ToroidalMixture parameter estimation about libdirectional HOT 2 CLOSED

Hi,
if you want to obtain a pdf based on your samples: I have added the capability to determine parameters of a multidimensional Fourier series (HypertoroidalFourierDistribution) based on samples. The change will be pushed to this repository soon.
Small example how you can use it:

noCoeffPerDim=7;
samples = [ToroidalWNDistribution([1;1],0.3*[1,0.7;0.7,1]).sample(1000),...
    ToroidalWNDistribution([4;4],0.3*[1,-0.3;-0.3,1]).sample(1000)];
wd = HypertoroidalWDDistribution(samples);
figure(1), wd.plot;
hfd = HypertoroidalFourierDistribution.fromDistribution(wd, [noCoeffPerDim, noCoeffPerDim], 'identity');
figure(2), hfd.plot;

Of course, you just use your samples instead of generating them as in the example. Choose noCoeffsPerDim wisely. Low values result in a smooth density, that, however, has little power to approximate complicated densities. High values can result in a distribution that is more "peaky" than you may desire. This is because, in essence, you can never know for sure how the density that generated the samples may look like. If you only have samples, then even a distribution that only has probability mass on these points may actually be plausible (although this is rarely the desired solution).
Hope it helps.

Greetings,
Florian

from libdirectional.

If there is no further need for discussion, I think this issue can be closed.

from libdirectional.