Hands-On Deep Learning with Apache Spark

RNN distributed training with Spark and DL4J

Let's reconsider the example that has been presented in Chapter 6, Recurrent Neural Networks, section RNNs with DL4J and Spark, about an LSTM that would be trained to generate text, one character at a time. For convenience, let's remind ourselves of the network configuration used there (an LSTM RNN implementation of the model proposed by Alex Graves):

val rng = new Random(12345) val lstmLayerSize: Int = 200 val tbpttLength: Int = 50 val nSamplesToGenerate: Int = 4 val nCharactersToSample: Int = 300 val generationInitialization: String = null val conf = new NeuralNetConfiguration.Builder()     .optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT)     .iterations(1)     .learningRate(0.1) .rmsDecay(0.95) ...