Class L1L2

java.lang.Object

org.tensorflow.framework.regularizers.AbstractRegularizer

org.tensorflow.framework.regularizers.L1L2

All Implemented Interfaces:

Regularizer

Direct Known Subclasses:

L1, L2

public class L1L2
extends AbstractRegularizer

A regularizer that applies both L1 and L2 regularization penalties.

The L1 regularization penalty is computed as:

loss = l1 * reduceSum(abs(x))

The L2 regularization penalty is computed as

loss = l2 * reduceSum(square(x))

Field Summary

Fields inherited from class AbstractRegularizer

DEFAULT_REGULARIZATION_PENALTY

Constructor Summary

Constructors

Constructor	Description
L1L2()	Creates an L1L2 regularizer with no l1 or l2 penalty with zero penalty
L1L2(float l1, float l2)	Creates an L1L2 regularizer
L1L2(String name, float l1, float l2)	Creates an L1L2 regularizer

Method Summary

Modifier and Type	Method	Description
<R extends TNumber> Operand<R>	call(Ops tf, Operand<R> input)	Computes a regularization penalty from an input.
float	getL1()	Gets the L1 regularization factor
float	getL2()	Gets the L2 regularization factor

Methods inherited from class AbstractRegularizer

asLoss, getName

Methods inherited from class Object

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

Constructor Details

L1L2

public L1L2()

Creates an L1L2 regularizer with no l1 or l2 penalty with zero penalty

L1L2

public L1L2(float l1,
 float l2)

Creates an L1L2 regularizer

Parameters:

l1 - L1 regularization factor, if null it is set to 0.

l2 - L2 regularization factor, if null it is set to 0.

Throws:

IllegalArgumentException - if the l1 or l2 regularization factor is Float.isNaN(float) of Float.isInfinite(float)

L1L2

public L1L2(String name,
 float l1,
 float l2)

Creates an L1L2 regularizer

Parameters:

name - the name for this regularizer, if null the class name will be used.

l1 - L1 regularization factor, if null it is set to 0.

l2 - L2 regularization factor, if null it is set to 0.

Throws:

IllegalArgumentException - if the l1 or l2 regularization factor is Float.isNaN(float) of Float.isInfinite(float)

Method Details

call

public <R extends TNumber> Operand<R> call(Ops tf,
 Operand<R> input)

Computes a regularization penalty from an input.

Type Parameters:

R - the data type of the input and result

Parameters:

tf - the TensorFlow Ops

input - the weighted input

Returns:

the result of computing the regularization penalty

getL1

public float getL1()

Gets the L1 regularization factor

Returns:

the L1 regularization factor

getL2

public float getL2()

Gets the L2 regularization factor

Returns:

the L2 regularization factor