de.jungblut.classification.nn

Class MultilayerPerceptron.MultilayerPerceptronBuilder

java.lang.Object

de.jungblut.classification.nn.MultilayerPerceptron.MultilayerPerceptronBuilder

Enclosing class:

MultilayerPerceptron

public static final class MultilayerPerceptron.MultilayerPerceptronBuilder
extends Object

Configuration for training a neural net through the Classifier

Method Summary

Modifier and Type	Method and Description
MultilayerPerceptron.MultilayerPerceptronBuilder	batchParallelism(int numThreads)
MultilayerPerceptron	build()
static MultilayerPerceptron.MultilayerPerceptronBuilder	create(int[] layer, ActivationFunction[] activations, LossFunction errorFunction, Minimizer minimizer, int maxIteration) Creates a new TrainingConfiguration with the mandatory configurations of the activation functions, the to be used minimizer and the maximum iterations.
MultilayerPerceptron.MultilayerPerceptronBuilder	hiddenLayerDropout(double d) Sets the hidden layer dropout probability.
MultilayerPerceptron.MultilayerPerceptronBuilder	inputLayerDropout(double d) Sets the input layer dropout probability.
MultilayerPerceptron.MultilayerPerceptronBuilder	lambda(double lambda) Sets the regularization parameter lambda, defaults to 0 if not set.
MultilayerPerceptron.MultilayerPerceptronBuilder	miniBatchSize(int size)
MultilayerPerceptron.MultilayerPerceptronBuilder	stochastic() Sets the training mode to stochastic.
MultilayerPerceptron.MultilayerPerceptronBuilder	stochastic(boolean stochastic) If verbose is true, stochastic training will be used.
MultilayerPerceptron.MultilayerPerceptronBuilder	trainingType(TrainingType type) Sets the training type, it defaults to CPU- so only use if you want to use the GPU.
MultilayerPerceptron.MultilayerPerceptronBuilder	verbose() Sets verbose to true.
MultilayerPerceptron.MultilayerPerceptronBuilder	verbose(boolean verbose) If verbose is true, progress indicators will be printed to STDOUT.
MultilayerPerceptron.MultilayerPerceptronBuilder	withWeights(WeightMatrix[] weights) Sets the initial weights, maybe from an already trained network, or from a fancy random initialization technique.

Methods inherited from class java.lang.Object

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

Method Detail

trainingType

public MultilayerPerceptron.MultilayerPerceptronBuilder trainingType(TrainingType type)

Sets the training type, it defaults to CPU- so only use if you want to use the GPU.

lambda

public MultilayerPerceptron.MultilayerPerceptronBuilder lambda(double lambda)

Sets the regularization parameter lambda, defaults to 0 if not set.

verbose

public MultilayerPerceptron.MultilayerPerceptronBuilder verbose()

Sets verbose to true. Progress indicators will be printed to STDOUT.

verbose

public MultilayerPerceptron.MultilayerPerceptronBuilder verbose(boolean verbose)

If verbose is true, progress indicators will be printed to STDOUT.

hiddenLayerDropout

public MultilayerPerceptron.MultilayerPerceptronBuilder hiddenLayerDropout(double d)

Sets the hidden layer dropout probability.

inputLayerDropout

public MultilayerPerceptron.MultilayerPerceptronBuilder inputLayerDropout(double d)

Sets the input layer dropout probability.

withWeights

public MultilayerPerceptron.MultilayerPerceptronBuilder withWeights(WeightMatrix[] weights)

Sets the initial weights, maybe from an already trained network, or from a fancy random initialization technique.

miniBatchSize

public MultilayerPerceptron.MultilayerPerceptronBuilder miniBatchSize(int size)

Parameters:

size - the minibatch size to use. Batches are calculated in parallel on every cpu core if not overridden by batchParallelism(int).

batchParallelism

public MultilayerPerceptron.MultilayerPerceptronBuilder batchParallelism(int numThreads)

Parameters:

numThreads - set the number of threads where batches should be calculated in parallel.

stochastic

public MultilayerPerceptron.MultilayerPerceptronBuilder stochastic()

Sets the training mode to stochastic.

stochastic

public MultilayerPerceptron.MultilayerPerceptronBuilder stochastic(boolean stochastic)

If verbose is true, stochastic training will be used.

build

public MultilayerPerceptron build()

Returns:

a new MultilayerPerceptron with the given configuration.

create

public static MultilayerPerceptron.MultilayerPerceptronBuilder create(int[] layer,
                                                                      ActivationFunction[] activations,
                                                                      LossFunction errorFunction,
                                                                      Minimizer minimizer,
                                                                      int maxIteration)

Creates a new TrainingConfiguration with the mandatory configurations of the activation functions, the to be used minimizer and the maximum iterations.

Parameters:

layer - the number of neurons for each layer, each index denotes a layer.

activations - the activation functions to be used, each index denotes a layer.

errorFunction - the error function on the last layer.

minimizer - the minimizer to be used.

maxIterations - how many iterations (epochs) to run.

Returns:

a brand new training configuration with the given parameters set.