com.weighscore.neuro
Class Teacher
- public abstract class Teacher
- extends com.weighscore.neuro.ParameterHolder
- This class can should be extended to implement the specific teaching action. The descent class may define teaching parameters as protected class fields, and redefine record methods to work with parameters. Parameters become automatically available to the neural system by ParameterHolder methods.
- Version:
- 2.0
- Author:
- Fyodor Kravchenko
teach
public abstract void teach(double[] error,
com.weighscore.neuro.Signal signal)
- This method is called when the signal was forth and the error was back propagated and the gradient members are known and kept in Signal object. This method can compute the correction of neurons' weights. It can call neurons' teach() method to change the weights
- Parameters:
error
- double[]
signal
- object that holds the values passed through the neurons and the values of back-propagated errors
nextEpoch
public void nextEpoch()
- This method is called from the teaching program (nnt) when the teaching case set (epoch) reached its end. It should be overriden by teachers like conjugate gradient teachers, that parform any action when the epoch changes.
getNeededStatisticNames
public java.lang.String[] getNeededStatisticNames()
- Normally the statistic objects appear automatically after the first attempt of using it. If the teacher needs some statistics to be processed before the first use (like some averages that need to start to be stat computed beforehand), the teacher may override this method to return an array of the needed statistic class names.
- Returns:
- An array of the statistic class names needed to be set to every neuron and synapse of the neural network