New Post: Negative Samples from Geometric Distribution

Yes, that looks wrong. Neither Geometric nor NegativeBinomial distributions are currently covered by the tests to verify the distribution of the generated samples, and when added quickly they indeed both fail (VapnikChervonenkisTest). We should extend these tests to verify all sample distributions and make sure they hold.

Note that the side-by-side test you mentioned is invalid with Math.NET Numerics. The Sample routines are only supposed to guarantee the correct distribution but not any strict mapping from the uniform samples provided by the RNG (in fact, some of them take multiple samples from the RNG to generate a single sample).

Thanks,
Christoph