Package ai.djl.training.dataset

Class BatchSampler

java.lang.Object

ai.djl.training.dataset.BatchSampler

All Implemented Interfaces:

Sampler

public class BatchSampler
extends java.lang.Object
implements Sampler

BatchSampler is a Sampler that returns a single epoch over the data.

BatchSampler wraps another Sampler.SubSampler to yield a mini-batch of indices.

Nested Class Summary

Nested classes/interfaces inherited from interface ai.djl.training.dataset.Sampler

Sampler.SubSampler

Constructor Summary

Constructors

Constructor	Description
BatchSampler(Sampler.SubSampler subSampler, int batchSize)	Creates a new instance of BatchSampler that samples from the given Sampler.SubSampler, and yields a mini-batch of indices.
BatchSampler(Sampler.SubSampler subSampler, int batchSize, boolean dropLast)	Creates a new instance of BatchSampler that samples from the given Sampler.SubSampler, and yields a mini-batch of indices.

Method Summary

Modifier and Type	Method	Description
int	getBatchSize()	Returns the batch size of the Sampler.
java.util.Iterator<java.util.List<java.lang.Long>>	sample(RandomAccessDataset dataset)	Fetches an iterator that iterates through the given RandomAccessDataset in mini-batches of indices.

Methods inherited from class java.lang.Object

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

Constructor Detail

BatchSampler

public BatchSampler(Sampler.SubSampler subSampler,
                    int batchSize)

Creates a new instance of BatchSampler that samples from the given Sampler.SubSampler, and yields a mini-batch of indices.

The last batch will not be dropped. The size of the last batch maybe smaller than batch size in case the size of the dataset is not a multiple of batch size.

Parameters:

subSampler - the Sampler.SubSampler to sample from

batchSize - the required batch size

BatchSampler

public BatchSampler(Sampler.SubSampler subSampler,
                    int batchSize,
                    boolean dropLast)

Creates a new instance of BatchSampler that samples from the given Sampler.SubSampler, and yields a mini-batch of indices.

Parameters:

subSampler - the Sampler.SubSampler to sample from

batchSize - the required batch size

dropLast - whether the BatchSampler should drop the last few samples in case the size of the dataset is not a multiple of batch size

Method Detail

sample

public java.util.Iterator<java.util.List<java.lang.Long>> sample(RandomAccessDataset dataset)

Fetches an iterator that iterates through the given RandomAccessDataset in mini-batches of indices.

Specified by:

sample in interface Sampler

Parameters:

dataset - the RandomAccessDataset to sample from

Returns:

an iterator that iterates through the given RandomAccessDataset in mini-batches of indices

getBatchSize

public int getBatchSize()

Returns the batch size of the Sampler.

Specified by:

getBatchSize in interface Sampler

Returns:

the batch size of the Sampler, -1 if batch size is not fixed