带有自定义矩阵的CNN的DeepLearning4J IllegalArgumentException

Question

I have a custom 7(height) and 24(width) matrix input to be used for training. The output are labels with Age ( Young, Mature, Old). I would like to go with Deeplearning4J Convolutional Neural Networks.

After building a very basic Convolutional Neural Network the very first training item gives the following error and i have no clue what is this about.

Exception in thread "main" java.lang.IllegalArgumentException: Invalid size index 2 wher it's >= rank 2
at org.nd4j.linalg.api.ndarray.BaseNDArray.size(BaseNDArray.java:4066)
at org.deeplearning4j.nn.layers.convolution.ConvolutionLayer.preOutput(ConvolutionLayer.java:192)
at org.deeplearning4j.nn.layers.convolution.ConvolutionLayer.activate(ConvolutionLayer.java:247)
at org.deeplearning4j.nn.graph.vertex.impl.LayerVertex.doForward(LayerVertex.java:88)
at org.deeplearning4j.nn.graph.ComputationGraph.feedForward(ComputationGraph.java:983)
at org.deeplearning4j.nn.graph.ComputationGraph.computeGradientAndScore(ComputationGraph.java:889)

My DL4J code

//Model Config here
MultiLayerConfiguration.Builder builder = new NeuralNetConfiguration.Builder()
    .seed(seed)
    .iterations(iterations)
    .regularization(true).l2(0.0005)
    .learningRate(0.01)//.biasLearningRate(0.02)
    //.learningRateDecayPolicy(LearningRatePolicy.Inverse).lrPolicyDecayRate(0.001).lrPolicyPower(0.75)
    .weightInit(WeightInit.XAVIER)
    .optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT)
    .updater(Updater.NESTEROVS).momentum(0.9) 
    .list()
    .layer(0, new ConvolutionLayer.Builder(4, 1)
        //nIn and nOut specify depth. nIn here is the nChannels and nOut is the number of filters to be applied
            .name("hzvt1")
        .nIn(nChannels)
        .stride(1, 1)
        .nOut(26)
        .activation("relu")//.activation("identity")
        .build())
    .layer(1, new OutputLayer.Builder(LossFunctions.LossFunction.NEGATIVELOGLIKELIHOOD)
        .nOut(outputNum)
        .activation("softmax")
        .build())  
    .setInputType(InputType.convolutional(nChannels,height,width))
    .backprop(true).pretrain(false);

//Model build here            
model.fit(wmTrain);MultiLayerConfiguration conf = builder.build();
model.fit(wmTrain);MultiLayerNetwork model = new MultiLayerNetwork(conf);
model.init();            

//Training data creation here 
INDArray weekMatrix = Nd4j.ones(DLAgeGender.nChannels,DLAgeGender.height*DLAgeGender.width);       
double[] vector = new double[] { 0.0, 1.0, 0.0 };
INDArray intLabels = Nd4j.create(vector);
DataSet ds=new DataSet(weekMatrix,intLabels);
//Train the first item
model.fit(wmTrain);

I am using DL4J version 0.6 , Java Version 1.8, maven 3.3+

I suspect a bug in the library.

Answer 1

With the help of gitter support. I found out that the model and input is not matching. The correct working code is as follows.

I hope those DL4J error/exception messages are more clear in the next releases.

log.info("Build model....");
System.out.println("Building model...");
MultiLayerConfiguration.Builder builder = new NeuralNetConfiguration.Builder()
        .seed(seed)
        .iterations(iterations)
        .regularization(true).l2(0.0005)
        .learningRate(0.01)//.biasLearningRate(0.02)
        //.learningRateDecayPolicy(LearningRatePolicy.Inverse).lrPolicyDecayRate(0.001).lrPolicyPower(0.75)
        .weightInit(WeightInit.XAVIER)
        .optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT)
        .updater(Updater.NESTEROVS).momentum(0.9)
        .list()
        .layer(0, new ConvolutionLayer.Builder(4, 1)
            //nIn and nOut specify depth. nIn here is the nChannels and nOut is the number of filters to be applied
            .name("hzvt1")
            .nIn(nChannels)
            .stride(1, 1)
            .nOut(26)
            .activation("relu")//.activation("identity")
            .build())
        .layer(1, new OutputLayer.Builder(LossFunctions.LossFunction.NEGATIVELOGLIKELIHOOD)
            .nOut(classes)
            .activation("softmax")
            .build())
        .setInputType(InputType.convolutional(height,width,nChannels))
        .backprop(true).pretrain(false);

//Model build here            
model.fit(wmTrain);MultiLayerConfiguration conf = builder.build();
model.fit(wmTrain);MultiLayerNetwork model = new MultiLayerNetwork(conf);
model.init();            

//Training data creation here 
    INDArray weekMatrix = Nd4j.ones(new int[]{1,DLAgeGender.nChannels,DLAgeGender.height,DLAgeGender.width});
    INDArray intLabels;
    double[] vector = new double[] { 0.0, 1.0, };
    intLabels = Nd4j.create(vector);
DataSet ds=new DataSet(weekMatrix,intLabels);

log.info("Train model....");
model.setListeners(new ScoreIterationListener(1));
model.fit(wmTrain);
System.out.println("Data train OK.");