org.biojava.bio.dist
Interface DistributionTrainerContext

All Known Subinterfaces:
ModelTrainer
All Known Implementing Classes:
SimpleDistributionTrainerContext, SimpleModelTrainer

public interface DistributionTrainerContext

A context within a group of DistributionTrainers can be trained together.

Author:
Matthew Pocock

Method Summary
 void addCount(Distribution dist, Symbol sym, double times)
           Registers that sym was counted in this state.
 void clearCounts()
          Clears all of the counts to zero.
 double getCount(Distribution dist, Symbol sym)
          Return the number of counts of a particular symbol which will be used to train the specified distribution.
 double getNullModelWeight()
          Return the number of pseudocounts added to the distribution when training.
 DistributionTrainer getTrainer(Distribution dist)
          Return the Distribution trainer object from the current context.
 void registerDistribution(Distribution dist)
           Register a distribution object with this context.
 void registerTrainer(Distribution dist, DistributionTrainer trainer)
           Register a Distribution and an associated DistributionTrainer object.
 void setNullModelWeight(double weight)
          Set the number of pseudocounts to add when training the distribution.
 void train()
           Trains the Distribution, given a null model.
 

Method Detail

getNullModelWeight

public double getNullModelWeight()
Return the number of pseudocounts added to the distribution when training.

Returns:
the null model weight

setNullModelWeight

public void setNullModelWeight(double weight)
Set the number of pseudocounts to add when training the distribution. These counts are added in proportion to the null model of the distribution being trained.

Parameters:
weight - the new null model weight

registerDistribution

public void registerDistribution(Distribution dist)

Register a distribution object with this context.

This method is a request to the context to register dist. If dist is already registered then this method should do nothing. If it is not registered, then it should invoke dist.registerWithTrainer

Parameters:
dist - the Distribution to register

registerTrainer

public void registerTrainer(Distribution dist,
                            DistributionTrainer trainer)

Register a Distribution and an associated DistributionTrainer object.

In the registerWithTrainer method of a Distribution, it should associate itself with a trainer using this method.

Parameters:
dist - the distribution to be registered.
trainer - the distribution's trainer object to be registered.

getTrainer

public DistributionTrainer getTrainer(Distribution dist)
Return the Distribution trainer object from the current context.

Parameters:
dist - the Distribution whose trainer is required.

addCount

public void addCount(Distribution dist,
                     Symbol sym,
                     double times)
              throws IllegalSymbolException

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:
dist - the Distribution that the symbol was associated with
sym - the Symbol seen
times - the number of times to add
Throws:
IllegalSymbolException

getCount

public double getCount(Distribution dist,
                       Symbol sym)
                throws IllegalSymbolException
Return the number of counts of a particular symbol which will be used to train the specified distribution.

Throws:
IllegalSymbolException

train

public void train()
           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.

Throws:
ChangeVetoException - if any of the distributions can't be trained

clearCounts

public void clearCounts()
Clears all of the counts to zero.