BLOCKS_PER_COLUMN

Method Detail

apply

public MatrixBlock apply(CacheBlock in,
                         MatrixBlock out,
                         int outputCol)

Apply Functions are only used in Single Threaded or Multi-Threaded Dense context. That's why there is no regard for MT sparse!

Specified by:

apply in interface Encoder

Parameters:

in - Input Block

out - Output Matrix

outputCol - The output column for the given column

Returns:

same as out

apply

public MatrixBlock apply(CacheBlock in,
                         MatrixBlock out,
                         int outputCol,
                         int rowStart,
                         int blk)

isApplicable

public boolean isApplicable()

Indicates if this encoder is applicable, i.e, if there is a column to encode.

Returns:

true if a colID is set

isApplicable

public boolean isApplicable(int colID)

Indicates if this encoder is applicable for the given column ID, i.e., if it is subject to this transformation.

Parameters:

colID - column ID

Returns:

true if encoder is applicable for given column

prepareBuildPartial

public void prepareBuildPartial()

Allocates internal data structures for partial build.

Specified by:

prepareBuildPartial in interface Encoder

buildPartial

public void buildPartial(FrameBlock in)

Partial build of internal data structures (e.g., in distributed spark operations).

Specified by:

buildPartial in interface Encoder

Parameters:

in - input frame block

build

public void build(CacheBlock in,
                  double[] equiHeightMaxs)

build

public void build(CacheBlock in,
                  Map<Integer,double[]> equiHeightMaxs)

mergeAt

public void mergeAt(ColumnEncoder other)

Merges another encoder, of a compatible type, in after a certain position. Resizes as necessary. ColumnEncoders are compatible with themselves and EncoderComposite is compatible with every other ColumnEncoders. MultiColumnEncoders are compatible with every encoder

Parameters:

other - the encoder that should be merged in

updateIndexRanges

public void updateIndexRanges(long[] beginDims,
                              long[] endDims,
                              int colOffset)

Update index-ranges to after encoding. Note that only Dummycoding changes the ranges.

Specified by:

updateIndexRanges in interface Encoder

Parameters:

beginDims - begin dimensions of range

endDims - end dimensions of range

colOffset - is applied to begin and endDims

getColMapping

public MatrixBlock getColMapping(FrameBlock meta)

Obtain the column mapping of encoded frames based on the passed meta data frame.

Parameters:

meta - meta data frame block

Returns:

matrix with column mapping (one row per attribute)

writeExternal

public void writeExternal(ObjectOutput os)
                   throws IOException

Redirects the default java serialization via externalizable to our default hadoop writable serialization for efficient broadcast/rdd serialization.

Specified by:

writeExternal in interface Externalizable

Parameters:

os - object output

Throws:

IOException - if IOException occurs

readExternal

public void readExternal(ObjectInput in)
                  throws IOException

Redirects the default java serialization via externalizable to our default hadoop writable serialization for efficient broadcast/rdd deserialization.

Specified by:

readExternal in interface Externalizable

Parameters:

in - object input

Throws:

IOException - if IOException occur

getColID

public int getColID()

setColID

public void setColID(int colID)

shiftCol

public void shiftCol(int columnOffset)

setEstMetaSize

public void setEstMetaSize(long estSize)

getEstMetaSize

public long getEstMetaSize()

setEstNumDistincts

public void setEstNumDistincts(int numDistincts)

getEstNumDistincts

public int getEstNumDistincts()

compareTo

public int compareTo(ColumnEncoder o)

Specified by:

compareTo in interface Comparable<ColumnEncoder>

getBuildTasks

public List<DependencyTask<?>> getBuildTasks(CacheBlock in)

getBuildTask

public Callable<Object> getBuildTask(CacheBlock in)

getPartialBuildTask

public Callable<Object> getPartialBuildTask(CacheBlock in,
                                            int startRow,
                                            int blockSize,
                                            HashMap<Integer,Object> ret)

getPartialMergeBuildTask

public Callable<Object> getPartialMergeBuildTask(HashMap<Integer,?> ret)

getApplyTasks

public List<DependencyTask<?>> getApplyTasks(CacheBlock in,
                                             MatrixBlock out,
                                             int outputCol)

getSparseRowsWZeros

public Set<Integer> getSparseRowsWZeros()