Java example source code file (BatchNormalizationTest.java)
The BatchNormalizationTest.java Java example source codepackage org.deeplearning4j.nn.layers.normalization;
import org.deeplearning4j.berkeley.Pair;
import org.nd4j.linalg.dataset.api.iterator.DataSetIterator;
import org.deeplearning4j.datasets.iterator.impl.MnistDataSetIterator;
import org.deeplearning4j.nn.api.Layer;
import org.deeplearning4j.nn.api.OptimizationAlgorithm;
import org.deeplearning4j.nn.conf.Updater;
import org.deeplearning4j.nn.conf.layers.*;
import org.deeplearning4j.nn.conf.MultiLayerConfiguration;
import org.deeplearning4j.nn.conf.NeuralNetConfiguration;
import org.deeplearning4j.nn.conf.layers.BatchNormalization;
import org.deeplearning4j.nn.conf.preprocessor.FeedForwardToRnnPreProcessor;
import org.deeplearning4j.nn.conf.preprocessor.RnnToCnnPreProcessor;
import org.deeplearning4j.nn.conf.preprocessor.RnnToFeedForwardPreProcessor;
import org.deeplearning4j.nn.gradient.Gradient;
import org.deeplearning4j.nn.layers.factory.LayerFactories;
import org.deeplearning4j.nn.multilayer.MultiLayerNetwork;
import org.deeplearning4j.nn.params.BatchNormalizationParamInitializer;
import org.deeplearning4j.nn.weights.WeightInit;
import org.junit.Before;
import org.junit.Ignore;
import org.junit.Test;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.dataset.DataSet;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.lossfunctions.LossFunctions;
import java.util.Random;
import static org.junit.Assert.*;
/**
*/
public class BatchNormalizationTest {
protected INDArray dnnInput = Nd4j.linspace(0,31,32).reshape(2,16);
protected INDArray dnnEpsilon = Nd4j.linspace(0,31,32).reshape(2,16);
protected INDArray cnnInput = Nd4j.linspace(0,63,64).reshape(2, 2, 4, 4);
protected INDArray cnnEpsilon = Nd4j.linspace(0,63,64).reshape(2, 2, 4, 4);
@Before
public void doBefore() {
}
protected Layer setupActivations(int nIn, int nOut){
BatchNormalization bN = new BatchNormalization.Builder().nIn(nIn).nOut(nOut).build();
NeuralNetConfiguration conf = new NeuralNetConfiguration.Builder()
.iterations(1).layer(bN).build();
int numParams = LayerFactories.getFactory(conf).initializer().numParams(conf,true);
INDArray params = Nd4j.create(1, numParams);
Layer layer = LayerFactories.getFactory(conf).create(conf, null, 0, params, true);
return layer;
}
@Test
public void testDnnShapeBatchNormForward() {
Layer layer = setupActivations(2, 16);
// Confirm param initial shape before override
assertArrayEquals(new int[]{1,16}, layer.getParam("gamma").shape());
assertArrayEquals(new int[]{1,16}, layer.getParam("beta").shape());
layer.setParam("gamma", Nd4j.linspace(0,15,16));
layer.setParam("beta", Nd4j.linspace(0,15,16));
INDArray activationsActual = layer.preOutput(dnnInput);
INDArray activationsExpected = Nd4j.create(new double[] {
0.00000000e+00, 7.81248399e-11, 1.56249680e-10,
2.34374298e-10, 3.12499360e-10, 3.90624422e-10,
4.68748595e-10, 5.46873657e-10, 6.24998719e-10,
7.03122893e-10, 7.81248843e-10, 8.59373017e-10,
9.37497191e-10, 1.01562314e-09, 1.09374731e-09,
1.17187327e-09, 0.00000000e+00, 2.00000000e+00,
4.00000000e+00, 6.00000000e+00, 8.00000000e+00,
1.00000000e+01, 1.20000000e+01, 1.40000000e+01,
1.60000000e+01, 1.80000000e+01, 2.00000000e+01,
2.20000000e+01, 2.40000000e+01, 2.60000000e+01,
2.80000000e+01, 3.00000000e+01
},new int[]{2, 16});
assertEquals(activationsExpected, activationsActual);
assertArrayEquals(activationsExpected.shape(), activationsActual.shape());
}
@Test
public void testDnnShapeBatchNormBack(){
Layer layer = setupActivations(2, 16);
layer.setParam("gamma", Nd4j.linspace(0,15,16));
layer.setParam("beta", Nd4j.linspace(0,15,16));
layer.preOutput(dnnInput);
layer.setBackpropGradientsViewArray(Nd4j.create(1,32));
Pair<Gradient, INDArray> actualOut = layer.backpropGradient(dnnEpsilon);
INDArray dnnExpectedEpsilonOut = Nd4j.create(new double[] {
0.00000000e+00, -1.56249680e-10, -3.12499360e-10,
-4.68748595e-10, -6.24998719e-10, -7.81248843e-10,
-9.37497191e-10, -1.09374731e-09, -1.24999744e-09,
-1.40624934e-09, -1.56249413e-09, -1.71874603e-09,
-1.87499438e-09, -2.03124273e-09, -2.18749818e-09,
-2.34373942e-09, 0.00000000e+00, 1.56249680e-10,
3.12499804e-10, 4.68748595e-10, 6.24997831e-10,
7.81248843e-10, 9.37497191e-10, 1.09374731e-09,
1.24999744e-09, 1.40624579e-09, 1.56249769e-09,
1.71874603e-09, 1.87499438e-09, 2.03124983e-09,
2.18749818e-09, 2.34374653e-09
},new int[]{2, 16});
// short calculation expected output
INDArray dnnExpectedEpsilonOutOther = Nd4j.create(new double[] {
16., 15., 14., 13., 12., 11., 10., 9., 8., 7., 6.,
5., 4., 3., 2., 1., -16., -15., -14., -13., -12., -11.,
-10., -9., -8., -7., -6., -5., -4., -3., -2., -1.
},new int[]{2, 16});
INDArray expectedGGamma = Nd4j.create(new double[]
{
16., 16., 16., 16., 16., 16., 16., 16., 16., 16., 16.,
16., 16., 16., 16., 16.
}, new int[] {1, 16});
INDArray expectedBeta = Nd4j.create(new double[]
{
16., 18., 20., 22., 24., 26., 28., 30., 32., 34., 36.,
38., 40., 42., 44., 46.
}, new int[] {1, 16});
// arrays are the same but assert does not see that
assertEquals(dnnExpectedEpsilonOut, actualOut.getSecond());
assertEquals(expectedGGamma, actualOut.getFirst().getGradientFor("gamma"));
assertEquals(expectedBeta, actualOut.getFirst().getGradientFor("beta"));
}
@Test
public void testCnnShapeBatchNormForward() {
Layer layer = setupActivations(2, 2);
// Confirm param initial shape before override
assertArrayEquals(new int[]{1,2}, layer.getParam("gamma").shape());
assertArrayEquals(new int[]{1,2}, layer.getParam("beta").shape());
layer.setParam("gamma", Nd4j.linspace(2,3,2));
layer.setParam("beta", Nd4j.linspace(2,3,2));
INDArray activationsActual = layer.preOutput(cnnInput);
INDArray activationsExpected = Nd4j.create(new double[] {
3.90625310e-11, 3.90625310e-11, 3.90625310e-11,
3.90625310e-11, 3.90625310e-11, 3.90625310e-11,
3.90625310e-11, 3.90625310e-11, 3.90625310e-11,
3.90625310e-11, 3.90625310e-11, 3.90625310e-11,
3.90625310e-11, 3.90625310e-11, 3.90625310e-11,
3.90625310e-11, 5.85940185e-11, 5.85940185e-11,
5.85940185e-11, 5.85940185e-11, 5.85940185e-11,
5.85940185e-11, 5.85940185e-11, 5.85940185e-11,
5.85940185e-11, 5.85940185e-11, 5.85940185e-11,
5.85940185e-11, 5.85940185e-11, 5.85940185e-11,
5.85940185e-11, 5.85940185e-11, 4.00000000e+00,
4.00000000e+00, 4.00000000e+00, 4.00000000e+00,
4.00000000e+00, 4.00000000e+00, 4.00000000e+00,
4.00000000e+00, 4.00000000e+00, 4.00000000e+00,
4.00000000e+00, 4.00000000e+00, 4.00000000e+00,
4.00000000e+00, 4.00000000e+00, 4.00000000e+00,
6.00000000e+00, 6.00000000e+00, 6.00000000e+00,
6.00000000e+00, 6.00000000e+00, 6.00000000e+00,
6.00000000e+00, 6.00000000e+00, 6.00000000e+00,
6.00000000e+00, 6.00000000e+00, 6.00000000e+00,
6.00000000e+00, 6.00000000e+00, 6.00000000e+00,
6.00000000e+00
},new int[]{2,2,4,4});
assertEquals(activationsExpected, activationsActual);
assertArrayEquals(activationsExpected.shape(), activationsActual.shape());
}
@Test
public void testCnnShapeBatchNormBack(){
Layer layer = setupActivations(2, 2);
layer.setParam("gamma", Nd4j.linspace(2,3,2));
layer.setParam("beta", Nd4j.linspace(2,3,2));
layer.preOutput(cnnInput);
layer.setBackpropGradientsViewArray(Nd4j.create(1,4));
Pair<Gradient, INDArray> actualOut = layer.backpropGradient(cnnEpsilon);
INDArray expectedEpsilonOut = Nd4j.create(new double[] {
-7.81250620e-11, -7.81250620e-11, -7.81250620e-11,
-7.81250620e-11, -7.81250620e-11, -7.81250620e-11,
-7.81250620e-11, -7.81250620e-11, -7.81250620e-11,
-7.81250620e-11, -7.81250620e-11, -7.81250620e-11,
-7.81250620e-11, -7.81250620e-11, -7.81250620e-11,
-7.81250620e-11, -1.17187149e-10, -1.17187149e-10,
-1.17187149e-10, -1.17187149e-10, -1.17188037e-10,
-1.17187149e-10, -1.17187149e-10, -1.17187149e-10,
-1.17187149e-10, -1.17187149e-10, -1.17187149e-10,
-1.17188037e-10, -1.17186261e-10, -1.17186261e-10,
-1.17188037e-10, -1.17188037e-10, 7.81250620e-11,
7.81250620e-11, 7.81246179e-11, 7.81255061e-11,
7.81250620e-11, 7.81250620e-11, 7.81246179e-11,
7.81255061e-11, 7.81255061e-11, 7.81250620e-11,
7.81246179e-11, 7.81246179e-11, 7.81255061e-11,
7.81255061e-11, 7.81250620e-11, 7.81246179e-11,
1.17187149e-10, 1.17187149e-10, 1.17188037e-10,
1.17187149e-10, 1.17188037e-10, 1.17187149e-10,
1.17187149e-10, 1.17188037e-10, 1.17187149e-10,
1.17188037e-10, 1.17187149e-10, 1.17186261e-10,
1.17188037e-10, 1.17188037e-10, 1.17188037e-10,
1.17186261e-10
},new int[]{2, 2, 4, 4});
INDArray expectedGGamma = Nd4j.create(new double[]
{
512, 512
}, new int[] {1, 2});
INDArray expectedBeta = Nd4j.create(new double[]
{
752, 1264
}, new int[] {1, 2});
// arrays are the same but assert does not see that
assertEquals(expectedEpsilonOut, actualOut.getSecond());
assertEquals(expectedGGamma, actualOut.getFirst().getGradientFor("gamma"));
assertEquals(expectedBeta, actualOut.getFirst().getGradientFor("beta"));
}
@Test
public void testMultiCNNBNLayer() throws Exception {
MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
.optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT)
.iterations(1)
.seed(123)
.list()
.layer(0, new ConvolutionLayer.Builder().nIn(1).nOut(6).weightInit(WeightInit.XAVIER).activation("relu").build())
.layer(1, new BatchNormalization.Builder().build())
.layer(2, new DenseLayer.Builder().nOut(2).build())
.layer(3, new OutputLayer.Builder(LossFunctions.LossFunction.MCXENT)
.weightInit(WeightInit.XAVIER)
.activation("softmax")
.nIn(2).nOut(10).build())
.backprop(true).pretrain(false)
.cnnInputSize(28,28,1)
.build();
MultiLayerNetwork network = new MultiLayerNetwork(conf);
network.init();
DataSetIterator iter = new MnistDataSetIterator(2, 2);
DataSet next = iter.next();
network.setInput(next.getFeatureMatrix());
INDArray activationsActual = network.preOutput(next.getFeatureMatrix());
assertEquals(10, activationsActual.shape()[1], 1e-2);
network.fit(next);
INDArray actualGammaParam = network.getLayer(1).getParam(BatchNormalizationParamInitializer.GAMMA);
INDArray actualBetaParam = network.getLayer(1).getParam(BatchNormalizationParamInitializer.BETA);
assertTrue(actualGammaParam != null);
assertTrue(actualBetaParam != null);
}
@Test
public void testDBNBNMultiLayer() throws Exception {
DataSetIterator iter = new MnistDataSetIterator(2, 2);
DataSet next = iter.next();
// Run with separate activation layer
MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
.optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT)
.iterations(2)
.seed(123)
.list()
.layer(0, new DenseLayer.Builder().nIn(28*28*1).nOut(10).weightInit(WeightInit.XAVIER).activation("relu").build())
.layer(1, new BatchNormalization.Builder().nOut(10).build())
.layer(2, new ActivationLayer.Builder().activation("relu").build())
.layer(3, new OutputLayer.Builder(LossFunctions.LossFunction.MCXENT)
.weightInit(WeightInit.XAVIER)
.activation("softmax")
.nIn(10).nOut(10).build())
.backprop(true).pretrain(false)
.build();
MultiLayerNetwork network = new MultiLayerNetwork(conf);
network.init();
network.setInput(next.getFeatureMatrix());
INDArray activationsActual = network.preOutput(next.getFeatureMatrix());
assertEquals(10, activationsActual.shape()[1], 1e-2);
network.fit(next);
INDArray actualGammaParam = network.getLayer(1).getParam(BatchNormalizationParamInitializer.GAMMA);
INDArray actualBetaParam = network.getLayer(1).getParam(BatchNormalizationParamInitializer.BETA);
assertTrue(actualGammaParam != null);
assertTrue(actualBetaParam != null);
}
@Ignore@Test
public void testMultiLSTMBNLayer() throws Exception {
// TODO once BatchNorm setup for RNN, expand this test
int nChannelsIn = 3;
int inputSize = 10*10*nChannelsIn; //10px x 10px x 3 channels
int miniBatchSize = 4;
int timeSeriesLength = 10;
int nClasses = 3;
Nd4j.getRandom().setSeed(12345);
INDArray input = Nd4j.rand(new int[]{miniBatchSize,inputSize,timeSeriesLength});
INDArray labels = Nd4j.zeros(miniBatchSize, nClasses, timeSeriesLength);
Random r = new Random(12345);
for( int i = 0; i < miniBatchSize; i++ ){
for(int j = 0; j < timeSeriesLength; j++) {
int idx = r.nextInt(nClasses);
labels.putScalar(new int[]{i,idx,j}, 1.0);
}
}
MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
.seed(12345)
.list()
.layer(0, new ConvolutionLayer.Builder(5, 5).nIn(3).nOut(5).stride(1, 1)
.activation("identity").weightInit(WeightInit.XAVIER).updater(Updater.NONE).build()) //Out: (10-5)/1+1 = 6 -> 6x6x5
.layer(1, new BatchNormalization.Builder().build())
.layer(2, new ActivationLayer.Builder().activation("relu").build())
.layer(3, new SubsamplingLayer.Builder(SubsamplingLayer.PoolingType.MAX)
.kernelSize(2, 2).stride(1, 1).build()) //Out: (6-2)/1+1 = 5 -> 5x5x5
.layer(4, new DenseLayer.Builder().nIn(5 * 5 * 5).nOut(4)
.updater(Updater.NONE).weightInit(WeightInit.XAVIER).activation("relu")
.build())
.layer(5, new GravesLSTM.Builder().nIn(4).nOut(3)
.activation("identity").updater(Updater.NONE).weightInit(WeightInit.XAVIER)
.build())
// .layer(6, new BatchNormalization.Builder().build())
// .layer(7, new ActivationLayer.Builder().activation("tanh").build())
.layer(6, new RnnOutputLayer.Builder().nIn(3).nOut(nClasses)
.activation("softmax").lossFunction(LossFunctions.LossFunction.MCXENT)
.updater(Updater.NONE).build())
.cnnInputSize(10, 10, 3)
.pretrain(false).backprop(true)
.build();
conf.getInputPreProcessors().put(0,new RnnToCnnPreProcessor(10, 10, 3));
MultiLayerNetwork mln = new MultiLayerNetwork(conf);
mln.init();
mln.setInput(input);
mln.setLabels(labels);
mln.fit();
}
@Test
public void testCNNBNActivationCombo() throws Exception {
DataSetIterator iter = new MnistDataSetIterator(2, 2);
DataSet next = iter.next();
MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
.optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT)
.iterations(2)
.seed(123)
.list()
.layer(0, new ConvolutionLayer.Builder().nIn(1).nOut(6).weightInit(WeightInit.XAVIER).activation("identity").build())
.layer(1, new BatchNormalization.Builder().build())
.layer(2, new ActivationLayer.Builder().activation("relu").build())
.layer(3, new OutputLayer.Builder(LossFunctions.LossFunction.MCXENT)
.weightInit(WeightInit.XAVIER)
.activation("softmax")
.nOut(10).build())
.backprop(true).pretrain(false)
.cnnInputSize(28,28,1)
.build();
MultiLayerNetwork network = new MultiLayerNetwork(conf);
network.init();
network.fit(next);
assertNotEquals(null, network.getLayer(0).getParam("W"));
assertNotEquals(null, network.getLayer(0).getParam("b"));
}
@Ignore@Test
public void testMultiLSTMLayer() throws Exception {
// TODO use this to test when batch norm implemented for RNN
int inputSize = 10*10;
int miniBatchSize = 4;
int timeSeriesLength = 10;
int nClasses = 3;
Nd4j.getRandom().setSeed(12345);
INDArray input = Nd4j.rand(new int[]{miniBatchSize,inputSize,timeSeriesLength});
INDArray labels = Nd4j.zeros(miniBatchSize, nClasses, timeSeriesLength);
Random r = new Random(12345);
for( int i = 0; i < miniBatchSize; i++ ){
for(int j = 0; j < timeSeriesLength; j++) {
int idx = r.nextInt(nClasses);
labels.putScalar(new int[]{i,idx,j}, 1.0);
}
}
// Run without separate activation layer
MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
.seed(12345)
.list()
.layer(0, new GravesLSTM.Builder().nIn(100).nOut(3)
.activation("tanh").updater(Updater.NONE).weightInit(WeightInit.XAVIER)
.build())
.layer(1, new RnnOutputLayer.Builder().nIn(3).nOut(nClasses)
.activation("softmax").lossFunction(LossFunctions.LossFunction.MCXENT)
.updater(Updater.NONE).build())
.pretrain(false).backprop(true)
.build();
MultiLayerNetwork mln = new MultiLayerNetwork(conf);
mln.init();
mln.setInput(input);
mln.setLabels(labels);
mln.fit();
// Run with separate activation layer
MultiLayerConfiguration conf2 = new NeuralNetConfiguration.Builder()
.seed(12345)
.list()
.layer(0, new GravesLSTM.Builder().nIn(100).nOut(3)
.activation("identity").updater(Updater.NONE).weightInit(WeightInit.XAVIER)
.build())
// .layer(, new BatchNormalization.Builder().build())
.layer(1, new ActivationLayer.Builder().activation("tanh").build())
.layer(2, new RnnOutputLayer.Builder().nIn(3).nOut(nClasses)
.activation("softmax").lossFunction(LossFunctions.LossFunction.MCXENT)
.updater(Updater.NONE).build())
.pretrain(false).backprop(true)
.build();
conf2.getInputPreProcessors().put(1,new RnnToFeedForwardPreProcessor());
conf2.getInputPreProcessors().put(2,new FeedForwardToRnnPreProcessor());
MultiLayerNetwork mln2 = new MultiLayerNetwork(conf2);
mln2.init();
mln2.setInput(input);
mln2.setLabels(labels);
mln2.fit();
// assertEquals(mln.getLayer(0).getParam("W"), mln2.getLayer(0).getParam("W"));
// assertEquals(mln.getLayer(3).getParam("W"), mln2.getLayer(4).getParam("W"));
// assertEquals(mln.getLayer(0).getParam("b"), mln2.getLayer(0).getParam("b"));
// assertEquals(mln.getLayer(3).getParam("b"), mln2.getLayer(4).getParam("b"));
}
@Ignore@Test
public void testMultiCNNLSTMLayer() throws Exception {
// TODO use this to test when batch norm implemented for RNN
int nChannelsIn = 3;
int inputSize = 10*10*nChannelsIn; //10px x 10px x 3 channels
int miniBatchSize = 4;
int timeSeriesLength = 10;
int nClasses = 3;
Nd4j.getRandom().setSeed(12345);
INDArray input = Nd4j.rand(new int[]{miniBatchSize,inputSize,timeSeriesLength});
INDArray labels = Nd4j.zeros(miniBatchSize, nClasses, timeSeriesLength);
Random r = new Random(12345);
for( int i = 0; i < miniBatchSize; i++ ){
for(int j = 0; j < timeSeriesLength; j++) {
int idx = r.nextInt(nClasses);
labels.putScalar(new int[]{i,idx,j}, 1.0);
}
}
// Run without separate activation layer
MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
.seed(12345)
.list()
.layer(0, new ConvolutionLayer.Builder(5, 5).nIn(3).nOut(5).stride(1, 1)
.activation("relu").weightInit(WeightInit.XAVIER).updater(Updater.NONE).build()) //Out: (10-5)/1+1 = 6 -> 6x6x5
.layer(1, new SubsamplingLayer.Builder(SubsamplingLayer.PoolingType.MAX)
.kernelSize(2, 2).stride(1, 1).build()) //Out: (6-2)/1+1 = 5 -> 5x5x5
.layer(2, new DenseLayer.Builder().nIn(5 * 5 * 5).nOut(4)
.updater(Updater.NONE).weightInit(WeightInit.XAVIER).activation("relu")
.build())
.layer(3, new GravesLSTM.Builder().nIn(4).nOut(3)
.activation("tanh").updater(Updater.NONE).weightInit(WeightInit.XAVIER)
.build())
.layer(4, new RnnOutputLayer.Builder().nIn(3).nOut(nClasses)
.activation("softmax").lossFunction(LossFunctions.LossFunction.MCXENT)
.updater(Updater.NONE).build())
.cnnInputSize(10, 10, 3)
.pretrain(false).backprop(true)
.build();
conf.getInputPreProcessors().put(0,new RnnToCnnPreProcessor(10, 10, 3));
MultiLayerNetwork mln = new MultiLayerNetwork(conf);
mln.init();
mln.setInput(input);
mln.setLabels(labels);
mln.fit();
// Run with separate activation layer
MultiLayerConfiguration conf2 = new NeuralNetConfiguration.Builder()
.seed(12345)
.list()
.layer(0, new ConvolutionLayer.Builder(5, 5).nIn(3).nOut(5).stride(1, 1)
.activation("identity").weightInit(WeightInit.XAVIER).updater(Updater.NONE).build()) //Out: (10-5)/1+1 = 6 -> 6x6x5
.layer(1, new BatchNormalization.Builder().build())
.layer(2, new ActivationLayer.Builder().activation("relu").build())
.layer(3, new SubsamplingLayer.Builder(SubsamplingLayer.PoolingType.MAX)
.kernelSize(2, 2).stride(1, 1).build()) //Out: (6-2)/1+1 = 5 -> 5x5x5
.layer(4, new DenseLayer.Builder().nIn(5 * 5 * 5).nOut(4)
.updater(Updater.NONE).weightInit(WeightInit.XAVIER).activation("relu")
.build())
.layer(5, new GravesLSTM.Builder().nIn(4).nOut(3)
.activation("identity").updater(Updater.NONE).weightInit(WeightInit.XAVIER)
.build())
.layer(6, new BatchNormalization.Builder().build())
.layer(7, new ActivationLayer.Builder().activation("tanh").build())
.layer(8, new RnnOutputLayer.Builder().nIn(3).nOut(nClasses)
.activation("softmax").lossFunction(LossFunctions.LossFunction.MCXENT)
.updater(Updater.NONE).build())
.cnnInputSize(10, 10, 3)
.pretrain(false).backprop(true)
.build();
conf2.getInputPreProcessors().put(0,new RnnToCnnPreProcessor(10, 10, 3));
MultiLayerNetwork mln2 = new MultiLayerNetwork(conf2);
mln2.init();
mln2.setInput(input);
mln2.setLabels(labels);
mln2.fit();
//
// assertEquals(mln.getLayer(0).getParam("W"), mln2.getLayer(0).getParam("W"));
// assertEquals(mln.getLayer(3).getParam("W"), mln2.getLayer(4).getParam("W"));
// assertEquals(mln.getLayer(0).getParam("b"), mln2.getLayer(0).getParam("b"));
// assertEquals(mln.getLayer(3).getParam("b"), mln2.getLayer(4).getParam("b"));
}
}
Other Java examples (source code examples)Here is a short list of links related to this Java BatchNormalizationTest.java source code file: |
The search page
Other Java source code examples at this package level
Click here to learn more about this project
