public static MemoryGroupByOperation AddLinearRegression(this MemoryGroupByOperation op, string sourceXcolumn, string sourceYcolumn, string alphaColumn, string bColumn, string countColumn)
{
return(op.AddAggregator((aggregate, groupRows) =>
{
var count = 0;
double sumX = 0.0, sumY = 0.0, sumX2 = 0.0, sumXy = 0.0;
foreach (var row in groupRows)
{
if (!row.HasValue(sourceXcolumn) || !row.HasValue(sourceYcolumn))
{
return;
}
var x = row.GetAs <double>(sourceXcolumn);
var y = row.GetAs <double>(sourceYcolumn);
count++;
sumX += x;
sumX2 += x * x;
sumY += y;
sumXy += x * y;
}
var divider = (count * sumX2) - (sumX * sumX);
double?alpha;
double?b = null;
alpha = count == 1 || Math.Abs(divider) < 0.0001
? 0.0
: ((count * sumXy) - (sumX * sumY)) / divider;
if (count == 1)
{
b = sumY;
}
else if (count > 1)
{
var avgX = sumX / count; // need to be the average of the full aggregate group
var avgY = sumY / count;
b = avgY - (alpha * avgX);
}
aggregate[bColumn] = b;
aggregate[alphaColumn] = alpha;
aggregate[countColumn] = groupRows.Count;
}));
}
/// <summary>
/// New value will be int.
/// </summary>
public static MemoryGroupByOperation AddIntMax(this MemoryGroupByOperation op, string sourceColumn, string targetColumn = null)
{
return(op.AddAggregator((aggregate, groupRows) => aggregate[targetColumn ?? sourceColumn] = groupRows.Max(x => x.GetAs(sourceColumn, 0))));
}