private void ParametersetSGB_Click(object sender, RoutedEventArgs e)
{
var request = new UserDecision();
request.Options = new string[] { "A", "B", "C" };
request.Request();
//CRFToolData.IsingData = new SoftwareGraphIsing();
//var window = new MainView();
//window.Show();
}
public void Execute()
{
// - Prerequisites:
// - Graphstructure
// - Training Data
// - 2 Node Classifications
TrainingData = new List <GWGraph <CRFNodeData, CRFEdgeData, CRFGraphData> >();
EvaluationData = new List <GWGraph <CRFNodeData, CRFEdgeData, CRFGraphData> >();
// decision wether user wants to train or load pre-trained data
var requestTraining = new UserDecision("Use Training.", "Load Training Result.");
requestTraining.Request();
if (requestTraining.Decision == 0) // use training
{
// -n characteristics for each node
{
var request = new UserInput(UserInputLookFor.Folder);
request.DefaultPath = "..\\..\\CRFToolApp\\bin\\Graphs";
request.TextForUser = "******";
request.Request();
GraphDataFolder = request.UserText;
foreach (var file in Directory.EnumerateFiles(GraphDataFolder))
{
var graph = JSONX.LoadFromJSON <GWGraph <CRFNodeData, CRFEdgeData, CRFGraphData> >(file);
TrainingData.Add(graph);
}
}
// - Step 1:
// - discretize characteristics
#region Use Training Pede
{
// create features
CreateFeatures(Dataset, TrainingData);
InitCRFScores(TrainingData);
var request = new OLMRequest(OLMVariant.Ising, TrainingData);
request.BasisMerkmale.AddRange(Dataset.NodeFeatures);
request.BasisMerkmale.AddRange(Dataset.EdgeFeatures);
request.LossFunctionValidation = OLM.LossRatio;
request.Request();
// zugehörige Scores erzeugen für jeden Graphen (auch Evaluation)
CreateCRFScores(TrainingData, Dataset.NodeFeatures, request.Result.ResultingWeights);
// store trained Weights
Dataset.NumberIntervals = NumberIntervals;
Dataset.Characteristics = TrainingData.First().Data.Characteristics.ToArray();
Dataset.EdgeCharacteristic = "IsingEdgeCharacteristic";
Dataset.Weights = request.Result.ResultingWeights;
Dataset.SaveAsJSON("results.json");
}
#endregion
}
else
{ // load pre-trained data
var request = new UserInput();
request.TextForUser = "******";
request.Request();
var file = request.UserText;
var trainingResult = JSONX.LoadFromJSON <WorkflowOneDataset>(file);
Dataset = trainingResult;
}
//- Step2:
// User Choice here
{
var request = new UserInput(UserInputLookFor.Folder);
request.TextForUser = "******";
request.Request();
GraphDataFolder = request.UserText;
foreach (var file in Directory.EnumerateFiles(GraphDataFolder))
{
var graph = JSONX.LoadFromJSON <GWGraph <CRFNodeData, CRFEdgeData, CRFGraphData> >(file);
EvaluationData.Add(graph);
}
}
// remove double edges
{
var edgesToRemove = new LinkedList <GWEdge <CRFNodeData, CRFEdgeData, CRFGraphData> >();
foreach (var graph in EvaluationData)
{
foreach (var edge in graph.Edges.ToList())
{
if (graph.Edges.Any(e => (e != edge) && ((e.Foot == edge.Head && e.Head == edge.Foot && e.GWId.CompareTo(edge.GWId) < 0) || (e.Foot == edge.Foot && e.Head == edge.Head && e.GWId.CompareTo(edge.GWId) < 0))))
{
edgesToRemove.Add(edge);
graph.Edges.Remove(edge);
}
}
}
foreach (var edge in edgesToRemove)
{
edge.Foot.Edges.Remove(edge);
edge.Head.Edges.Remove(edge);
}
}
//scores erzeugen
CreateCRFScores(EvaluationData, Dataset.NodeFeatures, Dataset.Weights);
// - Create ROC Curve
{
}
// - Give Maximum with Viterbi
{
foreach (var graph in EvaluationData)
{
var request = new SolveInference(graph, null, 2);
request.Request();
graph.Data.AssginedLabeling = request.Solution.Labeling;
}
//show results in 3D Viewer
{
//var request = new ShowGraphs();
//request.Graphs = EvaluationData;
//request.Request();
}
}
// - Give Sample with MCMC
{
foreach (var graph in EvaluationData)
{
SoftwareGraphLearningParameters parameters = new SoftwareGraphLearningParameters();
parameters.NumberOfGraphs = 60;
parameters.NumberNodes = 50;
parameters.NumberLabels = 2;
parameters.NumberCategories = 4;
parameters.IntraConnectivityDegree = 0.15;
parameters.InterConnectivityDegree = 0.01;
//sample parameters
var samplerParameters = new MHSamplerParameters();
var sglGraph = graph.Convert((nodeData) => new SGLNodeData()
{
}, (edgeData) => new SGLEdgeData(), (graphData) => new SGLGraphData());
samplerParameters.Graph = sglGraph;
samplerParameters.NumberChains = 1;
//sampler starten
var gibbsSampler = new MHSampler();
gibbsSampler.Do(samplerParameters);
}
}
}