/// <summary>
/// <p>
/// Called to train the network. Iterates until the terminating flag is
/// set. If there are no training event listeners set then there is a
/// hard termination limit of 10000 (by default) iterations.
/// </p>
/// <p>
/// The training algorithm is simple. It presents the training examples
/// in order, up to epochSize before ending the epoch. At that point the
/// error for the epoch is calculated and the EndEpochEvent is sent to all
/// the listeners. The weights of the network are then adjusted. Training
/// stops when the hard limit is reached (given no listeners) or by a listener
/// requesting a termination.
/// </p>
/// </summary>
/// <param name="network"></param>
public void Train(Network network)
{
Dictionary <String, double> results;
terminating = false;
network.BeginTraining(this);
epochCount = 0;
if (errorManager.Count == 0)
{
errorManager.AddCalculator("MSE", new MSEErrorCalculator(trainingExamples.Count));
defaultErrorName = "MSE";
}
if (StartTrainingDelegates != null)
{
StartTrainingDelegates(new TrainingEpochEvent(this, network));
}
while (!terminating)
{
try
{
for (int i = 0; i < epochSize; i++)
{
Random r = new Random();
Example example = trainingExamples[r.Next(0, trainingExamples.Count - 1)];
results = network.Process(example.Inputs);
errorManager.AccumulateError(example.Expected, results);
network.SetFeedback(results, example.Expected);
if (EndPresentationDelegates != null)
{
EndPresentationDelegates(new TrainingEpochEvent(this, network));
}
}
epochCount++;
if (EndEpochDelegates != null)
{
EndEpochDelegates(new TrainingEpochEvent(this, network));
}
UpdateNetwork(network);
if (PostUpdateDelegates != null)
{
PostUpdateDelegates(new TrainingEpochEvent(this, network));
}
if ((epochEventListeners.Count == 0) && (epochCount > hardMax))
{
terminating = true;
}
} catch (Exception e)
{
Console.WriteLine(e.StackTrace);
throw new TrainingException("Unable to train netowrk due to exception", e);
}
}
network.EndTraining();
}
/// <summary>
/// Evaluates the network, returning the evaluation of the error.
/// </summary>
/// <param name="examples">The examples to evaluate</param>
/// <param name="network">The network to evalute</param>
/// <returns>An error report for the valuation.</returns>
public Dictionary <String, double> Evaluate(Network network, ExampleSet examples)
{
ZeroErrors();
for (int i = 0; i < examples.Count; i++)
{
Example example = examples[i];
Dictionary <String, double> results = network.Process(example.Inputs);
CalculateErrors(example.Expected, results);
}
return(errorValues);
}
/// <summary>
/// <p>
/// Called to train the network. Iterates until the terminating flag is
/// set. If there are no training event listeners set then there is a
/// hard termination limit of 10000 (by default) iterations.
/// </p>
/// <p>
/// The training algorithm is simple. It presents the training examples
/// in order, up to epochSize before ending the epoch. At that point the
/// error for the epoch is calculated and the EndEpochEvent is sent to all
/// the listeners. The weights of the network are then adjusted. Training
/// stops when the hard limit is reached (given no listeners) or by a listener
/// requesting a termination.
/// </p>
/// </summary>
/// <param name="network"></param>
public void Train(Network network)
{
Dictionary<String, double> results;
terminating = false;
network.BeginTraining(this);
epochCount = 0;
if(errorManager.Count == 0)
{
errorManager.AddCalculator("MSE", new MSEErrorCalculator(trainingExamples.Count));
defaultErrorName = "MSE";
}
if(StartTrainingDelegates != null) {
StartTrainingDelegates(new TrainingEpochEvent(this, network));
}
while(!terminating)
{
try
{
for(int i = 0; i < epochSize; i++)
{
Random r = new Random();
Example example = trainingExamples[r.Next(0, trainingExamples.Count - 1)];
results = network.Process(example.Inputs);
errorManager.AccumulateError(example.Expected, results);
network.SetFeedback(results, example.Expected);
if(EndPresentationDelegates != null) {
EndPresentationDelegates(new TrainingEpochEvent(this, network));
}
}
epochCount++;
if(EndEpochDelegates != null) {
EndEpochDelegates(new TrainingEpochEvent(this, network));
}
UpdateNetwork(network);
if(PostUpdateDelegates != null) {
PostUpdateDelegates(new TrainingEpochEvent(this, network));
}
if((epochEventListeners.Count == 0) && (epochCount > hardMax))
{
terminating = true;
}
} catch(Exception e)
{
Console.WriteLine(e.StackTrace);
throw new TrainingException("Unable to train netowrk due to exception", e);
}
}
network.EndTraining();
}
/// <summary>
/// Evaluates the network, returning the evaluation of the error.
/// </summary>
/// <param name="examples">The examples to evaluate</param>
/// <param name="network">The network to evalute</param>
/// <returns>An error report for the valuation.</returns>
public Dictionary<String, double> Evaluate(Network network, ExampleSet examples)
{
ZeroErrors();
for(int i = 0; i < examples.Count; i++)
{
Example example = examples[i];
Dictionary<String, double> results = network.Process(example.Inputs);
CalculateErrors(example.Expected, results);
}
return errorValues;
}
