public static RocRow GetInstance(long crossValIndex, KeyValuePair <MerAndHlaToLength, bool> merAndHlaToLengthWithLabel, double prediction)
{
RocRow rocRow = new RocRow();
rocRow.CrossValIndex = crossValIndex;
rocRow.MerAndHlaToLength = merAndHlaToLengthWithLabel.Key;
rocRow.Label = merAndHlaToLengthWithLabel.Value;
rocRow.Prediction = prediction;
return(rocRow);
}
static public List <double> RankPositives(List <RocRow> rocRowCollection)
{
rocRowCollection.Sort(
delegate(RocRow x, RocRow y)
{
return(y.Prediction.CompareTo(x.Prediction)); //opposite of the report
});
List <double> rankPositives = new List <double>();
for (long iRocRow = 0; iRocRow < rocRowCollection.Count; ++iRocRow)
{
RocRow rocRow = rocRowCollection[(int)iRocRow];
if (rocRow.Label)
{
double rank = Rank(rocRowCollection, rocRow);
rankPositives.Add(rank);
}
}
return(rankPositives);
}
/// <summary>
/// !!!could be done much faster
/// </summary>
/// <param name="rocRowCollection"></param>
/// <param name="rocRow"></param>
/// <returns></returns>
private static double Rank(List <RocRow> rocRowCollection, RocRow rocRowTarget)
{
int rowsBetterThanTarget = 0;
int rowsSameThanTarget = 0;
foreach (RocRow rocRow in rocRowCollection)
{
if (rocRow.Prediction > rocRowTarget.Prediction)
{
++rowsBetterThanTarget;
}
else if (rocRow.Prediction == rocRowTarget.Prediction)
{
++rowsSameThanTarget;
}
}
int bestRank = rowsBetterThanTarget + 1;
int worseRank = rowsBetterThanTarget + rowsSameThanTarget;
double rank = (double)(bestRank + worseRank) / 2.0;
return(rank);
}
public static void ReportRocCurve(List <RocRow> rocRowCollection, string fileName, bool isComplete, long lineLimit, HlaFilter hlaFilterOrNull)
{
if (hlaFilterOrNull != null)
{
fileName = fileName + "FilteredOn" + hlaFilterOrNull.Name;
rocRowCollection = hlaFilterOrNull.Filter(rocRowCollection);
}
Debug.WriteLine(fileName);
using (TextWriter evalStream = File.CreateText(fileName))
{
rocRowCollection.Sort(
delegate(RocRow x, RocRow y)
{
return(y.Prediction.CompareTo(x.Prediction));
});
long positiveTotal = CountPositives(rocRowCollection);
long negativeTotal = rocRowCollection.Count - positiveTotal;
long positiveSoFar = positiveTotal;
long negativeSoFar = 0;
double prevPrediction = 1.0;
if (!isComplete)
{
evalStream.WriteLine("WARNING: The weights files were incomplete");
}
evalStream.WriteLine("{0}\t{1}\t{2}\t{3}\t{4}\t{5}\t{6}\t{7}\t{8}",
"CrossValIndex", "Label",
"Mer", "Hla",
"Prediction",
"FalsePositiveCount", "FalseNegativeCount",
rocRowCollection.Count > lineLimit ? "SampledFalsePositiveFraction" : "FalsePositiveFraction",
rocRowCollection.Count > lineLimit ? "SampledFalseNegativeFraction" : "FalseNegativeFraction");
evalStream.WriteLine("{0}\t{1}\t{2}\t{3}\t{4}\t{5}\t{6}\t{7}\t{8}",
"", "",
"", "",
1.0,
positiveSoFar, negativeSoFar,
(double)positiveSoFar / (double)positiveTotal,
(double)negativeSoFar / (double)negativeTotal);
for (long iRocRow = 0; iRocRow < rocRowCollection.Count; ++iRocRow)
{
RocRow rocRow = rocRowCollection[(int)iRocRow];
Debug.Assert(prevPrediction >= rocRow.Prediction); // real assert
//if (prevPrediction > rocRow.Prediction)
//{
// Console.WriteLine("WARNING: There needs to be a code change for when there are predictions with ties. Consider switching to SpecialFunctions.Roc (but it doesn't currently have sampling or filtering)");
//}
prevPrediction = rocRow.Prediction;
if (rocRow.Label)
{
--positiveSoFar;
}
else
{
++negativeSoFar;
}
if (DoOutput(iRocRow, rocRowCollection.Count, lineLimit))
{
evalStream.WriteLine("{0}\t{1}\t{2}\t{3}\t{4}\t{5}\t{6}\t{7}\t{8}",
rocRow.CrossValIndex, rocRow.Label,
rocRow.MerAndHlaToLength.Mer, rocRow.MerAndHlaToLength.HlaToLength,
rocRow.Prediction,
positiveSoFar, negativeSoFar,
(double)positiveSoFar / (double)positiveTotal,
(double)negativeSoFar / (double)negativeTotal);
}
}
Debug.Assert(positiveSoFar == 0); // real assert
Debug.Assert(negativeSoFar == negativeTotal);
}
}
