protected void DoLearning(LearningTask learningTask)
{
if (learningTask.LearningState == ComputationState.Done)
{
return;
}
// Grab options.
var options = learningTask.Options;
// Grab training set.
var trainingSet = learningTask.TrainingSet;
// Grab variable names.
var firstObservation = learningTask.TrainingSet.First();
var variableNames = firstObservation.VariableNames;
// Build a starter Bayesian network.
var bn = new BayesianNetwork(learningTask.TrainingSet.Name);
// Build variables for each variable in the training set.
foreach (var variableName in variableNames)
{
var space = trainingSet.Variables.TryFind(variableName).Value;
var rv = new RandomVariable(variableName, space);
bn.AddVariable(rv);
}
// Store in task so that task can send updates to its listeners.
learningTask.BayesianNetwork = bn;
learningTask.LearningState = ComputationState.Computing;
// Build Dirichlet parameters
IDictionary<string, DirichletDistribution> priors = new Dictionary<string, DirichletDistribution>();
foreach (var variableName in variableNames)
{
var space = trainingSet.Variables.TryFind(variableName).Value;
DirichletDistribution prior;
if (options.DistributionDirichletAlpha > 0)
{
prior = new DirichletDistribution();
foreach (var value in space.Values)
{
prior.SetParameter(value, options.DistributionDirichletAlpha);
}
priors[variableName] = prior;
}
}
// Initialize a sufficient statistics.
SufficientStatistics sufficientStatistics
= new SufficientStatistics(trainingSet, Utils.Some(priors));
// Learn structure.
switch (options.Structure)
{
case LearningOptions.StructureEnum.DisconnectedStructure:
break;
case LearningOptions.StructureEnum.RandomStructure:
bn.GenerateStructure(
StructureClass.Random,
Utils.Some(options.StructureSeed),
Utils.Some(options.StructureParentLimit));
break;
case LearningOptions.StructureEnum.TreeStructure:
bn.LearnStructure(
sufficientStatistics,
StructureClass.Tree,
Utils.None<int>(),
Utils.None<int>());
break;
case LearningOptions.StructureEnum.GeneralStructure:
bn.LearnStructure(
sufficientStatistics,
StructureClass.General,
Utils.None<int>(),
Utils.Some(options.StructureParentLimit));
break;
}
// Learn distributions.
bn.LearnDistributions(sufficientStatistics);
// Done.
learningTask.LearningState = ComputationState.Done;
}
internal void RequestTraining(IObservationSet trainingSet, LearningOptions options)
{
if (trainingSet == null)
{
throw new ArgumentNullException();
}
ILearningTask learningTask
= new LearningTask(
Guid.NewGuid().ToString(),
trainingSet,
options);
learningTask.BayesianNetworkStarted += OnLearningStarted;
learningTask.BayesianNetworkFinished += OnLearningFinished;
Model.LearningTasks.Clear();
Model.LearningTasks.Add(learningTask);
xLearningInspector.SetIsLearning(true);
}
protected void DoLearning(LearningTask learningTask)
{
if (learningTask.LearningState == ComputationState.Done)
{
return;
}
// Grab options.
var options = learningTask.Options;
// Grab training set.
var trainingSet = learningTask.TrainingSet;
// Grab variable names.
var firstObservation = learningTask.TrainingSet.First();
var variableNames = firstObservation.VariableNames;
// Build a starter Bayesian network.
var bn = new BayesianNetwork(learningTask.TrainingSet.Name);
// Build variables for each variable in the training set.
foreach (var variableName in variableNames)
{
var space = trainingSet.Variables.TryFind(variableName).Value;
var rv = new RandomVariable(variableName, space);
bn.AddVariable(rv);
}
// Store in task so that task can send updates to its listeners.
learningTask.BayesianNetwork = bn;
learningTask.LearningState = ComputationState.Computing;
// Build Dirichlet parameters
IDictionary <string, DirichletDistribution> priors = new Dictionary <string, DirichletDistribution>();
foreach (var variableName in variableNames)
{
var space = trainingSet.Variables.TryFind(variableName).Value;
DirichletDistribution prior;
if (options.DistributionDirichletAlpha > 0)
{
prior = new DirichletDistribution();
foreach (var value in space.Values)
{
prior.SetParameter(value, options.DistributionDirichletAlpha);
}
priors[variableName] = prior;
}
}
// Initialize a sufficient statistics.
SufficientStatistics sufficientStatistics
= new SufficientStatistics(trainingSet, Utils.Some(priors));
// Learn structure.
switch (options.Structure)
{
case LearningOptions.StructureEnum.DisconnectedStructure:
break;
case LearningOptions.StructureEnum.RandomStructure:
bn.GenerateStructure(
StructureClass.Random,
Utils.Some(options.StructureSeed),
Utils.Some(options.StructureParentLimit));
break;
case LearningOptions.StructureEnum.TreeStructure:
bn.LearnStructure(
sufficientStatistics,
StructureClass.Tree,
Utils.None <int>(),
Utils.None <int>());
break;
case LearningOptions.StructureEnum.GeneralStructure:
bn.LearnStructure(
sufficientStatistics,
StructureClass.General,
Utils.None <int>(),
Utils.Some(options.StructureParentLimit));
break;
}
// Learn distributions.
bn.LearnDistributions(sufficientStatistics);
// Done.
learningTask.LearningState = ComputationState.Done;
}