private static List <Tuple <string, Parameter> > GetPointPredictionValues(Dictionary <int, float> pointIdOverallDensity, Dictionary <int, Dictionary <string, double> > pointIdIncidentDensity)
{
List <Tuple <string, Parameter> > pointPredictionValues = new List <Tuple <string, Parameter> >(pointIdIncidentDensity.Count);
foreach (int pointId in pointIdIncidentDensity.Keys)
{
string timeParameterName = "@time_" + pointId;
pointPredictionValues.Add(new Tuple <string, Parameter>(PointPrediction.GetValue(pointId, timeParameterName, pointIdIncidentDensity[pointId], pointIdOverallDensity[pointId]), new Parameter(timeParameterName, NpgsqlDbType.Timestamp, DateTime.MinValue)));
}
return(pointPredictionValues);
}
public override void Apply(Prediction prediction)
{
List <PointPrediction> pointPredictions = prediction.PointPredictions;
if (pointPredictions.Count > 0)
{
Dictionary <int, Point> idPoint = new Dictionary <int, Point>();
foreach (Point p in prediction.Points)
{
idPoint.Add(p.Id, p);
}
IEnumerable <PostGIS.Point> kdeEvalPoints = pointPredictions.Select(p => idPoint[p.PointId].Location);
List <PostGIS.Point> kdeInputPoints = new List <PostGIS.Point>();
foreach (string incident in pointPredictions[0].IncidentScore.Keys.ToArray())
{
if (incident != PointPrediction.NullLabel)
{
double minScore = pointPredictions.Min(p => p.IncidentScore[incident]);
kdeInputPoints.Clear();
foreach (PointPrediction pointPrediction in pointPredictions)
{
PostGIS.Point pointPredictionLocation = idPoint[pointPrediction.PointId].Location;
double replicates = pointPrediction.IncidentScore[incident] / minScore;
for (int i = 0; i < replicates; ++i)
{
kdeInputPoints.Add(pointPredictionLocation);
}
}
List <float> density = KernelDensityDCM.GetDensityEstimate(kdeInputPoints, _sampleSize, false, 0, 0, kdeEvalPoints, _normalize);
for (int i = 0; i < density.Count; ++i)
{
pointPredictions[i].IncidentScore[incident] = density[i];
}
}
}
foreach (PointPrediction pointPrediction in pointPredictions)
{
pointPrediction.TotalThreat = pointPrediction.IncidentScore.Keys.Sum(incident => incident == PointPrediction.NullLabel ? 0 : pointPrediction.IncidentScore[incident]);
}
PointPrediction.UpdateThreatScores(pointPredictions, prediction);
}
prediction.SmoothingDetails = GetSmoothingDetails();
}
protected override void Run(Prediction prediction)
{
List <PostGIS.Point> predictionPoints = new List <PostGIS.Point>();
Area predictionArea = prediction.PredictionArea;
double areaMinX = predictionArea.BoundingBox.MinX;
double areaMaxX = predictionArea.BoundingBox.MaxX;
double areaMinY = predictionArea.BoundingBox.MinY;
double areaMaxY = predictionArea.BoundingBox.MaxY;
for (double x = areaMinX + prediction.PredictionPointSpacing / 2d; x <= areaMaxX; x += prediction.PredictionPointSpacing) // place points in the middle of the square boxes that cover the region - we get display errors from pixel rounding if the points are exactly on the boundaries
{
for (double y = areaMinY + prediction.PredictionPointSpacing / 2d; y <= areaMaxY; y += prediction.PredictionPointSpacing)
{
predictionPoints.Add(new PostGIS.Point(x, y, predictionArea.Shapefile.SRID));
}
}
List <PostGIS.Point> incidentPoints = new List <PostGIS.Point>(Incident.Get(TrainingStart, TrainingEnd, predictionArea, IncidentTypes.ToArray()).Select(i => i.Location));
predictionPoints.AddRange(incidentPoints);
Console.Out.WriteLine("Filtering prediction points to prediction area");
predictionPoints = predictionArea.Intersects(predictionPoints, prediction.PredictionPointSpacing / 2f).Select(i => predictionPoints[i]).ToList();
NpgsqlConnection connection = DB.Connection.OpenConnection;
try
{
Console.Out.WriteLine("Inserting points into prediction");
Point.CreateTable(prediction, predictionArea.Shapefile.SRID);
List <int> predictionPointIds = Point.Insert(connection, predictionPoints.Select(p => new Tuple <PostGIS.Point, string, DateTime>(p, PointPrediction.NullLabel, DateTime.MinValue)), prediction, predictionArea, false);
Console.Out.WriteLine("Running overall KDE for " + IncidentTypes.Count + " incident type(s)");
List <float> density = GetDensityEstimate(incidentPoints, _trainingSampleSize, false, 0, 0, predictionPoints, _normalize);
Dictionary <int, float> pointIdOverallDensity = new Dictionary <int, float>(predictionPointIds.Count);
int pointNum = 0;
foreach (int predictionPointId in predictionPointIds)
{
pointIdOverallDensity.Add(predictionPointId, density[pointNum++]);
}
Dictionary <int, Dictionary <string, double> > pointIdIncidentDensity = new Dictionary <int, Dictionary <string, double> >(pointIdOverallDensity.Count);
if (IncidentTypes.Count == 1)
{
string incident = IncidentTypes.First();
foreach (int pointId in pointIdOverallDensity.Keys)
{
Dictionary <string, double> incidentDensity = new Dictionary <string, double>();
incidentDensity.Add(incident, pointIdOverallDensity[pointId]);
pointIdIncidentDensity.Add(pointId, incidentDensity);
}
}
else
{
foreach (string incidentType in IncidentTypes)
{
Console.Out.WriteLine("Running KDE for incident \"" + incidentType + "\"");
incidentPoints = new List <PostGIS.Point>(Incident.Get(TrainingStart, TrainingEnd, predictionArea, incidentType).Select(i => i.Location));
density = GetDensityEstimate(incidentPoints, _trainingSampleSize, false, 0, 0, predictionPoints, _normalize);
if (density.Count > 0)
{
pointNum = 0;
foreach (int predictionPointId in predictionPointIds)
{
pointIdIncidentDensity.EnsureContainsKey(predictionPointId, typeof(Dictionary <string, double>));
pointIdIncidentDensity[predictionPointId].Add(incidentType, density[pointNum++]);
}
}
}
}
PointPrediction.CreateTable(prediction);
PointPrediction.Insert(GetPointPredictionValues(pointIdOverallDensity, pointIdIncidentDensity), prediction, false);
Smooth(prediction);
}
finally
{
DB.Connection.Return(connection);
}
}
public override void Apply(Prediction prediction)
{
List <PointPrediction> pointPredictions = prediction.PointPredictions;
if (pointPredictions.Count > 0)
{
Dictionary <int, Point> idPoint = new Dictionary <int, Point>();
foreach (Point p in prediction.Points)
{
idPoint.Add(p.Id, p);
}
List <Tuple <PointPrediction, Dictionary <string, double> > > pointPredictionIncidentScore = new List <Tuple <PointPrediction, Dictionary <string, double> > >(pointPredictions.Count);
Set <Thread> threads = new Set <Thread>(Configuration.ProcessorCount);
for (int i = 0; i < Configuration.ProcessorCount; ++i)
{
Thread t = new Thread(new ParameterizedThreadStart(core =>
{
List <Tuple <PointPrediction, Dictionary <string, double> > > threadPointPredictionIncidentScore = new List <Tuple <PointPrediction, Dictionary <string, double> > >((int)((pointPredictions.Count / (float)Configuration.ProcessorCount) + 1));
for (int j = (int)core; j < pointPredictions.Count; j += Configuration.ProcessorCount)
{
PointPrediction pointPrediction = pointPredictions[j];
Dictionary <PointPrediction, double> neighborInvDist = new Dictionary <PointPrediction, double>();
foreach (PointPrediction neighbor in pointPredictions)
{
double distance = idPoint[pointPrediction.PointId].Location.DistanceTo(idPoint[neighbor.PointId].Location);
if (pointPrediction == neighbor || (distance >= _minimum && distance <= _maximum))
{
neighborInvDist.Add(neighbor, _maximum - distance);
}
}
double totalInvDistance = neighborInvDist.Values.Sum();
Dictionary <string, double> incidentScore = new Dictionary <string, double>(pointPrediction.IncidentScore.Count);
foreach (string incident in pointPrediction.IncidentScore.Keys)
{
incidentScore.Add(incident, neighborInvDist.Keys.Sum(neighbor => (neighborInvDist[neighbor] / totalInvDistance) * neighbor.IncidentScore[incident]));
}
threadPointPredictionIncidentScore.Add(new Tuple <PointPrediction, Dictionary <string, double> >(pointPrediction, incidentScore));
}
lock (pointPredictionIncidentScore) { pointPredictionIncidentScore.AddRange(threadPointPredictionIncidentScore); }
}));
t.Start(i);
threads.Add(t);
}
foreach (Thread t in threads)
{
t.Join();
}
foreach (Tuple <PointPrediction, Dictionary <string, double> > pointPredictionScores in pointPredictionIncidentScore)
{
pointPredictionScores.Item1.IncidentScore = pointPredictionScores.Item2;
pointPredictionScores.Item1.TotalThreat = pointPredictionScores.Item2.Values.Sum();
}
PointPrediction.UpdateThreatScores(pointPredictions, prediction);
prediction.SmoothingDetails = GetSmoothingDetails();
}
}
public override void Apply(Prediction prediction)
{
List <PointPrediction> pointPredictions = prediction.PointPredictions;
if (pointPredictions.Count > 0)
{
Dictionary <int, Point> idPoint = new Dictionary <int, Point>();
foreach (Point p in prediction.Points)
{
idPoint.Add(p.Id, p);
}
string inputPointsPath = Path.GetTempFileName();
string evalPointsPath = Path.GetTempFileName();
string outputPath = Path.GetTempFileName();
using (StreamWriter evalPointsFile = new StreamWriter(evalPointsPath))
{
foreach (PostGIS.Point p in pointPredictions.Select(p => idPoint[p.PointId].Location))
{
evalPointsFile.WriteLine(p.X + "," + p.Y);
}
evalPointsFile.Close();
}
foreach (string incident in pointPredictions[0].IncidentScore.Keys.ToArray())
{
if (incident != PointPrediction.NullLabel)
{
using (StreamWriter inputPointsFile = new StreamWriter(inputPointsPath))
{
inputPointsFile.WriteLine("threat,x,y");
foreach (PointPrediction pointPrediction in pointPredictions)
{
PostGIS.Point location = idPoint[pointPrediction.PointId].Location;
inputPointsFile.WriteLine(pointPrediction.IncidentScore[incident] + "," + location.X + "," + location.Y);
}
inputPointsFile.Close();
}
R.Execute(@"
library(earth)
input.points = read.csv(""" + inputPointsPath.Replace(@"\", @"\\") + @""",header=TRUE)
model = earth(threat ~ ., data = input.points, " + (_numberOfKnots == -1 ? "" : "nk = " + _numberOfKnots + ", ") + "fast.k = " + _consideredParentTerms + ", degree = " + _interactionDegree + @")
eval.points = read.csv(""" + evalPointsPath.Replace(@"\", @"\\") + @""",header=FALSE)
prediction = predict(model, eval.points, type=""response"")
prediction = (prediction - min(prediction)) / (max(prediction) - min(prediction))
write.table(prediction,file=""" + outputPath.Replace(@"\", @"\\") + @""",row.names=FALSE,col.names=FALSE)", false);
int pointNum = 0;
foreach (string line in File.ReadLines(outputPath))
{
pointPredictions[pointNum++].IncidentScore[incident] = double.Parse(line);
}
}
}
File.Delete(inputPointsPath);
File.Delete(evalPointsPath);
File.Delete(outputPath);
foreach (PointPrediction pointPrediction in pointPredictions)
{
pointPrediction.TotalThreat = pointPrediction.IncidentScore.Keys.Sum(incident => incident == PointPrediction.NullLabel ? 0 : pointPrediction.IncidentScore[incident]);
}
PointPrediction.UpdateThreatScores(pointPredictions, prediction);
prediction.SmoothingDetails = GetSmoothingDetails();
}
}
