3

I have a rectangular matrix thus:

1, 3, 2, 4, n..
4, 2, 1, 5, n..
n..

and a vector thus:
1, 2, 5, 6, 7, n..

I need to solve the least-squares equation for all columns in the matrix, but I want to constrain the results such that all answers are greater than zero.

I've added the Math.NET package and got as far as
matrix.QR().Solve(...

Presumably there is some way to iteratively solve this but there doesn't appear to be a way to specify constraints/conditions to the Solve method, and I'm not sure what other method(s) I should use.

Partial code added below:

//compounds is Dictionary<int, List<double>> 
var xdata = compounds.Values.Select(v => v.ToArray()).ToArray();
var ydata = new DenseVector(someKnownValues.ToArray());
var matrix = DenseMatrix.OfColumns(ydata.Count(), xdata.Count(), xdata);
var factors = matrix.QR().Solve(ydata);
TylerH
  • 20,799
  • 66
  • 75
  • 101
Simon
  • 151
  • 5
  • Would you mind adding your code so we can see where you are with solving this on your own? – tnw Mar 04 '14 at 14:39
  • @tnw - brief code added; it's difficult to add more because there is a lot of other processing going into gathering the 'compounds' together. – Simon Mar 04 '14 at 14:55

1 Answers1

4

Unfortunately solving non-negative (or positive) least squares problems is not supported by Math.NET Numerics out of the box.

Some further information and an algorithm: Non-negative least squares (Wikipedia)

Christoph Rüegg
  • 4,626
  • 1
  • 20
  • 34
  • Thanks Christoph - I will investigate some other libraries – Simon Mar 04 '14 at 16:35
  • Hi again Christoph, I am just wondering whether I am able to use any of the other solving methods in Math.NET, such as the WeightedRegression? – Simon Mar 12 '14 at 15:03