Package ai.djl.translate

Class PaddingStackBatchifier.Builder

java.lang.Object

ai.djl.translate.PaddingStackBatchifier.Builder

Enclosing class:

PaddingStackBatchifier

public static final class PaddingStackBatchifier.Builder
extends Object

Builder to build a PaddingStackBatchifier.

Method Summary

Modifier and Type	Method	Description
PaddingStackBatchifier.Builder	addPad(int array, int dim, NDArraySupplier supplier)	Adds a new dimension to be padded in the input NDList.
PaddingStackBatchifier.Builder	addPad(int array, int dim, NDArraySupplier supplier, int paddingSize)	Adds a new dimension to be padded in the input NDList.
PaddingStackBatchifier	build()	Builds the PaddingStackBatchifier.
PaddingStackBatchifier.Builder	optIncludeValidLengths(boolean includeValidLengths)	Sets whether to include the valid lengths (length of non-padded data) for each array.

Methods inherited from class java.lang.Object

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

Method Details

optIncludeValidLengths

public PaddingStackBatchifier.Builder optIncludeValidLengths(boolean includeValidLengths)

Sets whether to include the valid lengths (length of non-padded data) for each array.

Parameters:

includeValidLengths - true to include valid lengths

Returns:

this builder

addPad

public PaddingStackBatchifier.Builder addPad(int array,
 int dim,
 NDArraySupplier supplier)

Adds a new dimension to be padded in the input NDList.

Parameters:

array - which array in the NDList to pad

dim - which dimension in the array to pad

supplier - a supplier that produces the padding array. The padding array shape should include both the batch and a 1 for the padded dimension. For batch array shape NTC, the padding shape should be N x 1 x C

Returns:

this builder

addPad

public PaddingStackBatchifier.Builder addPad(int array,
 int dim,
 NDArraySupplier supplier,
 int paddingSize)

Adds a new dimension to be padded in the input NDList.

Parameters:

array - which array in the NDList to pad

dim - which dimension in the array to pad

supplier - a supplier that produces the padding array. The padding array shape should include both the batch and a 1 for the padded dimension. For batch array shape NTC, the padding shape should be N x 1 x C

paddingSize - the minimum padding size to use. All sequences to pad must be less than this size

Returns:

this builder

build

public PaddingStackBatchifier build()

Builds the PaddingStackBatchifier.

Returns:

the constructed PaddingStackBatchifier