rbm

Type Members

1. class BernoulliUnitLayer extends UnscaledSigmoidUnitLayer with RbmLayer with Serializable

2. class CompetitiveRetryTrainingStrategy extends RbmTrainingStrategy

   This training strategy generates multiple Rbm training configurations at random, and selects the Rbm which has best reconstruction error on a held out validation set.

3. case class ConstantConfigurationEarlyStoppingTrainingStrategy(config: RbmTrainingConfiguration, relativeValidationSetSize: Double) extends RbmTrainingStrategy with Product with Serializable

   Simple training strategy that just takes a fixed training configuration, and trains the Rbm with it, until no improvement can be observed.

4. case class ConstantConfigurationFixedEpochsTrainingStrategy(config: RbmTrainingConfiguration) extends RbmTrainingStrategy with Product with Serializable

   Dumbest imaginable training strategy: it just takes an RBM and an RBM training configuration, and does exactly what's told in the configuration, without monitoring the progress in any way.

5. class DefaultRbmTrainingConfiguration extends RbmTrainingConfiguration

   Implements a reasonable training strategy for Rbms.

6. class GaussianUnitLayer extends LinearUnitLayer with RbmLayer with Serializable

7. class RandomRetryTrainingStrategy extends RbmTrainingStrategy

   This training strategy generates multiple Rbm training configurations at random, and selects the Rbm which has best reconstruction error on a held out validation set.

8. class Rbm extends NeuralNet with NeuralNetLike[Rbm]

   Represents a restricted Boltzmann machine, which consists of two layers of neurons connected with a full bipartite graph.

9. sealed trait RbmLayer extends Layer with Cloneable with Serializable

10. class RbmStack extends Visualizable

    Encapsulates training procedures for stacks of Rbm's.

11. abstract class RbmTrainingConfiguration extends AnyRef

    This class provides settings necessary for rbm training.

12. trait RbmTrainingStrategy extends AnyRef

13. class TournamentTrainingStrategy extends RbmTrainingStrategy

    This training strategy generates multiple Rbm training configurations at random, and then trains multiple Rbm's with these configurations in rounds with specified number of epochs.