VarianceScaling

public class VarianceScaling

Known Direct Subclasses Glorot<T extends TFloating>, He<T extends TFloating>, LeCun<T extends TFloating> Glorot<T extends TFloating> The Glorot initializer, also called Xavier initializer.  He<T extends TFloating> He initializer.  LeCun<T extends TFloating> LeCun normal initializer.

Initializer capable of adapting its scale to the shape of weights tensors.

With distribution=TRUNCATED_NORMAL or NORMAL, samples are drawn from a truncated/untruncated normal distribution with a mean of zero and a standard deviation (after truncation, if used) stddev = Math.sqrt(scale / n), where n is:

• number of input units in the weight tensor, if mode=FAN_IN
• number of output units, if mode=FAN_OUT
• average of the numbers of input and output units, if mode=FAN_AVG

With distribution=UNIFORM, samples are drawn from a uniform distribution within [-limit, limit], where limit = Math.sqrt(3 * scale / n);.

Examples:

      long seed = 1234l;
      float scale = 0.1f;
      VarianceScaling<TFloat32, TFloat32> initializer =
          new org.tensorflow.framework.initializers.VarianceScaling<>(
              tf, scale, Mode.FAN_IN, Distribution.UNIFORM, seed);
      Operand<TFloat32> values =
          initializer.call(tf.constant(Shape.of(2,2)), TFloat32.class);
 

See Also

• VarianceScaling.Mode
• VarianceScaling.Distribution

Nested Classes

enum	VarianceScaling.Distribution		The random distribution to use when initializing the values.
enum	VarianceScaling.Mode		The mode to use for calculating the fan values.

Constants

double

SCALE_DEFAULT

Fields

public static final VarianceScaling.Distribution	DISTRIBUTION_DEFAULT
public static final VarianceScaling.Mode	MODE_DEFAULT

Public Constructors

	VarianceScaling(Ops tf, long seed) Creates a VarianceScaling Initializer
	VarianceScaling(Ops tf, double scale, VarianceScaling.Mode mode, VarianceScaling.Distribution distribution, long seed) Creates a VarianceScaling Initializer

Public Methods

Operand<T>

call(Operand<TInt64> dims, Class<T> type)

Inherited Methods

From class org.tensorflow.framework.initializers.BaseInitializer

Ops	getTF() Gets the TensorFlow Ops

From class java.lang.Object

boolean	equals(Object arg0)
final Class<?>	getClass()
int	hashCode()
final void	notify()
final void	notifyAll()
String	toString()
final void	wait(long arg0, int arg1)
final void	wait(long arg0)
final void	wait()

From interface org.tensorflow.framework.initializers.Initializer

abstract Operand<T>

call(Operand<TInt64> dims, Class<T> type) Generates the operation used to perform the initialization.

Constants

public static final double SCALE_DEFAULT

Constant Value: 1.0

Fields

public static final VarianceScaling.Distribution DISTRIBUTION_DEFAULT

public static final VarianceScaling.Mode MODE_DEFAULT

Public Constructors

public VarianceScaling (Ops tf, long seed)

Creates a VarianceScaling Initializer

Parameters

tf	the TensorFlow Ops
seed	sed to create random seeds.

public VarianceScaling (Ops tf, double scale, VarianceScaling.Mode mode, VarianceScaling.Distribution distribution, long seed)

Creates a VarianceScaling Initializer

Parameters

tf	the TensorFlow Ops
scale	Scaling factor (positive float).
mode	the mode for the variance
distribution	Random distribution to use.
seed	Used to create random seeds.

Public Methods

public Operand<T> call (Operand<TInt64> dims, Class<T> type)