private void RefreshCumulativeProbabilityChart()
{
// Get the cumulative probabilities for marked and unmarked batches and add them to the graph
InitializeCumulativeProbabilityDistributionChart();
//// Format the chart
////standardSeries.ChartType = SeriesChartType.RangeColumn;
////standardSeries.BorderWidth = 1;
////standardSeries.BorderDashStyle = ChartDashStyle.Solid;
////standardSeries.BorderColor = Color.Black;
////standardSeries.Color = Color.Blue;
//CdfChart.Series.Clear();
//CdfChart.Titles.Clear();
//CdfChart.Titles.Add("Cumulative Probability Distribution");
////CdfChart.Legends[0].Position.Auto = true; //ElementPosition
//CdfChart.Legends[0].IsDockedInsideChartArea = true;
//CdfChart.Legends[0].Docking = Docking.Bottom;
//CdfChart.Legends[0].Alignment = StringAlignment.Center;
//CdfChart.ChartAreas[0].AxisX.Title = "Packets per Interval";
//CdfChart.ChartAreas[0].AxisX.Minimum = 0;
////CumulativeChart.ChartAreas[0].AxisX.Maximum =
//// Get the type of chart to display
//string chartType = ChartTypeComboBox.Items[ChartTypeComboBox.SelectedIndex].ToString();
//// Marked probabilities series
//CdfChart.Series.Add("MarkedProbabilities");
//CdfChart.Series["MarkedProbabilities"].ChartType = SeriesChartType.Line;
////CdfChart.Series["MarkedProbabilities"].ChartType = chartType == "Bar" ? SeriesChartType.Column : SeriesChartType.Line;
//CdfChart.Series["MarkedProbabilities"].IsVisibleInLegend = true;
//CdfChart.Series["MarkedProbabilities"].LegendText = "Marked";
//// Unmarked probabilities series
//CdfChart.Series.Add("UnmarkedProbabilities");
//CdfChart.Series["UnmarkedProbabilities"].ChartType = SeriesChartType.Line;
////CdfChart.Series["UnmarkedProbabilities"].ChartType = chartType == "Bar" ? SeriesChartType.Column : SeriesChartType.Line;
//CdfChart.Series["UnmarkedProbabilities"].IsVisibleInLegend = true;
//CdfChart.Series["UnmarkedProbabilities"].LegendText = "Unmarked";
//// Get the batch intervals
//BindingList<BatchIntervalMarked> unmarkedBatchIntervals = new BindingList<BatchIntervalMarked>();
//BindingList<BatchIntervalMarked> markedBatchIntervals = new BindingList<BatchIntervalMarked>();
////int histogramBinSize = Convert.ToInt32(HistogramBinSizeTextBox.Text);
//SortedDictionary<int, decimal> markedProbabilities = new CalculateProbability(markedBatchIntervals).GetCumulativeProbabilityDistribution(_CumulativeMarkedProbabilities);
//SortedDictionary<int, decimal> unmarkedProbabilities = new CalculateProbability(unmarkedBatchIntervals).GetCumulativeProbabilityDistribution(_CumulativeUnmarkedProbabilities);
// Retrieve the cumulative histogram probabilities from the database
ProcessCapturePackets pcp = new ProcessCapturePackets();
BindingList<CumulativeProbabilityDistribution> markedCumulativeProbabilityDistributionList = new BindingList<CumulativeProbabilityDistribution>();
markedCumulativeProbabilityDistributionList = pcp.GetCumulativeProbabilityDistributionData(CaptureState.Marked);
SortedDictionary<int, decimal> markedCumulativeProbabilityDistribution = new SortedDictionary<int, decimal>();
foreach (CumulativeProbabilityDistribution cpd in markedCumulativeProbabilityDistributionList)
{
markedCumulativeProbabilityDistribution.Add(cpd.Interval, cpd.Probability);
}
CdfChart.Series["MarkedProbabilities"].Color = Color.CornflowerBlue;
foreach (KeyValuePair<int, decimal> pair in markedCumulativeProbabilityDistribution)
{
CdfChart.Series["MarkedProbabilities"].Points.AddXY(Convert.ToDouble(pair.Key), Convert.ToDouble(pair.Value));
}
BindingList<CumulativeProbabilityDistribution> unmarkedCumulativeProbabilityDistributionList = new BindingList<CumulativeProbabilityDistribution>();
unmarkedCumulativeProbabilityDistributionList = pcp.GetCumulativeProbabilityDistributionData(CaptureState.Unmarked);
SortedDictionary<int, decimal> unmarkedCumulativeProbabilityDistribution = new SortedDictionary<int, decimal>();
foreach (CumulativeProbabilityDistribution cpd in unmarkedCumulativeProbabilityDistributionList)
{
unmarkedCumulativeProbabilityDistribution.Add(cpd.Interval, cpd.Probability);
}
CdfChart.Series["UnmarkedProbabilities"].Color = Color.Red;
foreach (KeyValuePair<int, decimal> pair in unmarkedCumulativeProbabilityDistribution)
{
CdfChart.Series["UnmarkedProbabilities"].Points.AddXY(Convert.ToDouble(pair.Key), Convert.ToDouble(pair.Value));
}
}
public HypothesisTest GetHypothesisTestResult()
{
HypothesisTest ht = new HypothesisTest(); ;
// Get cumulative probability distribution data and find the max difference between marked and unmarked distributions
ProcessCapturePackets pcp = new ProcessCapturePackets();
BindingList<CumulativeProbabilityDistribution> markedCPD = new BindingList<CumulativeProbabilityDistribution>();
BindingList<CumulativeProbabilityDistribution> unmarkedCPD = new BindingList<CumulativeProbabilityDistribution>();
markedCPD = pcp.GetCumulativeProbabilityDistributionData(CaptureState.Marked);
unmarkedCPD = pcp.GetCumulativeProbabilityDistributionData(CaptureState.Unmarked);
if (markedCPD.Count > 0 && unmarkedCPD.Count > 0)
{
decimal maxVariance = 0M;
int intervalCount = 0;
// Only compare intervals from each distribution with a corresponding interval in the other distribution
if (unmarkedCPD.Count > markedCPD.Count)
{
intervalCount = markedCPD.Count;
}
else
{
intervalCount = unmarkedCPD.Count;
}
// Expand each distribution into equal discrete steps for comparison of cumulative probabilities
// First, find the largest cumulative packet count (= interval)
int maxPacketCount = 0;
if (markedCPD[markedCPD.Count - 1].Interval >= unmarkedCPD[unmarkedCPD.Count - 1].Interval)
{
maxPacketCount = markedCPD[markedCPD.Count - 1].Interval;
}
else
{
maxPacketCount = unmarkedCPD[unmarkedCPD.Count - 1].Interval;
}
// Second, expand the packet counts by interpolating between packet counts (intervals) using an average probability
// for each packet count in the range and successively adding up to the next packet count (interval); add these
// interpolated packets to a dictionary; outcome is a dictionary for each distribution containing packet counts and
// probabilities from packet count = 0 to packet count = largest packet count (interval) of both distributions and
// the associated probabilities for each packet count. We are basically calculating a linear estimate of packet
// counts and probabilities between each packet count and probability in the actual distributions.
//// Third, check for packet counts that are less than the maximum packet count and assign a probability of 1
//// to any that are found
ExpandPacketCountLinear markedExpPktCount = new ExpandPacketCountLinear(markedCPD, maxPacketCount);
ExpandPacketCountLinear unmarkedExpPktCount = new ExpandPacketCountLinear(unmarkedCPD, maxPacketCount);
SortedDictionary<int, decimal> markedCPDExpanded = new SortedDictionary<int, decimal>();
SortedDictionary<int, decimal> unmarkedCPDExpanded = new SortedDictionary<int, decimal>();
//markedCPDExpanded = ExpandPacketCount(markedCPD, maxPacketCount);
//unmarkedCPDExpanded = ExpandPacketCount(unmarkedCPD, maxPacketCount);
markedCPDExpanded = markedExpPktCount.ExpandPacketCount();
unmarkedCPDExpanded = unmarkedExpPktCount.ExpandPacketCount();
// Find the maximum variance between the cumulative probabilities in each distribution
for (int i = 0; i < maxPacketCount; i++)
{
#region Debug
#if(DEBUG)
System.Diagnostics.Debug.WriteLine("unmarkedCPDExpanded[{0}]:[{1}] - markedCPDExpanded[{2}]:[{3}] = {4}", i, unmarkedCPDExpanded[i], i, markedCPDExpanded[i], Math.Abs(unmarkedCPDExpanded[i] - markedCPDExpanded[i]));
#endif
#endregion
if (Math.Abs(unmarkedCPDExpanded[i] - markedCPDExpanded[i]) > maxVariance)
{
maxVariance = Math.Abs(unmarkedCPDExpanded[i] - markedCPDExpanded[i]);
}
}
// Multiply by the square root of the sample size factor
maxVariance = maxVariance * Convert.ToDecimal(Math.Sqrt((markedCPD.Count * unmarkedCPD.Count) / (markedCPD.Count + unmarkedCPD.Count)));
// Compare the maximum variance with the hypothesis test threshold
// For significance level alpha = 0.05, the K-S statistic is computed as 1.36/N^(1/2), where N is the number of samples
decimal ksStatistic = Convert.ToDecimal(1.36 / Math.Pow(intervalCount, 0.5));
ht.KsStatistic = ksStatistic;
ht.MaxCpdVariance = maxVariance;
if (maxVariance > ksStatistic)
{
// Reject the null hypothesis
ht.KsTestResult = true;
}
}
else
{
// Not enough data to perform the test
ht.KsStatistic = 0;
ht.MaxCpdVariance = 0;
ht.KsTestResult = false;
}
return ht;
}
