private static void StartTrainingCycle()
{
// erzeugt die RequestListener
CRFToolApp.Build.Do();
CRFBase.Build.Do();
CRFGraphVis.Build.Do();
// starting of TrainingCycle:
var trainingCycle = new TrainingEvaluationCycleZellner();
var parameters = new TrainingEvaluationCycleInputParameters();
// take OLM variants we want to test
List <OLMVariant> variants = new List <OLMVariant>
{
//OLMVariant.Ising
//OLMVariant.IsingII
OLMVariant.IsingIII
//OLMVariant.Default
};
Log.Post("Training variant: " + variants[0]);
// setting of transition probabilities to create observation from reference labeling
double[,] transition = SetTransitionProbabilities();
#region modify graphs
// do this for all graphs (currently saved in form of pdbfiles)
List <GWGraph <CRFNodeData, CRFEdgeData, CRFGraphData> > crfGraphList = new List <GWGraph <CRFNodeData, CRFEdgeData, CRFGraphData> >();
var id = 0;
foreach (String file in File.ReadLines(fileNames))
{
RandomlySelectedPDBFile = fileFolder + "/" + file;
RandomlySelectedPDBFileName = file.Substring(0, 4);
var pdbFile = PDBExt.Parse(RandomlySelectedPDBFile);
pdbFile.Name = RandomlySelectedPDBFileName;
// set IsCore nodes, cuz we only need the interface nodes (the core nodes cannot interact)
var proteinGraph = SetIsCore(pdbFile);
// trim graph -> remove isCore nodes
var trimmedGraph = TrimGraph(pdbFile, proteinGraph);
// set real reference label of the graph
var crfGraph = SetReferenceLabel(trimmedGraph);
SetEdgeMaxDiffValues(crfGraph);
crfGraph.Id = id++;
crfGraphList.Add(crfGraph);
}
#endregion
// set parameters for the training cycle
parameters = new TrainingEvaluationCycleInputParameters(crfGraphList, crfGraphList.Count, variants, IsingConformityParameter,
PottsConformityParameters, IsingCorrelationParameter, PottsCorrelationParameters, NumberOfIntervals,
transition, NumberOfLabels, BufferSizeViterbi, amplifierControlParameter, Threshold);
// running the cycle
trainingCycle.RunCycle(parameters);
}
private static void startTestCycle()
{
// erzeugt die RequestListener
Build.Do();
// einfache Graphen erzeugen (12 clone)
//var graph = CreateGraphs.createGraph_Grid2D();
var pdbFile = PDBExt.Parse(RandomlySelectedPDBFile);
var proteinGraph = CreateProteinGraph.Do(pdbFile, 6.0, ResidueNodeCenterDefinition.CAlpha).Values.First();
IGWGraph <ICRFNodeData, ICRFEdgeData, ICRFGraphData> crfGraph = proteinGraph.Convert(nd => new CRFNodeData(nd.Data.Residue.Id), ed => new CRFEdgeData(), gd => new CRFGraphData());
// seeding
// 12 clone erzeugen
var clone = new GWGraph <ICRFNodeData, ICRFEdgeData, ICRFGraphData> [clones];
for (int i = 0; i < clones; i++)
{
var cloneGraph = crfGraph.Clone(ConvertData.convertNodeData, ConvertData.convertEdgeData, ConvertData.convertGraphData);
// für jeden graphen (clone): Erzeugung eines Labelings (rdm) => als int[]
Labeling.randomLabeling(cloneGraph);
Labeling.assignedLabeling(cloneGraph);
clone[i] = cloneGraph;
}
// calculation of tp, fp, tn, fn, sensitivity, specificity, mcc
double[][] keysForAllClones = new double[clones][];
for (int i = 0; i < clones; i++)
{
double[] keys = Compute.computeKeys(crfGraph, clone, i);
keysForAllClones[i] = keys;
}
// test
//outputKeys(keysForAllClones);
outputLabelingsScores(clone);
}
private static void startTrainingCycle()
{
// erzeugt die RequestListener
Build.Do();
// calculation of the termination condition for the seeding method
double[] averageRatioItoS = calculateTerminationConditionForSeeding();
// starting of TrainingCycle:
var trainingCycle = new TrainingEvaluationCycle();
var parameters = new TrainingEvaluationCycleInputParameters();
// take OLM variants we want to test, ISING and OLM_III (Default)
List <OLMVariant> variants = new List <OLMVariant>();
variants.Add(OLMVariant.Ising);
variants.Add(OLMVariant.Default);
// setting of transition probabilities
//var a = 0.5 + 0.15 * rdm.NextDouble();
var a = 0.85;
var b = 0.85;
double[,] transition = new double[2, 2] {
{ a, 1 - a }, { 1 - b, b }
};
#region modify graph
var pdbFile = PDBExt.Parse(RandomlySelectedPDBFile);
pdbFile.Name = RandomlySelectedPDBFileName;
// setzen der IsCore
var rasaRequest = new RequestRasa(pdbFile);
rasaRequest.RequestInDefaultContext();
foreach (var rasa in rasaRequest.Rasavalues)
{
rasa.Key.IsCore = rasa.Value <= 0.15;
}
var proteinGraph = CreateProteinGraph.Do(pdbFile, 7.0, ResidueNodeCenterDefinition.CAlpha).Values.First();
// graph trimmen
var trimmedGraph = trimGraph(pdbFile, proteinGraph);
// set real reference label of the graph
var nameWithChain = RandomlySelectedPDBFile.Substring(fileFolder.Length + 1, 6);
var interfacesList = new List <string>();
using (var reader = new StreamReader(InterfaceDefLocation))
{
var line = "";
while ((line = reader.ReadLine()) != null)
{
var nuss = nameWithChain;
if (line.StartsWith(nameWithChain))
{
interfacesList.Add(line.Substring(8));
}
}
}
foreach (var interfaceEntry in interfacesList)
{
var node = trimmedGraph.Nodes.FirstOrDefault(n => n.Data.Residue.Id.Equals(interfaceEntry));
if (node != null)
{
node.Data.ReferenceLabel = 1;
}
}
#endregion
//test
if (GraphVisualization)
{
var graph3D = trimmedGraph.Wrap3D(nd => new Node3DWrap <ResidueNodeData>(nd.Data)
{
ReferenceLabel = nd.Data.ReferenceLabel, X = nd.Data.X, Y = nd.Data.Y, Z = nd.Data.Z
});
new ShowGraph3D(graph3D).Request(RequestRunType.Background);
Thread.Sleep(120000);
}
var crfGraph = trimmedGraph.Convert <ResidueNodeData, CRFNodeData, SimpleEdgeData, CRFEdgeData,
ProteinGraphData, CRFGraphData>(nd => new CRFNodeData(nd.Data.Residue.Id)
{
X = nd.Data.X, Y = nd.Data.Y, Z = nd.Data.Z
}, ed => new CRFEdgeData(),
gd => new CRFGraphData());
crfGraph.SaveAsJSON("testGraph.txt");
// setting the MaximumTotalPatchSize based on the ratio calculated with the hermann data
int min = (int)(crfGraph.NodeCounter * averageRatioItoS[1]);
int max = (int)(crfGraph.NodeCounter * averageRatioItoS[2]);
int av = (int)(crfGraph.NodeCounter * averageRatioItoS[0]);
int MaximumTotalPatchSize = rdm.Next(min, max);
// set parameters for the training cycle
parameters = new TrainingEvaluationCycleInputParameters(crfGraph, NumberOfGraphInstances,
NumberOfSeedsForPatchCreation, MaximumTotalPatchSize, variants,
IsingConformityParameter, IsingCorrelationParameter, transition, NumberOfLabels, BufferSizeViterbi);
// running the cycle
trainingCycle.RunCycle(parameters);
}
private static double[] calculateTerminationConditionForSeeding()
{
double[] averageRatioItoS = new double[3];
if (ComputeTerminationCondition)
{
List <double> ratios = new List <double>();
// Getting the ratio of all surface nodes to the interface nodes
// to get a reality near parameter when to stop the seeding
foreach (var file in Directory.GetFiles(fileFolder))
{
var pdbFile = PDBExt.Parse(file);
pdbFile.Name = file.Substring(fileFolder.Length + 1, 4);
var nameWithChain = file.Substring(fileFolder.Length + 1, 6);
// counting all the nodes
var rasaRequest = new RequestRasa(pdbFile);
rasaRequest.RequestInDefaultContext();
// counting all the surface nodes
int SurfaceCounter = 0;
foreach (var rasa in rasaRequest.Rasavalues)
{
SurfaceCounter += rasa.Value > 0.15 ? 1 : 0;
rasa.Key.IsCore = rasa.Value <= 0.15;
}
// counting the interface nodes for each file
int InterfaceCounter = 0;
using (var reader = new StreamReader(InterfaceDefLocation))
{
var line = "";
while ((line = reader.ReadLine()) != null)
{
var nuss = nameWithChain;
InterfaceCounter += line.StartsWith(nameWithChain) ? 1 : 0;
}
}
double ratio = 1.0 * InterfaceCounter / SurfaceCounter;
ratios.Add(ratio);
}
// calculation of the average ratio between surface and interface nodes
double ratioSum = 0.0;
for (int i = 0; i < ratios.Count; i++)
{
ratioSum += ratios[i];
}
double max = ratios.Max();
double min = ratios.Min();
averageRatioItoS[0] = ratioSum / ratios.Count;
averageRatioItoS[1] = min;
averageRatioItoS[2] = max;
averageRatioItoS.SaveAsJSON("terminationCondition.txt");
}
else
{
averageRatioItoS = JSONX.LoadFromJSON <double[]>("terminationCondition.txt");
}
return(averageRatioItoS);
}
