org.deeplearning4j.nn.conf

Class NeuralNetConfiguration

java.lang.Object

org.deeplearning4j.nn.conf.NeuralNetConfiguration

All Implemented Interfaces:

java.io.Serializable, java.lang.Cloneable

public class NeuralNetConfiguration
extends java.lang.Object
implements java.io.Serializable, java.lang.Cloneable

A Serializable configuration for neural nets that covers per layer parameters

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	NeuralNetConfiguration.Builder
static class	NeuralNetConfiguration.ListBuilder Fluent interface for building a list of configurations

Field Summary

Fields

Modifier and Type	Field and Description
static java.lang.String	CUSTOM_FUNCTIONALITY System property for custom layers, preprocessors, graph vertices etc.
protected int	iterationCount
protected java.util.Map<java.lang.String,java.lang.Double>	l1ByParam
protected java.util.Map<java.lang.String,java.lang.Double>	l2ByParam
protected Layer	layer
protected double	leakyreluAlpha Deprecated.
protected java.util.Map<java.lang.String,java.lang.Double>	learningRateByParam
protected LearningRatePolicy	learningRatePolicy
protected double	lrPolicyDecayRate
protected double	lrPolicyPower
protected double	lrPolicySteps
protected int	maxNumLineSearchIterations
protected boolean	miniBatch
protected boolean	minimize
protected int	numIterations
protected OptimizationAlgorithm	optimizationAlgo
protected boolean	pretrain
protected long	seed
protected StepFunction	stepFunction
protected boolean	useDropConnect
protected boolean	useRegularization
protected java.util.List<java.lang.String>	variables

Constructor Summary

Constructors

Constructor and Description
NeuralNetConfiguration()

Method Summary

Modifier and Type	Method and Description
void	addVariable(java.lang.String variable)
void	clearVariables()
NeuralNetConfiguration	clone() Creates and returns a deep copy of the configuration.
static NeuralNetConfiguration	fromJson(java.lang.String json) Create a neural net configuration from json
static NeuralNetConfiguration	fromYaml(java.lang.String json) Create a neural net configuration from json
double	getL1ByParam(java.lang.String variable)
double	getL2ByParam(java.lang.String variable)
double	getLearningRateByParam(java.lang.String variable)
static org.nd4j.shade.jackson.databind.ObjectMapper	mapper() Object mapper for serialization of configurations
static org.nd4j.shade.jackson.databind.ObjectMapper	mapperYaml() Object mapper for serialization of configurations
static org.nd4j.shade.jackson.databind.ObjectMapper	reinitMapperWithSubtypes(java.util.Collection<org.nd4j.shade.jackson.databind.jsontype.NamedType> additionalTypes) Reinitialize and return the Jackson/json ObjectMapper with additional named types.
void	setLayerParamLR(java.lang.String variable)
void	setLearningRateByParam(java.lang.String variable, double rate)
java.lang.String	toJson() Return this configuration as json
java.lang.String	toYaml() Return this configuration as json
java.util.List<java.lang.String>	variables()
java.util.List<java.lang.String>	variables(boolean copy)

Methods inherited from class java.lang.Object

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

Field Detail

CUSTOM_FUNCTIONALITY

public static final java.lang.String CUSTOM_FUNCTIONALITY

System property for custom layers, preprocessors, graph vertices etc. Enabled by default. Run JVM with "-Dorg.deeplearning4j.config.custom.enabled=false" to disable classpath scanning for Overriding the default (i.e., disabling) this is only useful if (a) no custom layers/preprocessors etc will be used, and (b) minimizing startup/initialization time for new JVMs is very important. Results are cached, so there is no cost to custom layers after the first network has been constructed.

See Also:

Constant Field Values

layer

protected Layer layer

leakyreluAlpha

@Deprecated
protected double leakyreluAlpha

Deprecated.

miniBatch

protected boolean miniBatch

numIterations

protected int numIterations

maxNumLineSearchIterations

protected int maxNumLineSearchIterations

seed

protected long seed

optimizationAlgo

protected OptimizationAlgorithm optimizationAlgo

variables

protected java.util.List<java.lang.String> variables

stepFunction

protected StepFunction stepFunction

useRegularization

protected boolean useRegularization

useDropConnect

protected boolean useDropConnect

minimize

protected boolean minimize

learningRateByParam

protected java.util.Map<java.lang.String,java.lang.Double> learningRateByParam

l1ByParam

protected java.util.Map<java.lang.String,java.lang.Double> l1ByParam

l2ByParam

protected java.util.Map<java.lang.String,java.lang.Double> l2ByParam

learningRatePolicy

protected LearningRatePolicy learningRatePolicy

lrPolicyDecayRate

protected double lrPolicyDecayRate

lrPolicySteps

protected double lrPolicySteps

lrPolicyPower

protected double lrPolicyPower

pretrain

protected boolean pretrain

iterationCount

protected int iterationCount

Constructor Detail

NeuralNetConfiguration

public NeuralNetConfiguration()

Method Detail

clone

public NeuralNetConfiguration clone()

Creates and returns a deep copy of the configuration.

Overrides:

clone in class java.lang.Object

variables

public java.util.List<java.lang.String> variables()

variables

public java.util.List<java.lang.String> variables(boolean copy)

addVariable

public void addVariable(java.lang.String variable)

clearVariables

public void clearVariables()

setLayerParamLR

public void setLayerParamLR(java.lang.String variable)

getLearningRateByParam

public double getLearningRateByParam(java.lang.String variable)

setLearningRateByParam

public void setLearningRateByParam(java.lang.String variable,
                                   double rate)

getL1ByParam

public double getL1ByParam(java.lang.String variable)

getL2ByParam

public double getL2ByParam(java.lang.String variable)

toYaml

public java.lang.String toYaml()

Return this configuration as json

Returns:

this configuration represented as json

fromYaml

public static NeuralNetConfiguration fromYaml(java.lang.String json)

Create a neural net configuration from json

Parameters:

json - the neural net configuration from json

Returns:

toJson

public java.lang.String toJson()

Return this configuration as json

Returns:

this configuration represented as json

fromJson

public static NeuralNetConfiguration fromJson(java.lang.String json)

Create a neural net configuration from json

Parameters:

json - the neural net configuration from json

Returns:

mapperYaml

public static org.nd4j.shade.jackson.databind.ObjectMapper mapperYaml()

Object mapper for serialization of configurations

Returns:

mapper

public static org.nd4j.shade.jackson.databind.ObjectMapper mapper()

Object mapper for serialization of configurations

Returns:

reinitMapperWithSubtypes

public static org.nd4j.shade.jackson.databind.ObjectMapper reinitMapperWithSubtypes(java.util.Collection<org.nd4j.shade.jackson.databind.jsontype.NamedType> additionalTypes)

Reinitialize and return the Jackson/json ObjectMapper with additional named types. This can be used to add additional subtypes at runtime (i.e., for JSON mapping with types defined outside of the main DL4J codebase)