public void OnDataChanged()
{
if (DrawCost && DrawDistance)
{
throw new Exception("Only one of the values can be drawn at once, 'cost' or 'distance'.");
}
double[][] matrixValues = null;
if (DrawCost)
{
matrixValues = Dtw.GetCostMatrix();
}
if (DrawDistance)
{
matrixValues = Dtw.GetDistanceMatrix();
}
var dtwPath = Dtw.GetPath();
var xLength = Dtw.XLength;
var yLength = Dtw.YLength;
var cost = Dtw.GetCost();
var costNormalized = Dtw.GetCost() / Math.Sqrt(xLength * xLength + yLength * yLength);
var plotModel = new PlotModel(String.Format("Dtw norm by length: {0:0.00}, total: {1:0.00}", costNormalized, cost))
{
LegendTextColor = DrawCost || DrawDistance ? OxyColors.White : OxyColors.Black,
};
if (matrixValues != null)
{
var maxMatrixValue = 0.0;
for (int i = 0; i < xLength; i++)
{
for (int j = 0; j < yLength; j++)
{
maxMatrixValue = Math.Max(maxMatrixValue, Double.IsPositiveInfinity(matrixValues[i][j]) ? 0 : matrixValues[i][j]);
}
}
for (int i = 0; i < xLength; i++)
{
for (int j = 0; j < yLength; j++)
{
var value = matrixValues[i][j];
var isValuePositiveInfinity = Double.IsPositiveInfinity(value);
var intensityBytes = isValuePositiveInfinity ? new byte[] { 0, 0, 0 } : GetFauxColourRgbIntensity(value, 0, maxMatrixValue);
//var intensityByte = (byte)(255 - Math.Floor(255 * intensity));
plotModel.Annotations.Add(new PolygonAnnotation
{
Points =
new[]
{
new DataPoint(i - 0.5, j - 0.5), new DataPoint(i + 0.5, j - 0.5),
new DataPoint(i + 0.5, j + 0.5), new DataPoint(i - 0.5, j + 0.5),
},
StrokeThickness = 0,
Selectable = false,
Layer = AnnotationLayer.BelowAxes,
Fill = OxyColor.FromArgb(255, intensityBytes[0], intensityBytes[1], intensityBytes[2]),
});
}
}
for (int i = 0; i < 30; i++)
{
var intensityBytes = GetFauxColourRgbIntensity(i, 0, 29);
plotModel.Annotations.Add(new RectangleAnnotation
{
MinimumX = -39,
MaximumX = -25,
MinimumY = -i - 6,
MaximumY = -i - 5,
Selectable = false,
Fill = OxyColor.FromArgb(255, intensityBytes[0], intensityBytes[1], intensityBytes[2])
});
}
plotModel.Annotations.Add(new TextAnnotation
{
Position = new DataPoint(-24, -5),
HorizontalAlignment = HorizontalTextAlign.Left,
VerticalAlignment = VerticalTextAlign.Middle,
StrokeThickness = 0,
Text = "0"
});
plotModel.Annotations.Add(new TextAnnotation
{
Position = new DataPoint(-24, -34),
HorizontalAlignment = HorizontalTextAlign.Left,
VerticalAlignment = VerticalTextAlign.Middle,
StrokeThickness = 0,
Text = String.Format("{0:0.00}", maxMatrixValue),
});
}
var matrixPathSeries = new LineSeries("Path")
{
StrokeThickness = 1,
Color = OxyColors.Red,
};
for (int i = 0; i < dtwPath.Length; i++)
{
matrixPathSeries.Points.Add(new DataPoint(dtwPath[i].Item1, dtwPath[i].Item2));
}
plotModel.Series.Add(matrixPathSeries);
var seriesMatrixScale = (xLength + yLength) * 0.05;
for (int variableIndex = 0; variableIndex < Dtw.SeriesVariables.Length; variableIndex++)
{
var variableA = Dtw.SeriesVariables[variableIndex];
var variableASeries = variableA.OriginalXSeries;
var variableB = Dtw.SeriesVariables[variableIndex];
var variableBSeries = variableB.OriginalYSeries;
var minSeriesA = variableASeries.Min();
var maxSeriesA = variableASeries.Max();
var normalizedSeriesA = variableASeries.Select(x => (x - minSeriesA) / (maxSeriesA - minSeriesA)).ToList();
var matrixSeriesA = new LineSeries(variableA.VariableName);
for (int i = 0; i < normalizedSeriesA.Count; i++)
{
matrixSeriesA.Points.Add(new DataPoint(i, (-1 + normalizedSeriesA[i]) * seriesMatrixScale - 1 - seriesMatrixScale * (variableIndex + 1)));
}
plotModel.Series.Add(matrixSeriesA);
var minSeriesB = variableBSeries.Min();
var maxSeriesB = variableBSeries.Max();
var normalizedSeriesB = variableBSeries.Select(x => (x - minSeriesB) / (maxSeriesB - minSeriesB)).ToList();
var matrixSeriesB = new LineSeries(variableB.VariableName);
for (int i = 0; i < normalizedSeriesB.Count; i++)
{
matrixSeriesB.Points.Add(new DataPoint(-normalizedSeriesB[i] * seriesMatrixScale - 1 - seriesMatrixScale * (variableIndex + 1), i));
}
plotModel.Series.Add(matrixSeriesB);
}
plotModel.Axes.Add(new LinearAxis(AxisPosition.Bottom, " Series A")
{
Maximum = Math.Max(xLength, yLength), PositionAtZeroCrossing = true
});
plotModel.Axes.Add(new LinearAxis(AxisPosition.Left, " Series B")
{
Maximum = Math.Max(xLength, yLength), PositionAtZeroCrossing = true
});
MatrixPlot.Model = plotModel;
}
/// <summary>
/// Generate average of supplied series.
/// </summary>
public static double[] Average(List <double[]> series, int maxIterations = 100)
{
if (series == null || series.Count == 0)
{
throw new Exception("series null or empty");
}
if (series.Count == 1)
{
return(series[0]);
}
if (series.Select(x => x.Length).Distinct().Count() != 1)
{
throw new Exception("Series must be of equal length");
}
int length = series[0].Length;
//initializing the average to a simple mean of every point, without DTW
//double[] average = new double[length];
//for (int i = 0; i < series[0].Length; i++)
//{
// average[i] = series.Average(x => x[i]);
//}
//initialize to a random series from the input
//var r = new Random();
//double[] average = series[r.Next(0, series.Count - 1)];
//initialize to series closest to median min/max after detrending
List <double[]> tempSeries = series.Select(Detrend).ToList();
List <int> maxIndexes = tempSeries.Select(x => x.IndexOfMax()).ToList();
List <int> minIndexes = tempSeries.Select(x => x.IndexOfMin()).ToList();
double medianMaxIndex = maxIndexes.Median();
double medianMinIndex = minIndexes.Median();
var distances = maxIndexes.Select((x, i) => Math.Pow(x - medianMaxIndex, 2) + Math.Pow(minIndexes[i] - medianMinIndex, 2)).ToList();
int selectedSeries = distances.IndexOfMin();
double[] average = series[selectedSeries];
//this list will hold the values of each aligned point,
//later used to construct the aligned average
List <double>[] points = new List <double> [length];
for (int i = 0; i < length; i++)
{
points[i] = new List <double>();
}
double prevTotalDist = -1;
double totalDist = -2;
//sometimes the process gets "stuck" in a loop between two different states
//so we have to set a hard limit to end the loop
int count = 0;
//get the path between each series and the average
while (totalDist != prevTotalDist && count < maxIterations)
{
prevTotalDist = totalDist;
//clear the points from the last calculation
foreach (var list in points)
{
list.Clear();
}
//here we do the alignment for every series
foreach (double[] ts in series)
{
var dtw = new Dtw(new[] { new SeriesVariable(ts, average) });
Tuple <int, int>[] path = dtw.GetPath();
//use the path to distribute the points according to the warping
Array.ForEach(path, x => points[x.Item2].Add(ts[x.Item1]));
}
//Then simply construct the new average series by taking the mean of every List in points.
average = points.Select(x => x.Average()).ToArray();
//calculate Euclidean distance to stop the loop if no further improvement can be made
double[] average1 = average;
totalDist = series.Sum(x => x.Select((y, i) => Math.Pow(y - average1[i], 2)).Sum()); //we get convergence even though there's still work to be done
count++;
}
return(average);
}
static void DTWCsv(string[] files)
{
for (int fileId = 0; fileId < files.Length; fileId++)
{
if (File.Exists(files[fileId].Split('\\').Last() + "_pearsonDataCsv.txt"))
{
File.Copy(files[fileId].Split('\\').Last() + "_pearsonDataCsv.txt", files[fileId] + "_pearsonDataCsv.txt", true);
//Console.WriteLine("Copied " + files[fileId].Split('\\').Last() + "_pearsonDataCsv.txt" + " to " + files[fileId] + "_pearsonDataCsv.txt");
if (File.Exists(files[fileId].Split('\\').Last() + "dtwCostInfo.txt"))
{
File.Copy(files[fileId].Split('\\').Last() + "_pearsonDataCsv.txt", files[fileId] + "dtwCostInfo.txt", true);
}
Console.WriteLine(files[fileId].Split('\\').Last() + " is already done, skipping.");
continue;
}
var freed = GC.GetTotalMemory(false);
GC.Collect(GC.MaxGeneration, GCCollectionMode.Forced, true);
Console.WriteLine("Garbage Collection completed - memory:" + ((double)GC.GetTotalMemory(false) / 1024 / 1024 / 1024).ToString("0.0") + " GB (freed " + (freed / 1024 / 1024) + " MB)");
Console.WriteLine("Performing DTW on csv data " + fileId + " of " + files.Length + "..");
var watch = System.Diagnostics.Stopwatch.StartNew();
string[] data = File.ReadAllLines(files[fileId]);
Console.WriteLine("Data points: " + data.Length);
List <double> testDataPoints = new List <double>();
List <double> recallDataPoints = new List <double>();
foreach (var line in data)
{
var split = line.Replace(',', '.').Split(';');
double test = double.Parse(split[0], System.Globalization.CultureInfo.InvariantCulture);
double recall = double.Parse(split[1], System.Globalization.CultureInfo.InvariantCulture);
testDataPoints.Add(test);
recallDataPoints.Add(recall);
}
// Dtw dtw = new Dtw(testDataPoints.ToArray(), recallDataPoints.ToArray(), DistanceMeasure.Euclidean, true, true, null, null, 700);
Dtw dtw = new Dtw(testDataPoints.ToArray(), recallDataPoints.ToArray(), DistanceMeasure.Euclidean, true, true, slopeStepSizeDiagonal: 2, slopeStepSizeAside: 1);
var path = dtw.GetPath();
var cost = dtw.GetCost();
//var distanceMatrix = dtw.GetDistanceMatrix();
//var costMatrix = dtw.GetCostMatrix();
File.WriteAllText(files[fileId].Split('\\').Last() + "dtwCostInfo.txt",
"cost=" + cost +
"\nbefore_length=" + data.Length +
"\nbefore_cost=" + (cost / data.Length) +
"\nafter_length=" + path.Length +
"\nafter_cost=" + (cost / path.Length)
);
//PngExporter pngify = new PngExporter();
//pngify.Width = 36000;
//pngify.Height = 4000;
//var model = new PlotModel() { Title = "Red = test, blue = recall" };
//var aSeries = new OxyPlot.Series.LineSeries() { Color = OxyColors.Blue, MarkerSize = 10 };
//var bSeries = new OxyPlot.Series.LineSeries() { Color = OxyColors.Red, MarkerSize = 10 };
//for (int i = 0; i < testDataPoints.Count; i++)
//{
// aSeries.Points.Add(new DataPoint(i, testDataPoints[i]));
//}
//for (int i = 0; i < recallDataPoints.Count; i++)
//{
// bSeries.Points.Add(new DataPoint(i, recallDataPoints[i]));
//}
//List<string> pearsonData = new List<string>();
//foreach (var pairing in path)
//{
// var lineSeries = new OxyPlot.Series.LineSeries() { Color = OxyColors.Gray, MarkerSize = 0.05 };
// lineSeries.Points.Add(new DataPoint(pairing.Item1, testDataPoints[pairing.Item1]));
// lineSeries.Points.Add(new DataPoint(pairing.Item2, recallDataPoints[pairing.Item2]));
// model.Series.Add(lineSeries);
// pearsonData.Add(testDataPoints[pairing.Item1].ToString().Replace(',', '.') + ";" + recallDataPoints[pairing.Item2].ToString().Replace(',', '.'));
//}
//var pears = MathNet.Numerics.Statistics.Correlation.Pearson(path.Select(x => testDataPoints[x.Item1]).ToList(), path.Select(x => recallDataPoints[x.Item2]).ToList());
//Console.WriteLine("Pearson for " + files[fileId] + ":");
//Console.WriteLine(pears.ToString());
//File.WriteAllLines(files[fileId].Split('\\').Last() + "_pearsonDataCsv.txt", pearsonData);
//model.Series.Add(aSeries);
//model.Series.Add(bSeries);
//pngify.ExportToFile(model, files[fileId].Split('\\').Last() + "_csv.png");
watch.Stop();
Console.WriteLine("Done in " + watch.Elapsed);
Console.WriteLine("");
}
}
