config = new NeuralNetConfiguration.Builder() .weightInit(WeightInit.DISTRIBUTION) .dist(new NormalDistribution(0.0, 0.01)) .activation(Activation.RELU) .updater(new Nesterovs(new StepSchedule(ScheduleType.ITERATION, 1e-2, 0.1, 100000), 0.9))
.optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT) .updater(new RmsProp(0.1, 0.96, 0.001)).weightInit(WeightInit.DISTRIBUTION) .dist(new NormalDistribution(0.0, 0.5)).regularization(true).l2(5e-5).miniBatch(true) .convolutionMode(ConvolutionMode.Truncate).graphBuilder();
builder.weightInit(weightInit.getValue(values)); if (dist != null) builder.dist(dist.getValue(values)); if (learningRate != null) builder.learningRate(learningRate.getValue(values));
.weightInit(WeightInit.DISTRIBUTION).dist(new NormalDistribution(0.0, 0.01)) .activation(Activation.RELU).optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT) .updater(Updater.NESTEROVS).learningRate(1e-2).biasLearningRate(1e-2 * 2).regularization(true)
.seed(seed) .weightInit(WeightInit.DISTRIBUTION) .dist(new NormalDistribution(0.0, 0.01)) .activation(Activation.RELU) .updater(new Nesterovs(new StepSchedule(ScheduleType.ITERATION, 0.1, 0.1, 100000), 0.9))
.optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT) .updater(new RmsProp(0.1, 0.96, 0.001)).weightInit(WeightInit.DISTRIBUTION) .dist(new NormalDistribution(0.0, 0.5)).regularization(true).l1(1e-7).l2(5e-5).miniBatch(true) .convolutionMode(ConvolutionMode.Truncate).graphBuilder();
/** * Deliver access to the internal builder * * @return NeuralNetworkConfiguration */ public org.deeplearning4j.nn.conf.NeuralNetConfiguration.Builder builder() { Builder builder = new Builder(); // Set dist to null if Disabled was chosen as dl4j backend defaults to null builder .l1(l1) .l2(l2) .optimizationAlgo(optimizationAlgo) .seed(seed) .weightInit(weightInit) .dist(dist.getBackend()) .biasInit(biasInit) .updater(updater.getBackend()) .biasUpdater(biasUpdater.getBackend()) .dropOut(dropout.getBackend()) .miniBatch(miniBatch) .minimize(minimize) .weightNoise(weightNoise.getBackend()) .gradientNormalization(gradientNormalization.getBackend()) .gradientNormalizationThreshold(gradientNormalizationThreshold) .inferenceWorkspaceMode(inferenceWorkspaceMode) .trainingWorkspaceMode(trainingWorkspaceMode); builder.setPretrain(pretrain); return builder; }