Class AbstractConstraint

java.lang.Object

org.tensorflow.framework.constraints.AbstractConstraint

All Implemented Interfaces:

Constraint

Direct Known Subclasses:

MaxNorm, MinMaxNorm, NonNeg, UnitNorm

public abstract class AbstractConstraint
extends Object
implements Constraint

Base class for Constraints. AbstractConstraint subclasses impose constraints on weight values

Field Summary

Fields

Modifier and Type	Field	Description
static final float	EPSILON

Constructor Summary

Constructors

Constructor	Description
AbstractConstraint()	Creates a AbstractConstraint

Method Summary

Modifier and Type	Method	Description
protected <T extends TNumber> Operand<T>	clip(Ops tf, Operand<T> x, double minValue, double maxValue)	Gets the element-wise value clipping.
protected <T extends TNumber> Operand<T>	norm(Ops tf, Operand<T> weights, int[] axes)	Calculates the norm of the weights along the axes
protected <T extends TNumber> Operand<T>	sqrt(Ops tf, Operand<T> x)	Gets the element-wise square root.

Methods inherited from class Object

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

Methods inherited from interface Constraint

call

Field Details

EPSILON

public static final float EPSILON

See Also:

• Constant Field Values

Constructor Details

AbstractConstraint

public AbstractConstraint()

Creates a AbstractConstraint

Method Details

sqrt

protected <T extends TNumber> Operand<T> sqrt(Ops tf,
 Operand<T> x)

Gets the element-wise square root.

Type Parameters:

T - The data type for the operand and result.

Parameters:

tf - the TensorFlow Ops

x - the input Operand.

Returns:

the element-wise square root.

Throws:

IllegalArgumentException - if x is null

clip

protected <T extends TNumber> Operand<T> clip(Ops tf,
 Operand<T> x,
 double minValue,
 double maxValue)

Gets the element-wise value clipping.

Type Parameters:

T - The data type for the operand and result.

Parameters:

tf - the TensorFlow Ops

x - the Operand to clip

minValue - the minimum value

maxValue - the maximum value

Returns:

the operand with clipped values

Throws:

IllegalArgumentException - if x is null

norm

protected <T extends TNumber> Operand<T> norm(Ops tf,
 Operand<T> weights,
 int[] axes)

Calculates the norm of the weights along the axes

Type Parameters:

T - the data type for the weights and the result

Parameters:

tf - the TensorFlow Ops

weights - the weights used to calculate the norms

axes - the axes along which to calculate weight norms.

Returns:

the norms

Throws:

IllegalArgumentException - if weights is null