/// <summary>
/// Creates a new instance of the learning algorithm for a given
/// Markov sequence classifier using the specified configuration
/// function.
/// </summary>
///
public SequenceClassifierLearning(SequenceClassifier <TDistribution> classifier,
ClassifierLearningAlgorithmConfiguration algorithm)
: base(classifier, algorithm)
{
}
private void loadAllModels(string modelDirectoryName)
{
if (Directory.Exists(modelDirectoryName))
{
// Get all files of type .rec in the directory
string[] files = Directory.GetFiles(modelDirectoryName, "*.dat");
this.positionModels = new List<HiddenMarkovModel<MultivariateNormalDistribution>>();
this.jointModels = new List<HiddenMarkovModel<MultivariateNormalDistribution>>();
this.positionClasses = new List<string>();
this.jointClasses = new List<string>();
foreach (string s in files)
{
string actName = Path.GetFileNameWithoutExtension(s);
// strip off the hmm_ from the beginning
actName = actName.Substring(4);
bool isJointAction = Util.shouldUseJVals(actName);
// Deserialize the model and add it to the correct classifier
SerializableHmm serMod = new SerializableHmm(actName, modelDirectoryName);
HiddenMarkovModel<MultivariateNormalDistribution> hmm = serMod.LoadFromDisk();
if (isJointAction)
{
jointModels.Add(hmm);
jointClasses.Add(actName);
}
else
{
positionModels.Add(hmm);
positionClasses.Add(actName);
}
}
jointActionClassifier = new SequenceClassifier<MultivariateNormalDistribution>(jointModels.ToArray());
positionActionClassifier = new SequenceClassifier<MultivariateNormalDistribution>(positionModels.ToArray());
// TODO(pbrook) if there is time we should add a threshold model to these SequenceClassifiers
}
}