org.nd4j.autodiff.samediff

Class TrainingConfig.Builder

java.lang.Object

org.nd4j.autodiff.samediff.TrainingConfig.Builder

Enclosing class:

TrainingConfig

public static class TrainingConfig.Builder
extends Object

Constructor Summary

Constructors

Constructor and Description
Builder()

Method Summary

Modifier and Type	Method and Description
TrainingConfig.Builder	addEvaluations(boolean validation, @NonNull String variableName, int labelIndex, IEvaluation... evaluations) Add requested evaluations for a parm/variable, for either training or validation.
TrainingConfig.Builder	addRegularization(Regularization... regularizations) Add regularization to all trainable parameters in the network
TrainingConfig	build()
TrainingConfig.Builder	dataSetFeatureMapping(List<String> dataSetFeatureMapping) Set the name of the placeholders/variables that should be set using the feature INDArray(s) from the DataSet or MultiDataSet.
TrainingConfig.Builder	dataSetFeatureMapping(String... dataSetFeatureMapping) Set the name of the placeholders/variables that should be set using the feature INDArray(s) from the DataSet or MultiDataSet.
TrainingConfig.Builder	dataSetFeatureMaskMapping(List<String> dataSetFeatureMaskMapping) Set the name of the placeholders/variables that should be set using the feature mask INDArray(s) from the DataSet or MultiDataSet.
TrainingConfig.Builder	dataSetFeatureMaskMapping(String... dataSetFeatureMaskMapping) See dataSetFeatureMaskMapping(List)
TrainingConfig.Builder	dataSetLabelMapping(List<String> dataSetLabelMapping) Set the name of the placeholders/variables that should be set using the labels INDArray(s) from the DataSet or MultiDataSet.
TrainingConfig.Builder	dataSetLabelMapping(String... dataSetLabelMapping) Set the name of the placeholders/variables that should be set using the labels INDArray(s) from the DataSet or MultiDataSet.
TrainingConfig.Builder	dataSetLabelMaskMapping(List<String> dataSetLabelMaskMapping) Set the name of the placeholders/variables that should be set using the label mask INDArray(s) from the DataSet or MultiDataSet.
TrainingConfig.Builder	dataSetLabelMaskMapping(String... dataSetLabelMaskMapping) See dataSetLabelMaskMapping(List)
TrainingConfig.Builder	l1(double l1) Sets the L1 regularization coefficient for all trainable parameters.
TrainingConfig.Builder	l2(double l2) Sets the L2 regularization coefficient for all trainable parameters.
TrainingConfig.Builder	markLabelsUnused() Calling this method will mark the label as unused.
TrainingConfig.Builder	minimize(boolean minimize) Sets whether the loss function should be minimized (true) or maximized (false). The loss function is usually minimized in SGD. Default: true.
TrainingConfig.Builder	minimize(String... lossVariables)
TrainingConfig.Builder	regularization(List<Regularization> regularization) Set the regularization for all trainable parameters in the network.
TrainingConfig.Builder	regularization(Regularization... regularization) Set the regularization for all trainable parameters in the network.
TrainingConfig.Builder	skipBuilderValidation(boolean skip)
TrainingConfig.Builder	trainEvaluation(@NonNull SDVariable variable, int labelIndex, IEvaluation... evaluations) Add requested History training evaluations for a parm/variable.
TrainingConfig.Builder	trainEvaluation(@NonNull String variableName, int labelIndex, IEvaluation... evaluations) Add requested History training evaluations for a parm/variable.
TrainingConfig.Builder	updater(IUpdater updater) Set the updater (such as Adam, Nesterovs etc.
TrainingConfig.Builder	validationEvaluation(@NonNull SDVariable variable, int labelIndex, IEvaluation... evaluations) Add requested History validation evaluations for a parm/variable.
TrainingConfig.Builder	validationEvaluation(@NonNull String variableName, int labelIndex, IEvaluation... evaluations) Add requested History validation evaluations for a parm/variable.
TrainingConfig.Builder	weightDecay(double coefficient, boolean applyLR) Add weight decay regularization for all trainable parameters.

Methods inherited from class java.lang.Object

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

Constructor Detail

Builder

public Builder()

Method Detail

updater

public TrainingConfig.Builder updater(IUpdater updater)

Set the updater (such as Adam, Nesterovs etc. This is also how the learning rate (or learning rate schedule) is set.

Parameters:

updater - Updater to set

l1

public TrainingConfig.Builder l1(double l1)

Sets the L1 regularization coefficient for all trainable parameters. Must be >= 0.
See L1Regularization for more details

Parameters:

l1 - L1 regularization coefficient

l2

public TrainingConfig.Builder l2(double l2)

Sets the L2 regularization coefficient for all trainable parameters. Must be >= 0.
Note: Generally, WeightDecay (set via weightDecay(double,boolean) should be preferred to L2 regularization. See WeightDecay javadoc for further details.
Note: L2 regularization and weight decay usually should not be used together; if any weight decay (or L2) has been added for the biases, these will be removed first.

See Also:

weightDecay(double, boolean)

weightDecay

public TrainingConfig.Builder weightDecay(double coefficient,
                                          boolean applyLR)

Add weight decay regularization for all trainable parameters. See WeightDecay for more details.
Note: values set by this method will be applied to all applicable layers in the network, unless a different value is explicitly set on a given layer. In other words: values set via this method are used as the default value, and can be overridden on a per-layer basis.

Parameters:

coefficient - Weight decay regularization coefficient

applyLR - Whether the learning rate should be multiplied in when performing weight decay updates. See WeightDecay for more details.

addRegularization

public TrainingConfig.Builder addRegularization(Regularization... regularizations)

Add regularization to all trainable parameters in the network

Parameters:

regularizations - Regularization type(s) to add

regularization

public TrainingConfig.Builder regularization(Regularization... regularization)

Set the regularization for all trainable parameters in the network. Note that if any existing regularization types have been added, they will be removed

Parameters:

regularization - Regularization type(s) to add

regularization

public TrainingConfig.Builder regularization(List<Regularization> regularization)

Set the regularization for all trainable parameters in the network. Note that if any existing regularization types have been added, they will be removed

Parameters:

regularization - Regularization type(s) to add

minimize

public TrainingConfig.Builder minimize(boolean minimize)

Sets whether the loss function should be minimized (true) or maximized (false).
The loss function is usually minimized in SGD.
Default: true.

Parameters:

minimize - True to minimize, false to maximize

dataSetFeatureMapping

public TrainingConfig.Builder dataSetFeatureMapping(String... dataSetFeatureMapping)

Set the name of the placeholders/variables that should be set using the feature INDArray(s) from the DataSet or MultiDataSet. For example, if the network had 2 inputs called "input1" and "input2" and the MultiDataSet features should be mapped with MultiDataSet.getFeatures(0)->"input1" and MultiDataSet.getFeatures(1)->"input2", then this should be set to List<>("input1", "input2").

Parameters:

dataSetFeatureMapping - Name of the variables/placeholders that the feature arrays should be mapped to

dataSetFeatureMapping

public TrainingConfig.Builder dataSetFeatureMapping(List<String> dataSetFeatureMapping)

Set the name of the placeholders/variables that should be set using the feature INDArray(s) from the DataSet or MultiDataSet. For example, if the network had 2 inputs called "input1" and "input2" and the MultiDataSet features should be mapped with MultiDataSet.getFeatures(0)->"input1" and MultiDataSet.getFeatures(1)->"input2", then this should be set to "input1", "input2".

Parameters:

dataSetFeatureMapping - Name of the variables/placeholders that the feature arrays should be mapped to

dataSetLabelMapping

public TrainingConfig.Builder dataSetLabelMapping(String... dataSetLabelMapping)

Set the name of the placeholders/variables that should be set using the labels INDArray(s) from the DataSet or MultiDataSet. For example, if the network had 2 labels called "label1" and "label2" and the MultiDataSet labels should be mapped with MultiDataSet.getLabel(0)->"label1" and MultiDataSet.getLabels(1)->"label", then this should be set to "label1", "label2".

Parameters:

dataSetLabelMapping - Name of the variables/placeholders that the label arrays should be mapped to

dataSetLabelMapping

public TrainingConfig.Builder dataSetLabelMapping(List<String> dataSetLabelMapping)

Set the name of the placeholders/variables that should be set using the labels INDArray(s) from the DataSet or MultiDataSet. For example, if the network had 2 labels called "label1" and "label2" and the MultiDataSet labels should be mapped with MultiDataSet.getLabel(0)->"label1" and MultiDataSet.getLabels(1)->"label", then this should be set to "label1", "label2".

Parameters:

dataSetLabelMapping - Name of the variables/placeholders that the label arrays should be mapped to

markLabelsUnused

public TrainingConfig.Builder markLabelsUnused()

Calling this method will mark the label as unused. This is basically a way to turn off label mapping validation in TrainingConfig builder, for training models without labels.
Put another way: usually you need to call dataSetLabelMapping(String...) to set labels, this method allows you to say that the DataSet/MultiDataSet labels aren't used in training.

dataSetFeatureMaskMapping

public TrainingConfig.Builder dataSetFeatureMaskMapping(String... dataSetFeatureMaskMapping)

See dataSetFeatureMaskMapping(List)

dataSetFeatureMaskMapping

public TrainingConfig.Builder dataSetFeatureMaskMapping(List<String> dataSetFeatureMaskMapping)

Set the name of the placeholders/variables that should be set using the feature mask INDArray(s) from the DataSet or MultiDataSet. For example, if the network had 2 mask variables called "mask1" and "mask2" and the MultiDataSet features masks should be mapped with MultiDataSet.getFeatureMaskArray(0)->"mask1" and MultiDataSet.getFeatureMaskArray(1)->"mask2", then this should be set to "mask1", "mask2".

Parameters:

dataSetFeatureMaskMapping - Name of the variables/placeholders that the feature arrays should be mapped to

dataSetLabelMaskMapping

public TrainingConfig.Builder dataSetLabelMaskMapping(String... dataSetLabelMaskMapping)

See dataSetLabelMaskMapping(List)

dataSetLabelMaskMapping

public TrainingConfig.Builder dataSetLabelMaskMapping(List<String> dataSetLabelMaskMapping)

Set the name of the placeholders/variables that should be set using the label mask INDArray(s) from the DataSet or MultiDataSet. For example, if the network had 2 mask variables called "mask1" and "mask2" and the MultiDataSet label masks should be mapped with MultiDataSet.getLabelMaskArray(0)->"mask1" and MultiDataSet.getLabelMaskArray(1)->"mask2", then this should be set to "mask1", "mask2".

Parameters:

dataSetLabelMaskMapping - Name of the variables/placeholders that the feature arrays should be mapped to

skipBuilderValidation

public TrainingConfig.Builder skipBuilderValidation(boolean skip)

minimize

public TrainingConfig.Builder minimize(String... lossVariables)

trainEvaluation

public TrainingConfig.Builder trainEvaluation(@NonNull
                                              @NonNull String variableName,
                                              int labelIndex,
                                              @NonNull
                                              IEvaluation... evaluations)

Add requested History training evaluations for a parm/variable. These evaluations will be reported in the History object returned by fit.

Parameters:

variableName - The variable to evaluate

labelIndex - The index of the label to evaluate against

evaluations - The evaluations to run

trainEvaluation

public TrainingConfig.Builder trainEvaluation(@NonNull
                                              @NonNull SDVariable variable,
                                              int labelIndex,
                                              @NonNull
                                              IEvaluation... evaluations)

Add requested History training evaluations for a parm/variable. These evaluations will be reported in the History object returned by fit.

Parameters:

variable - The variable to evaluate

labelIndex - The index of the label to evaluate against

evaluations - The evaluations to run

validationEvaluation

public TrainingConfig.Builder validationEvaluation(@NonNull
                                                   @NonNull String variableName,
                                                   int labelIndex,
                                                   @NonNull
                                                   IEvaluation... evaluations)

Add requested History validation evaluations for a parm/variable. These evaluations will be reported in the History object returned by fit.

Parameters:

variableName - The variable to evaluate

labelIndex - The index of the label to evaluate against

evaluations - The evaluations to run

validationEvaluation

public TrainingConfig.Builder validationEvaluation(@NonNull
                                                   @NonNull SDVariable variable,
                                                   int labelIndex,
                                                   @NonNull
                                                   IEvaluation... evaluations)

Add requested History validation evaluations for a parm/variable. These evaluations will be reported in the History object returned by fit.

Parameters:

variable - The variable to evaluate

labelIndex - The index of the label to evaluate against

evaluations - The evaluations to run

addEvaluations

public TrainingConfig.Builder addEvaluations(boolean validation,
                                             @NonNull
                                             @NonNull String variableName,
                                             int labelIndex,
                                             @NonNull
                                             IEvaluation... evaluations)

Add requested evaluations for a parm/variable, for either training or validation. These evaluations will be reported in the History object returned by fit.

Parameters:

validation - Whether to add these evaluations as validation or training

variableName - The variable to evaluate

labelIndex - The index of the label to evaluate against

evaluations - The evaluations to run

build

public TrainingConfig build()