Math.NET Numerics

New Post: Non-Mathematician needs help getting started! (Please)

Gregor959 — Sun, 19 May 2013 13:36:25 GMT

I am assuming you know how to do a normal regression.
For a weighted regression, you would need to compute the correlation or covariance matrix in a weighted fashion and then do the regression .
I did this in Matlab a long time ago, but have not had a need to do so in MathNet Numerics. If the rest is Greek to you this could be too, but it really is just a bunch of weighted sums. There might be higher level statistical libraries that can do weighted regression already.

New Post: Is there a covariance method for 2 dataset?

Gregor959 — Sun, 19 May 2013 12:26:49 GMT

You can use the Pearsons correlation to compute the elements of a correlation matrix:
I am sure this is not the fastest method and ignores the fact that the matrix could be better implemented with a square symmetric matrix but show how to get started

public static Matrix<double> Correl(Matrix<double> x)

    {
        Matrix<double> correlM = new DenseMatrix(x.RowCount, x.RowCount);

        //off diagonal elements
        foreach (var rowx in x.RowEnumerator())
        {
            foreach (var rowy in x.RowEnumerator())
            {
                if (rowy.Item1 > rowx.Item1)
                {
                    correlM[rowx.Item1, rowy.Item1] = MathNet.Numerics.Statistics.Correlation.Pearson(rowx.Item2, rowy.Item2);
                }
                if (rowy.Item1 < rowx.Item1)
                {
                    correlM[rowx.Item1, rowy.Item1] = correlM[rowy.Item1, rowx.Item1];
                }

            }
        }
        //Diagonal elements
        foreach (var rowx in x.RowEnumerator())
        {
            correlM[rowx.Item1, rowx.Item1] = MathNet.Numerics.Statistics.Correlation.Pearson(rowx.Item2, rowx.Item2);
        }

        return correlM;
    }

New Post: Lognormal sampling

cdrnet — Fri, 10 May 2013 08:15:09 GMT

New Post: Lognormal sampling

cdrnet — Fri, 10 May 2013 08:11:20 GMT

From a first look it seems it is working correctly. The mu/sigma parameters do not specify the mean and standard deviation of the variable (these are called "Mean" and "StdDev"), but of the natural logarithm of the variable (thus confirming your observation in your second post). This is consistent with wikipedia, although admittedly quite confusing.

New Post: Lognormal sampling

mcmillab — Fri, 10 May 2013 02:52:17 GMT

It looks like I get the correct result if I do this:

double sample = Math.Log(LogNormal.Sample(aRandomGenerator, 0.000001, 0.000001));

Why's that?

New Post: Lognormal sampling

mcmillab — Fri, 10 May 2013 02:04:45 GMT

I'm trying to generate lognormal samples.
When I do something like this:

double sample = LogNormal.Sample(aRandomGenerator, 0.000001, 0.000001)

I get numbers back close to 1, but I was expected them to be distributed around my mean (0.000001).

What am I doing wrong?

New Post: Root solver

cdrnet — Wed, 01 May 2013 17:53:42 GMT

That worked nicely, I've pulled it to mainline already.

Thanks,
Christoph

Source code checked in, #74bc285807f8c5d68b805e7f45f43ce43abe49f9

Christoph Ruegg — Wed, 01 May 2013 17:48:44 GMT

Update Portable project

Source code checked in, #db29d03b9a6950e956e19ed6aede9907cdd01e02

Christoph Ruegg — Wed, 01 May 2013 17:48:43 GMT

Use the new NonConvergenceException in more places, where appropriate

Source code checked in, #f142f3abb70d739a0cb9414c07ae63e14f75ab0f

Christoph Ruegg — Wed, 01 May 2013 17:48:42 GMT

Common NonConvergenceException

Source code checked in, #96d87cb23226da82209845ec3c7a96c5d5f13517

Christoph Ruegg — Wed, 01 May 2013 17:47:53 GMT

Merge brent root finding algorithm

Source code checked in, #ddac0a0b615b62caf276ee029759424aa8af8646

Christoph Ruegg — Wed, 01 May 2013 17:26:24 GMT

RootFinding: algorithm cosmetics

New Post: Root solver

candychiu — Tue, 30 Apr 2013 20:29:49 GMT

I created a fork, and pushed changes there.

https://git01.codeplex.com/forks/candychiu/rootsolver

Internally I used MbUnit for unit testing. The simple test case needs modification and extension.

New Post: Root solver

cdrnet — Mon, 29 Apr 2013 13:30:54 GMT

Hi Candy,

Certainly, that would be welcome. How can we get hold of your work?

Thanks,
Christoph

New Post: Root solver

candychiu — Mon, 29 Apr 2013 12:33:30 GMT

Hi,

I added the Brent root solver to my copy of the source code. Would you add it to the repository for others to use?

Candy

New Post: API Reference out-of-date

cdrnet — Sat, 27 Apr 2013 17:39:20 GMT

I've fixed docu and uploaded a new build. It should now include inherited members, list static methods separately and warn prominently on anything that has been marked as obsolete.

New Post: API Reference out-of-date

cdrnet — Fri, 26 Apr 2013 20:08:10 GMT

Yes, I've noticed that as well. It also doesn't flag obsolete members yet which can be confusing (e.g. DenseMatrix.Data vs .Values). I'm working on a fix within docu, not too hard but needs some refactoring first (wish it was written in F#...).

I used to kind of hate doxygen-style docs 6-7 years ago when I last tried, but I guess it has improved since (or can be configured accordingly) and is worth a new trial :)

New Post: API Reference out-of-date

creinke — Fri, 26 Apr 2013 18:12:58 GMT

So it appears that docu doesn't put show documentation for inherited members -- it would be handy if it included their documentation (as doxygen can be instructed to do) or at least linked to their documentation (as doxygen otherwise does).

For example, looking at DenseMatrix.Add you wouldn't know that there actually is documentation for it as Matrix<T>.Add.

Maybe I can convince you to switch to doxygen? :P The results are less spartan (perhaps even bloated), but they're also quite a bit more useable in my opinion.

New Post: API Reference out-of-date

cdrnet — Fri, 26 Apr 2013 12:31:17 GMT

Indeed, I've just regenerated it for v2.5.0.

I also fixed a bug in docu (some open source doc generation engine we use) that caused a lot of methods to be omitted previously, so the reference should be much more complete now.

Thanks,
Christoph

New Post: API Reference out-of-date

creinke — Thu, 25 Apr 2013 19:36:23 GMT

The API Reference is terribly out of date -- based on 2.1.2.21 when the current version is 2.5.0.27. I say terribly because the source-embedded documentation has apparently gotten much better over that time! Why not update it where people look for it?