public static BPMDataModel CreateNoisyModel(BPMDataModel model, double epsilon)
{
Random generator = new Random();
BPMDataModel noisyModel = new BPMDataModel(model.GetInputs().Length, model.GetNoOfFeatures());
List <double> ranges = model.GetRanges();
double[] values = new double[model.GetInputs()[0].Count];
List <Vector> x = new List <Vector>();
foreach (Vector vector in model.GetInputs())
{
for (int i = 0; i < ranges.Count; i++)
{
double randomVal = (0.5 - generator.NextDouble());
double newVal = CreateLaplacian(epsilon, 0, vector[i], ranges[i], randomVal);
values[i] = newVal;
}
x.Add(Vector.FromArray(values));
}
noisyModel.SetAllVectorFeatures(x.ToArray());
noisyModel.SetClasses(model.GetClasses());
//noisyModel.CalculateFeatureLimits();
//noisyModel.ScaleFeatures();
return(noisyModel);
}
private static double RunBPMGeneral(BPMDataModel model, bool addBias, Vector[] testSet, int[] testResults)
{
int correctCount = 0;
VectorGaussian[] posteriorWeights = _bpm.Train(model.GetInputs(), model.GetClasses());
string actualWeights = posteriorWeights[1].ToString();
int breakLocation = actualWeights.IndexOf("\r", StringComparison.Ordinal);
actualWeights = actualWeights.Substring(0, breakLocation);
if (!_onlyWriteAggregateResults && _writeGaussians)
{
_writer.WriteLine("Weights= " + actualWeights);
}
Discrete[] predictions = _bpm.Test(testSet);
int i = 0;
foreach (Discrete prediction in predictions)
{
if (FindMaxValPosition(prediction.GetProbs().ToArray()) == testResults[i])
{
correctCount++;
}
i++;
}
double accuracy = ((double)correctCount / predictions.Length) * 100;
//double logEvidence = bpm.GetLogEvidence();
return(accuracy);
}
private void begin_Click(object sender, EventArgs e)
{
if (_maxSensitivity < _startSensitivity)
{
ShowDialog(@"The maximum sensitivity must be greater than or equal to the minimum sensitivity.",
"Error", false);
return;
}
if (!_performingCalcs)
{
//Disable input changes
ChangeStatusOfInputs(false);
beginButton.Text = @"Cancel";
_bpm = new BPM(_numOfClasses, _noisePrecision);
try
{
_trainingModel = FileUtils.ReadFile(_trainingFilePath, _labelAtStartOfLine, _noOfFeatures, _addBias);
_testModel = FileUtils.ReadFile(_testFilePath, _labelAtStartOfLine, _noOfFeatures, _addBias);
_writer = new StreamWriter(_resultsFilePath, _appendToFile);
_totalRuns = (int)Math.Ceiling(_noOfRuns * (1 + ((_maxSensitivity - _startSensitivity) / _sensitivityIncrement)));
_last = DateTime.Now;
bw.WorkerReportsProgress = true;
bw.WorkerSupportsCancellation = true;
bw.DoWork += bw_DoWork;
bw.ProgressChanged += bw_ProgressChanged;
bw.RunWorkerCompleted += bw_RunWorkerCompleted;
bw.RunWorkerAsync();
progressBar1.Maximum = _totalRuns;
_performingCalcs = true;
_prevRem = _totalRuns;
}
catch (Exception exception)
{
ShowDialog("Sorry, there was an error reading the input data" + exception.GetType(), "Error", true);
beginButton.Text = @"Begin processing";
ChangeStatusOfInputs(true);
}
}
else
{
bw.CancelAsync();
beginButton.Text = @"Begin processing";
//Tidy up
statusLabel.Text = @"";
}
}
public static BPMDataModel CreateNoisyModel(BPMDataModel model, double epsilon)
{
Random generator = new Random();
BPMDataModel noisyModel = new BPMDataModel(model.GetInputs().Length, model.GetNoOfFeatures());
List<double> ranges = model.GetRanges();
double[] values = new double[model.GetInputs()[0].Count];
List<Vector> x = new List<Vector>();
foreach (Vector vector in model.GetInputs())
{
for (int i = 0; i < ranges.Count; i++)
{
double randomVal = (0.5 - generator.NextDouble());
double newVal = CreateLaplacian(epsilon, 0, vector[i], ranges[i], randomVal);
values[i] = newVal;
}
x.Add(Vector.FromArray(values));
}
noisyModel.SetAllVectorFeatures(x.ToArray());
noisyModel.SetClasses(model.GetClasses());
//noisyModel.CalculateFeatureLimits();
//noisyModel.ScaleFeatures();
return noisyModel;
}
private static double RunBPMGeneral(BPMDataModel model, bool addBias, Vector[] testSet, int[] testResults)
{
int correctCount = 0;
VectorGaussian[] posteriorWeights = _bpm.Train(model.GetInputs(), model.GetClasses());
string actualWeights = posteriorWeights[1].ToString();
int breakLocation = actualWeights.IndexOf("\r", StringComparison.Ordinal);
actualWeights = actualWeights.Substring(0, breakLocation);
if (!_onlyWriteAggregateResults && _writeGaussians) _writer.WriteLine("Weights= " + actualWeights);
Discrete[] predictions = _bpm.Test(testSet);
int i = 0;
foreach (Discrete prediction in predictions)
{
if (FindMaxValPosition(prediction.GetProbs().ToArray()) == testResults[i]) correctCount++;
i++;
}
double accuracy = ((double)correctCount / predictions.Length) * 100;
//double logEvidence = bpm.GetLogEvidence();
return accuracy;
}
private void begin_Click(object sender, EventArgs e)
{
if (_maxSensitivity < _startSensitivity)
{
ShowDialog(@"The maximum sensitivity must be greater than or equal to the minimum sensitivity.",
"Error", false);
return;
}
if (!_performingCalcs)
{
//Disable input changes
ChangeStatusOfInputs(false);
beginButton.Text = @"Cancel";
_bpm = new BPM(_numOfClasses, _noisePrecision);
try
{
_trainingModel = FileUtils.ReadFile(_trainingFilePath, _labelAtStartOfLine, _noOfFeatures, _addBias);
_testModel = FileUtils.ReadFile(_testFilePath, _labelAtStartOfLine, _noOfFeatures, _addBias);
_writer = new StreamWriter(_resultsFilePath, _appendToFile);
_totalRuns = (int) Math.Ceiling(_noOfRuns*(1 + ((_maxSensitivity - _startSensitivity)/_sensitivityIncrement)));
_last = DateTime.Now;
bw.WorkerReportsProgress = true;
bw.WorkerSupportsCancellation = true;
bw.DoWork += bw_DoWork;
bw.ProgressChanged += bw_ProgressChanged;
bw.RunWorkerCompleted += bw_RunWorkerCompleted;
bw.RunWorkerAsync();
progressBar1.Maximum = _totalRuns;
_performingCalcs = true;
_prevRem = _totalRuns;
}
catch (Exception exception)
{
ShowDialog("Sorry, there was an error reading the input data" + exception.GetType(), "Error", true);
beginButton.Text = @"Begin processing";
ChangeStatusOfInputs(true);
}
}
else
{
bw.CancelAsync();
beginButton.Text = @"Begin processing";
//Tidy up
statusLabel.Text = @"";
}
}
public static BPMDataModel ReadFile(string filename, bool labelAtStart, int noOfFeatures, bool addBias)
{
int noOfInputs = 0;
StreamReader temp = new StreamReader(filename);
while (!temp.EndOfStream)
{
temp.ReadLine();
noOfInputs++;
}
temp.Dispose();
BPMDataModel newModel = new BPMDataModel(noOfInputs, noOfFeatures);
List <double[]> featureData = new List <double[]>();
//Holds the upper and lower limits for each feature being analysed
List <double[]> limits = new List <double[]>(noOfFeatures);
for (int i = 0; i < noOfFeatures; i++)
{
double[] lim = new double[2];
lim[0] = Double.NaN;
lim[1] = Double.NaN;
limits.Add(lim);
}
//Initialise feature data list
for (int i = 0; i < noOfFeatures; i++)
{
featureData.Add(new double[noOfInputs]);
}
//The variables to hold the data being read in
int[] classData = new int[noOfInputs];
List <Vector> x = new List <Vector>();
Vector[] featureVectors;
string line;
double[] values;
int index;
int labelIndex;
int currentLocation = 0;
using (StreamReader reader = new StreamReader(filename))
{
while (!reader.EndOfStream)
{
line = reader.ReadLine();
//Allow for splitting of data in file by either tab, space or comma
if (line != null)
{
string[] pieces = line.Split('\t', ' ', ',');
//This assumes class identifier AND ID no on each line
int n = pieces.Length - 2;
if (addBias)
{
values = new double[n + 1];
values[n] = 1;
}
else
{
values = new double[n];
}
// Read feature values and labels.
labelIndex = labelAtStart ? 0 : n + 1;
for (int i = 0; i < noOfFeatures; i++)
{
index = labelAtStart ? i + 1 : i;
values[i] = Double.Parse(pieces[index]);
featureData[i][currentLocation] = Double.Parse(pieces[index]);
if ((limits[i][0].Equals(Double.NaN)) || (values[i] < limits[i][0]))
{
limits[i][0] = values[i];
}
if ((limits[i][1].Equals(Double.NaN)) || (values[i] > limits[i][1]))
{
limits[i][1] = values[i];
}
}
classData[currentLocation] = Int32.Parse(pieces[labelIndex - 1]);
x.Add(Vector.FromArray(values));
}
currentLocation++;
}
//Clean up resources
reader.Dispose();
}
featureVectors = x.ToArray();
newModel.SetAllVectorFeatures(featureVectors);
newModel.SetClasses(classData);
newModel.SetInputLimits(limits);
return(newModel);
}
public static BPMDataModel ReadFile(string filename, bool labelAtStart, int noOfFeatures, bool addBias)
{
int noOfInputs = 0;
StreamReader temp = new StreamReader(filename);
while (!temp.EndOfStream)
{
temp.ReadLine();
noOfInputs++;
}
temp.Dispose();
BPMDataModel newModel = new BPMDataModel(noOfInputs, noOfFeatures);
List<double[]> featureData = new List<double[]>();
//Holds the upper and lower limits for each feature being analysed
List<double[]> limits = new List<double[]>(noOfFeatures);
for (int i = 0; i < noOfFeatures; i++)
{
double[] lim = new double[2];
lim[0] = Double.NaN;
lim[1] = Double.NaN;
limits.Add(lim);
}
//Initialise feature data list
for (int i = 0; i < noOfFeatures; i++)
{
featureData.Add(new double[noOfInputs]);
}
//The variables to hold the data being read in
int[] classData = new int[noOfInputs];
List<Vector> x = new List<Vector>();
Vector[] featureVectors;
string line;
double[] values;
int index;
int labelIndex;
int currentLocation = 0;
using (StreamReader reader = new StreamReader(filename))
{
while (!reader.EndOfStream)
{
line = reader.ReadLine();
//Allow for splitting of data in file by either tab, space or comma
if (line != null)
{
string[] pieces = line.Split('\t', ' ', ',');
//This assumes class identifier AND ID no on each line
int n = pieces.Length - 2;
if (addBias)
{
values = new double[n + 1];
values[n] = 1;
}
else
{
values = new double[n];
}
// Read feature values and labels.
labelIndex = labelAtStart ? 0 : n + 1;
for (int i = 0; i < noOfFeatures; i++)
{
index = labelAtStart ? i + 1 : i;
values[i] = Double.Parse(pieces[index]);
featureData[i][currentLocation] = Double.Parse(pieces[index]);
if ((limits[i][0].Equals(Double.NaN)) || (values[i] < limits[i][0]))
{
limits[i][0] = values[i];
}
if ((limits[i][1].Equals(Double.NaN)) || (values[i] > limits[i][1]))
{
limits[i][1] = values[i];
}
}
classData[currentLocation] = Int32.Parse(pieces[labelIndex - 1]);
x.Add(Vector.FromArray(values));
}
currentLocation++;
}
//Clean up resources
reader.Dispose();
}
featureVectors = x.ToArray();
newModel.SetAllVectorFeatures(featureVectors);
newModel.SetClasses(classData);
newModel.SetInputLimits(limits);
return newModel;
}
