Package org.apache.sysds.runtime.data

Class SparseRow

java.lang.Object

org.apache.sysds.runtime.data.SparseRow

All Implemented Interfaces:

Serializable

Direct Known Subclasses:

SparseRowScalar, SparseRowVector

public abstract class SparseRow
extends Object
implements Serializable

Base class for sparse row implementations such as sparse row vectors and sparse scalars (single value per row).

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor	Description
SparseRow()

Method Summary

Modifier and Type	Method	Description
abstract boolean	add(int col, double v)	Add a value to a specified column with awareness of potential insertions.
abstract SparseRow	append(int col, double v)	Appends a value to the end of the sparse row.
abstract void	compact()	In-place compaction of non-zero-entries; removes zero entries and shifts non-zero entries to the left if necessary.
abstract void	compact(double eps)	In-place compaction of values over eps away from zero; and shifts non-zero entries to the left if necessary.
abstract SparseRow	copy(boolean deep)	Make a copy of this row.
abstract double	get(int col)	Gets the value of a specified column.
abstract int[]	indexes()	Get the index array of non-zero entries, co-aligned with the array of values.
abstract boolean	isEmpty()	Indicates if the sparse row is empty, i.e., if is has size zero.
abstract void	reset(int estnns, int maxnns)	Resets the sparse row to empty, after this call size and isEmpty are guaranteed to return 0 and true, respectively.
abstract int	searchIndexesFirstGT(int col)
abstract int	searchIndexesFirstGTE(int col)	Get first index greater than equal column index.
abstract boolean	set(int col, double v)	Sets the value of a specified column with awareness of potential overwrites or deletes (set to value zero).
abstract int	size()	Get the number of non-zero values of the sparse row.
abstract void	sort()	In-place sort of column-index value pairs in order to allow binary search after constant-time append was used for reading unordered sparse rows.
String	toString()
abstract double[]	values()	Get the value array of non-zero entries, co-aligned with the array of indexes.

Methods inherited from class java.lang.Object

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

Constructor Detail

SparseRow

public SparseRow()

Method Detail

size

public abstract int size()

Get the number of non-zero values of the sparse row.

Returns:

number of non-zeros

isEmpty

public abstract boolean isEmpty()

Indicates if the sparse row is empty, i.e., if is has size zero.

Returns:

true if empty

values

public abstract double[] values()

Get the value array of non-zero entries, co-aligned with the array of indexes.

Returns:

array of values

indexes

public abstract int[] indexes()

Get the index array of non-zero entries, co-aligned with the array of values.

Returns:

array of indexes

reset

public abstract void reset(int estnns,
                           int maxnns)

Resets the sparse row to empty, after this call size and isEmpty are guaranteed to return 0 and true, respectively.

Parameters:

estnns - estimated number of non-zeros

maxnns - maximum number of non-zeros, e.g., number of columns

set

public abstract boolean set(int col,
                            double v)

Sets the value of a specified column with awareness of potential overwrites or deletes (set to value zero).

Parameters:

col - column index, zero-based

v - value

Returns:

true if the size of the sparse row changed

add

public abstract boolean add(int col,
                            double v)

Add a value to a specified column with awareness of potential insertions.

Parameters:

col - column index, zero-based

v - value

Returns:

true if the size of the sparse row changed

append

public abstract SparseRow append(int col,
                                 double v)

Appends a value to the end of the sparse row.

Parameters:

col - column index, zero-based

v - value

Returns:

the row with an appended element

get

public abstract double get(int col)

Gets the value of a specified column. If the column index does not exist in the sparse row, this call returns zero.

Parameters:

col - column index, zero-based

Returns:

value

sort

public abstract void sort()

In-place sort of column-index value pairs in order to allow binary search after constant-time append was used for reading unordered sparse rows. We first check if already sorted and subsequently sort if necessary in order to get O(n) best case. Note: In-place sort is necessary in order to guarantee the memory estimate for operations that implicitly read that data set.

compact

public abstract void compact()

In-place compaction of non-zero-entries; removes zero entries and shifts non-zero entries to the left if necessary.

compact

public abstract void compact(double eps)

In-place compaction of values over eps away from zero; and shifts non-zero entries to the left if necessary.

Parameters:

eps - epsilon value

copy

public abstract SparseRow copy(boolean deep)

Make a copy of this row.

Parameters:

deep - if the copy should be deep

Returns:

A copy

searchIndexesFirstGTE

public abstract int searchIndexesFirstGTE(int col)

Get first index greater than equal column index.

Parameters:

col - column to be greater than

Returns:

index

searchIndexesFirstGT

public abstract int searchIndexesFirstGT(int col)

toString

public String toString()

Overrides:

toString in class Object