private List <double> RunSingle(RqqPeltChangePointDetector detector, IReadOnlyList <CpdTestData> dataSet, bool printReports)
{
int maxNameLength = dataSet.Select(data => data.Name.Length).Max();
var penalties = new List <double>();
Parallel.ForEach(dataSet, testData =>
{
var changePointIndexes = detector.GetChangePointIndexes(testData.Values.ToArray());
var verification = CpdTestDataVerification.Verify(testData, changePointIndexes);
lock (detector)
{
penalties.Add(verification.Penalty);
PrintLine($"{testData.Name.PadRight(maxNameLength)} : {verification.Penalty}");
if (printReports)
{
PrintLine(verification.Report);
}
}
});
PrintLine(" Sum = " + penalties.Sum());
PrintLine(" Max = " + penalties.Max());
foreach (double p in new[] { 0.5, 0.90, 0.99 })
{
string metric = $"P{p * 100}".PadRight(4);
double estimate = HarrellDavisQuantileEstimator.Instance.GetQuantile(penalties, p);
PrintLine($" {metric} = {estimate:0.##}");
}
return(penalties);
}
public void Run(string[] args)
{
var stopwatch = Stopwatch.StartNew();
var fullDataSet = CpdReferenceDataSet.Generate(new Random(42), 2);
string dataSetArg = args.Length > 0 ? args[0] : "*";
bool printReports = args.Contains("--reports");
int limit = int.MaxValue;
int limitArgIndex = Array.IndexOf(args, "--limit");
if (limitArgIndex >= 0 && limitArgIndex < args.Length - 1)
{
if (int.TryParse(args[limitArgIndex + 1], out int actualLimit))
{
limit = actualLimit;
}
}
var dataSet = dataSetArg == "*" ? fullDataSet : fullDataSet.Where(data => data.Name.Contains(dataSetArg)).ToList();
if (limit < dataSet.Count)
{
new Shuffler(42).Shuffle(dataSet);
dataSet.RemoveRange(limit, dataSet.Count - limit);
}
if (dataSet.Count == 0)
{
PrintLine("DataSet is empty");
return;
}
dataSet.Sort((a, b) => string.Compare(a.Name, b.Name, StringComparison.Ordinal));
if (args.Contains("--tune"))
{
var heterogeneityFactors = new ArithmeticProgressionSequence(1.1, 0.05).GenerateArray(7);
var sensitivities = new ArithmeticProgressionSequence(0.4, 0.02).GenerateArray(8);
var quantileSets = new List <QuantileSet>
{
QuantileSet.Classic,
QuantileSet.ArithmeticProgression(12, 0),
QuantileSet.SteadyPlusArithmeticProgression(12, 7, -0.01),
QuantileSet.ArithmeticProgressionWithRepetitions(12, 4, -0.1)
};
int quantileSetMaxLength = quantileSets.Max(s => s.Name.Length);
var results =
new List <(double HeterogeneityFactor, double Sensitivity, string QuantileSet, double MaxPenalty, double SumPenalty)>();
foreach (double heterogeneityFactor in heterogeneityFactors)
{
foreach (double sensitivity in sensitivities)
{
foreach (var quantileSet in quantileSets)
{
double homogeneityFactor = heterogeneityFactor - 1;
PrintLine(Separator('@'));
PrintLine(
$"@ HeterogeneityFactor = {heterogeneityFactor:0.0}, Sensitivity = {sensitivity:0.00}, QuantileSet = {quantileSet.Name}");
var detector = new RqqPeltChangePointDetector(
quantileSet.Probabilities,
quantileSet.Factors,
sensitivity: sensitivity,
heterogeneityFactor: heterogeneityFactor,
homogeneityFactor: homogeneityFactor);
var penalties = RunSingle(detector, dataSet, printReports);
results.Add((heterogeneityFactor, sensitivity, quantileSet.Name, penalties.Max(), penalties.Sum()));
}
}
}
PrintLine(Separator('*'));
PrintLine(Separator('*'));
PrintLine(Separator('*'));
results.Sort((a, b) =>
Math.Sign(b.MaxPenalty.CompareTo(a.MaxPenalty)) * 10 + Math.Sign(b.SumPenalty.CompareTo(a.SumPenalty)));
foreach ((double heterogeneityFactor, double sensitivity, string quantileSet, double maxPenalty,
double sumPenalty) in results)
{
PrintLine(
$"{heterogeneityFactor:0.00} {sensitivity:0.00} {quantileSet.PadRight(quantileSetMaxLength)} : {maxPenalty} / {sumPenalty}");
}
}
else
{
RunSingle(RqqPeltChangePointDetector.Instance, dataSet, printReports);
}
stopwatch.Stop();
PrintLine();
PrintLine($"TotalTime = {stopwatch.Elapsed.TotalSeconds:0.0} sec");
}
