Interface DistributionTrainer

All Known Implementing Classes:
IgnoreCountsTrainer, SimpleDistribution.Trainer, SimpleDistributionTrainer

public interface DistributionTrainer

An object that can be used to train a distribution up.

This lets the distribution implementation handle counts or distributions in the best way possible.

Since:
1.0
Author:
Matthew Pocock
  • Method Details

    • addCount

      Registers that sym was counted in this state.

      This method may be called multiple times with the same symbol. In this case, the times should be summed.

      Parameters:
      dtc - the DistributionTrainerContext within which the count was added
      sym - the Symbol seen
      times - the number of times to add
      Throws:
      IllegalSymbolException - if sym is not recognised
    • getCount

      Get the current count for this state.

      This method may be called multiple times with the same symbol. Each time it should return the agregate of the counts added with addCount since the last invocation of clearCounts.

      Parameters:
      dtc - the DistributionTrainerContext within which the count was added
      sym - the Symbol seen
      Returns:
      the agregate of the counts
      Throws:
      IllegalSymbolException - if sym is not recognised
    • train

      void train(DistributionTrainerContext dtc, double weight) throws ChangeVetoException

      Trains the Distribution, given a null model.

      This will use the information collected with multiple addCount calls, and the null model to generate the new weights.

      This method should not modify the underlying counts.

      Parameters:
      dtc - the context to use
      weight - how many lots of the null model to add
      Throws:
      ChangeVetoException - if the distribution could not have its weights modified
    • clearCounts

      Clears all of the counts to zero.