initialisation

Type Members

1. case class RandomGaussianInitialisation(seed: Long = 42L, mu: Double = 0.0, sigma: Double = 1.0) extends WeightInitialisationStrategy with Product with Serializable

   

   weight initialisation from a random guassian distribution

2. class RandomWeightInitialisation extends WeightInitialisationStrategy

   

   simple weight initialisation choosing a random double between 0 and 1.0

   simple weight initialisation choosing a random double between 0 and 1.0

   Some suggested initialisations are:

   $$ W_{i,j} = U(-1/\sqrt{m}, 1/\sqrt{m}) $$

   Where $m$ is the number of columns in the input layer.

   $$ W_{i,j} = U(-\sqrt{6/(m+n), \sqrt{6/(m+n)}) $$

   Where $m$ is the number of columns in the input layer and $n$ the number of rows in the data.

3. trait WeightInitialisationStrategy extends AnyRef

   

   weight initialisation strategies may differ depending on method chosen.