public static void ReplaceMissingsByGaussianWholeMatrix(double width, double shift, IMatrixData data)
{
int[] colInds = ArrayUtils.ConsecutiveInts(data.ExpressionColumnCount);
List<float> allValues = new List<float>();
for (int i = 0; i < data.RowCount; i++){
foreach (int t in colInds){
float x = data[i, t];
if (!float.IsNaN(x) && !float.IsInfinity(x)){
allValues.Add(x);
}
}
}
double stddev;
double mean = ArrayUtils.MeanAndStddev(allValues.ToArray(), out stddev);
double m = mean - shift*stddev;
double s = stddev*width;
Random2 r = new Random2();
for (int i = 0; i < data.RowCount; i++){
foreach (int t in colInds){
if (float.IsNaN(data[i, t]) || float.IsInfinity(data[i, t])){
data[i, t] = (float) r.NextGaussian(m, s);
data.IsImputed[i, t] = true;
}
}
}
}
private static void ReplaceMissingsByGaussianForOneColumn(double width, double shift, IMatrixData data, int colInd)
{
List<float> allValues = new List<float>();
for (int i = 0; i < data.RowCount; i++){
float x = data[i, colInd];
if (!float.IsNaN(x) && !float.IsInfinity(x)){
allValues.Add(x);
}
}
double stddev;
double mean = ArrayUtils.MeanAndStddev(allValues.ToArray(), out stddev);
double m = mean - shift*stddev;
double s = stddev*width;
Random2 r = new Random2();
for (int i = 0; i < data.RowCount; i++){
if (float.IsNaN(data[i, colInd]) || float.IsInfinity(data[i, colInd])){
data[i, colInd] = (float) r.NextGaussian(m, s);
data.IsImputed[i, colInd] = true;
}
}
}
public void ProcessData(IMatrixData mdata, Parameters param, ref IMatrixData[] supplTables,
ref IDocumentData[] documents, ProcessInfo processInfo)
{
Random2 rand = new Random2();
double std = param.GetDoubleParam("Standard deviation").Value;
for (int i = 0; i < mdata.RowCount; i++){
for (int j = 0; j < mdata.ExpressionColumnCount; j++){
mdata[i, j] += (float) rand.NextGaussian(0, std);
}
}
}
