Package ai.djl.training.dataset

Class ArrayDataset

java.lang.Object

ai.djl.training.dataset.RandomAccessDataset

ai.djl.training.dataset.ArrayDataset

All Implemented Interfaces:

Dataset

public class ArrayDataset
extends RandomAccessDataset

ArrayDataset is an implementation of RandomAccessDataset that consist entirely of large NDArrays. It is recommended only for datasets small enough to fit in memory that come in array formats. Otherwise, consider directly using the RandomAccessDataset instead.

There can be multiple data and label NDArrays within the dataset. Each sample will be retrieved by indexing each NDArray along the first dimension.

The following is an example of how to use ArrayDataset:

     ArrayDataset dataset = new ArrayDataset.Builder()
                              .setData(data1, data2)
                              .optLabels(labels1, labels2, labels3)
                              .setSampling(20, false)
                              .build();
 

Suppose you get a Batch from trainer.iterateDataset(dataset) or dataset.getData(manager). In the data of this batch, it will be an NDList with one NDArray for each data input. In this case, it would be 2 arrays. Similarly, the labels would have 3 arrays.

See Also:

• Dataset

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static final class	ArrayDataset.Builder	The Builder to construct an ArrayDataset.

Nested classes/interfaces inherited from class ai.djl.training.dataset.RandomAccessDataset

RandomAccessDataset.BaseBuilder<T extends RandomAccessDataset.BaseBuilder<T>>

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

Dataset.Usage

Field Summary

Fields

Modifier and Type	Field	Description
protected NDArray[]	data
protected NDArray[]	labels

Fields inherited from class ai.djl.training.dataset.RandomAccessDataset

dataBatchifier, device, labelBatchifier, limit, pipeline, prefetchNumber, sampler, targetPipeline

Constructor Summary

Constructors

Constructor	Description
ArrayDataset(RandomAccessDataset.BaseBuilder<?> builder)	Creates a new instance of ArrayDataset with the arguments in ArrayDataset.Builder.

Method Summary

Modifier and Type	Method	Description
protected long	availableSize()	Returns the number of records available to be read in this Dataset.
Record	get(NDManager manager, long index)	Gets the Record for the given index from the dataset.
Batch	getByIndices(NDManager manager, long... indices)	Gets the Batch for the given indices from the dataset.
Batch	getByRange(NDManager manager, long fromIndex, long toIndex)	Gets the Batch for the given range from the dataset.
Iterable<Batch>	getData(NDManager manager, Sampler sampler, ExecutorService executorService)	Fetches an iterator that can iterate through the Dataset with a custom sampler multi-threaded.
protected RandomAccessDataset	newSubDataset(int[] indices, int from, int to)
protected RandomAccessDataset	newSubDataset(List<Long> subIndices)
void	prepare(ai.djl.util.Progress progress)	Prepares the dataset for use with tracked progress.

Methods inherited from class ai.djl.training.dataset.RandomAccessDataset

getData, getData, getData, randomSplit, size, subDataset, subDataset, subDataset, subDataset, toArray

Methods inherited from class java.lang.Object

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

Methods inherited from interface ai.djl.training.dataset.Dataset

matchingTranslatorOptions, prepare

Field Details

data

protected NDArray[] data

labels

protected NDArray[] labels

Constructor Details

ArrayDataset

public ArrayDataset(RandomAccessDataset.BaseBuilder<?> builder)

Creates a new instance of ArrayDataset with the arguments in ArrayDataset.Builder.

Parameters:

builder - a builder with the required arguments

Method Details

availableSize

protected long availableSize()

Returns the number of records available to be read in this Dataset.

Specified by:

availableSize in class RandomAccessDataset

Returns:

the number of records available to be read in this Dataset

get

public Record get(NDManager manager,
 long index)

Gets the Record for the given index from the dataset.

Specified by:

get in class RandomAccessDataset

Parameters:

manager - the manager used to create the arrays

index - the index of the requested data item

Returns:

a Record that contains the data and label of the requested data item

getByIndices

public Batch getByIndices(NDManager manager,
 long... indices)

Gets the Batch for the given indices from the dataset.

Parameters:

manager - the manager used to create the arrays

indices - indices of the requested data items

Returns:

a Batch that contains the data and label of the requested data items

getByRange

public Batch getByRange(NDManager manager,
 long fromIndex,
 long toIndex)

Gets the Batch for the given range from the dataset.

Parameters:

manager - the manager used to create the arrays

fromIndex - low endpoint (inclusive) of the dataset

toIndex - high endpoint (exclusive) of the dataset

Returns:

a Batch that contains the data and label of the requested data items

newSubDataset

protected RandomAccessDataset newSubDataset(int[] indices,
 int from,
 int to)

Overrides:

newSubDataset in class RandomAccessDataset

newSubDataset

protected RandomAccessDataset newSubDataset(List<Long> subIndices)

Overrides:

newSubDataset in class RandomAccessDataset

getData

public Iterable<Batch> getData(NDManager manager,
 Sampler sampler,
 ExecutorService executorService)
                        throws IOException,
TranslateException

Fetches an iterator that can iterate through the Dataset with a custom sampler multi-threaded.

Overrides:

getData in class RandomAccessDataset

Parameters:

manager - the manager to create the arrays

sampler - the sampler to use to iterate through the dataset

executorService - the executorService to multi-thread with

Returns:

an Iterable of Batch that contains batches of data from the dataset

Throws:

IOException - for various exceptions depending on the dataset

TranslateException - if there is an error while processing input

prepare

public void prepare(ai.djl.util.Progress progress)
             throws IOException

Prepares the dataset for use with tracked progress.

Parameters:

progress - the progress tracker

Throws:

IOException - for various exceptions depending on the dataset