Class TrainingConfiguration


  • final class TrainingConfiguration
    extends java.lang.Object
    • Field Detail

      • dropouts

        boolean dropouts
      • learningRate

        double learningRate
      • regularisationL1

        boolean regularisationL1
      • regularisationL1Factor

        double regularisationL1Factor
      • regularisationL2

        boolean regularisationL2
      • regularisationL2Factor

        double regularisationL2Factor
    • Constructor Detail

      • TrainingConfiguration

        TrainingConfiguration()
    • Method Detail

      • equals

        public boolean equals​(java.lang.Object obj)
        Overrides:
        equals in class java.lang.Object
      • hashCode

        public int hashCode()
        Overrides:
        hashCode in class java.lang.Object
      • doL1

        private double doL1​(double current)
      • doL2

        private double doL2​(double current)
      • probabilityDidKeepInput

        double probabilityDidKeepInput​(int layer)
        Used to scale the weights after training with dropouts, and also to adjut the learning rate
        Parameters:
        layer -
        Returns:
        The probabilityToKeep (as in not drop) the input nodes of this layer
      • probabilityWillKeepOutput

        double probabilityWillKeepOutput​(int layer,
                                         int depth)
        Used to modify the activation function – with this probabilty it will be used as is, for the other parts the output is 0.
        Parameters:
        layer -
        depth -
        Returns:
      • regularisation

        java.util.function.DoubleUnaryOperator regularisation()