As stated earlier, I wrote a class called RecurrentNets.java to build an LSTM network. We create a MultilayerNetwork LSTM network that consists of an input layer, four LSTM layers, three dense layers, and an output layer. The input consists of sequences of genetic variants.
We use the BuildBuildLstmNetworks() method with two parameters—number of input for input layers and number of output for output layers, as shown here:
private static final int lstmLayer1Size = 128;private static final int lstmLayer2Size = 128;private static final int denseLayerSize = 32;private static final double dropoutRatio = 0.5;private static final int truncatedBPTTLength = 22;
Now, before we start creating and building the network, let's ...