Package deepnetts.net.weights

Class RandomWeights

  • public class RandomWeights
extends java.lang.Object
    This class provides various randomization methods.
    Author:
    Zoran Sevarac

    • Constructor Summary

      Constructors 
      Constructor Description
      RandomWeights()  

    • Method Summary

      Modifier and Type Method Description
      static void gaussian​(float[] weights, float mean, float std)  
      static void he​(float[] weights, int numInputs)  
      static void initSeed​(long seed)  
      static void normal​(float[] weights)  
      static void randomize​(float[] array)
      Fills the specified array with random numbers in range [-0.5, 0.5] from the current random seed
      static void uniform​(float[] weights, float min, float max)  
      static void uniform​(float[] weights, int numInputs)
      Uniform U[-a,a] where a=1/sqrt(in).
      static void widrowHoff​(float[] array, float input, float hidden)  
      static void xavier​(float[] weights, int numIn, int numOut)
      Normalized uniform initialization U[-a,a] with a = sqrt(6/(in + out)).

      • Methods inherited from class java.lang.Object

        clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

    • Method Detail

      • initSeed

        public static void initSeed​(long seed)

      • randomize

        public static void randomize​(float[] array)
        Fills the specified array with random numbers in range [-0.5, 0.5] from the current random seed
        Parameters:
        array -

      • widrowHoff

        public static void widrowHoff​(float[] array,
                              float input,
                              float hidden)

      • uniform

        public static void uniform​(float[] weights,
                           int numInputs)
        Uniform U[-a,a] where a=1/sqrt(in).
        Parameters:
        weights - an array of weights
        numInputs - number of inputs, a size of the previous layer

      • uniform

        public static void uniform​(float[] weights,
                           float min,
                           float max)

      • he

        public static void he​(float[] weights,
                      int numInputs)

      • gaussian

        public static void gaussian​(float[] weights,
                            float mean,
                            float std)

      • normal

        public static void normal​(float[] weights)

      • xavier

        public static void xavier​(float[] weights,
                          int numIn,
                          int numOut)
        Normalized uniform initialization U[-a,a] with a = sqrt(6/(in + out)).
        Parameters:
        weights -
        numIn - size of the previous layer (number of inputs)
        numOut - size of initialized layer (number of outputs)