2548-conf-small-hl-working.java

	hiddenWidth = 20;
		
		MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
				.optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT).iterations(1)
				.learningRate(0.05476383804339002) // 0.02 with BPTT
				.momentum(0.5358333571512998)
				.seed(12345)
				.regularization(true)
				.l2(0.016384859214629816)
				.dropOut(0.8145498131858961)
				.gradientNormalization(GradientNormalization.ClipElementWiseAbsoluteValue)
				.gradientNormalizationThreshold(28.22828467793605)
				.weightInit(WeightInit.XAVIER)
				.updater(Updater.NESTEROVS)
				.list()
				.layer(0, new GravesLSTM.Builder()
						.forgetGateBiasInit(1.2662013265922676)
						.nIn(numInput)
						.nOut(hiddenWidth)
						.activation("softsign")
						.build())
				.layer(1, new GravesLSTM.Builder()
						.forgetGateBiasInit(1.2662013265922676)
						.nIn(hiddenWidth)
						.nOut(hiddenWidth)
						.activation("softsign")
						.build())
				.layer(2, new GravesLSTM.Builder()
						.forgetGateBiasInit(1.2662013265922676)
						.nIn(hiddenWidth)
						.nOut(hiddenWidth)
						.activation("tanh")
						.build())
				.layer(3, new RnnOutputLayer.Builder(LossFunction.MSE).activation("identity")        
						.nIn(hiddenWidth).nOut(1).build())
				.backpropType(BackpropType.TruncatedBPTT).tBPTTForwardLength(28).tBPTTBackwardLength(28)
				.pretrain(false).backprop(true)
				.build();
		
		MultiLayerNetwork net = new MultiLayerNetwork(conf);
		net.init();
		net.setListeners(new ScoreIterationListener(1000));