org.deeplearning4j.util

Class OutputLayerUtil

java.lang.Object

org.deeplearning4j.util.OutputLayerUtil

public class OutputLayerUtil
extends Object

Method Summary

Modifier and Type	Method and Description
static boolean	activationExceedsZeroOneRange(IActivation activation, boolean isLossLayer)
static boolean	lossFunctionExpectsProbability(ILossFunction lf)
static void	validateOutputLayer(String layerName, Layer layer) Validate the output layer (or loss layer) configuration, to detect invalid consfiugrations.
static void	validateOutputLayerConfiguration(String layerName, long nOut, boolean isLossLayer, IActivation activation, ILossFunction lossFunction) Validate the output layer (or loss layer) configuration, to detect invalid consfiugrations.
static void	validateOutputLayerForClassifierEvaluation(Layer outputLayer, Class<? extends IEvaluation> classifierEval) Validates if the output layer configuration is valid for classifier evaluation.

Methods inherited from class java.lang.Object

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

Method Detail

validateOutputLayer

public static void validateOutputLayer(String layerName,
                                       Layer layer)

Validate the output layer (or loss layer) configuration, to detect invalid consfiugrations. A DL4JInvalidConfigException will be thrown for invalid configurations (like softmax + nOut=1).
If the specified layer is not an output layer, this is a no-op

Parameters:

layerName - Name of the layer

layer - Layer

validateOutputLayerConfiguration

public static void validateOutputLayerConfiguration(String layerName,
                                                    long nOut,
                                                    boolean isLossLayer,
                                                    IActivation activation,
                                                    ILossFunction lossFunction)

Validate the output layer (or loss layer) configuration, to detect invalid consfiugrations. A DL4JInvalidConfigException will be thrown for invalid configurations (like softmax + nOut=1).

If the specified layer is not an output layer, this is a no-op

Parameters:

layerName - Name of the layer

nOut - Number of outputs for the layer

isLossLayer - Should be true for loss layers (no params), false for output layers

activation - Activation function

lossFunction - Loss function

lossFunctionExpectsProbability

public static boolean lossFunctionExpectsProbability(ILossFunction lf)

activationExceedsZeroOneRange

public static boolean activationExceedsZeroOneRange(IActivation activation,
                                                    boolean isLossLayer)

validateOutputLayerForClassifierEvaluation

public static void validateOutputLayerForClassifierEvaluation(Layer outputLayer,
                                                              Class<? extends IEvaluation> classifierEval)

Validates if the output layer configuration is valid for classifier evaluation. This is used to try and catch invalid evaluation - i.e., trying to use classifier evaluation on a regression model. This method won't catch all possible invalid cases, but should catch some common problems.

Parameters:

outputLayer - Output layer

classifierEval - Class for the classifier evaluation