Class FastElementSequence<T, U extends NdArray<T>>

java.lang.Object

org.tensorflow.ndarray.impl.sequence.FastElementSequence<T,U>

Type Parameters:

T - Type of the elements

U - Type of the NdArray with this sequence

All Implemented Interfaces:

Iterable<U>, NdArraySequence<U>

public final class FastElementSequence<T, U extends NdArray<T>>
extends Object
implements NdArraySequence<U>

A sequence recycling the same NdArray instance when iterating its elements

Constructor Summary

Constructors

Constructor	Description
FastElementSequence(org.tensorflow.ndarray.impl.AbstractNdArray<T,U> ndArray, int dimensionIdx, U element, DataBufferWindow<?> elementWindow)

Method Summary

Modifier and Type	Method	Description
NdArraySequence<U>	asSlices()	Returns each element as a new slice.
void	forEachIndexed(BiConsumer<long[],U> consumer)	Visit each elements of this iteration and their respective coordinates.
Iterator<U>	iterator()

Methods inherited from class Object

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

Methods inherited from interface Iterable

forEach, spliterator

Constructor Details

FastElementSequence

public FastElementSequence(org.tensorflow.ndarray.impl.AbstractNdArray<T,U> ndArray,
 int dimensionIdx,
 U element,
 DataBufferWindow<?> elementWindow)

Method Details

iterator

public Iterator<U> iterator()

Specified by:

iterator in interface Iterable<T>

forEachIndexed

public void forEachIndexed(BiConsumer<long[],U> consumer)

Description copied from interface: NdArraySequence

Visit each elements of this iteration and their respective coordinates.

Important: the consumer method should not keep a reference to the coordinates as they might be mutable and reused during the iteration to improve performance.

Specified by:

forEachIndexed in interface NdArraySequence<T>

Parameters:

consumer - method to invoke for each elements

asSlices

public NdArraySequence<U> asSlices()

Description copied from interface: NdArraySequence

Returns each element as a new slice.

Unlike conventional Java collections, elements of a NdArraySequence are transient, i.e. new NdArray instances are allocated for each iteration. To improve performance, the same instance can be recycled to view all elements of this sequence, using a DataBufferWindow.

In some cases though, it might be preferable to disable such optimizations to ensure that each element returned is a new slice of the original array. For example, if one or more elements visited must live beyond the scope of the sequence iteration, asSlices() makes sure that all elements returned by the sequence are unique instances.

    final List<IntNdArray> vectors = new ArrayList<>();
    IntNdArray matrix = NdArrays.ofInts(Shape.of(6, 6));
    ndArray.elements(0).forEach(e -> vectors::add);  // Not safe, as `e` might always be the same recycled instance
    ndArray.elements(0).asSlices().forEach(e -> vectors::add);  // Safe, each `e` is a distinct NdArray instance

Specified by:

asSlices in interface NdArraySequence<T>

Returns:

a sequence that returns each elements iterated as a new slice

See Also:

• DataBufferWindow