Hi, I’m working on my app that uses LeapMotion sensor. Data I am working on is a distance between two fingers (in CSV file it is a single column, where rows represent every Nth frame). I trained my model with few different examples (fingers moving, steady, moving halfway) and evaluated some test data (also from CSV file) with almost 100% accuracy. But I have 2 problems. I want to gather live data from the sensor and make a prediction after specified amount of frames without use of CSV. How can I transfer this data (list of single distances) into 3 dimensional INDArray? Data from CSV files is 3 dimensional but maximum I managed to create from live data is 2. And the second problem is when I have output from my network (also 3 dimensional INDArray - [ [ [data], [data], [data] ] ] I can’t manage to extract the maximum value from this array to recognize the label - I know I can do this with Evaluation but for this I need labels which I don’t have (I want to recognize the movement). I’ve searched for solution on Stack, in official examples for DL4J on GitHub but haven’t found any solution.
Thank you for your help, below is my network configuration and small part of CSV file.
MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
.seed(123)
.weightInit(WeightInit.XAVIER)
.updater(new Nadam())
.gradientNormalization(GradientNormalization.ClipElementWiseAbsoluteValue)
.gradientNormalizationThreshold(0.5)
.list()
.layer(new LSTM.Builder().activation(Activation.TANH).nIn(1).nOut(1).build())
.layer(new RnnOutputLayer.Builder(LossFunctions.LossFunction.MCXENT)
.activation(Activation.SOFTMAX).nIn(1).nOut(3).build())
.build();
And CSV sample:
85.37913
85.297356
85.69315
86.18485
86.437485
84.70124
79.05874
75.48497
73.63566
72.05912