Distributions (Gamma, Poisson)

[benman1]

Hi
Thanks for creating the repo (nicely maintained and good to read). I have a problem, however.

I am trying to use other distribution families, like Poisson or Gamma. So I create the GLM as follows:

Glm glm = new Glm(GlmSolverType.GlmIrlsSvd, GlmDistributionFamily.Gamma) glm.fit(dMatrix)

or

Glm glm = new Glm() glm.distributionFamily = GlmDistributionFamily.Poisson glm.fit(dMatrix)

I only get nans in either case for predictions. Is there anything I am missing here? If I change this to linear. I get non-nan results on my test dataset (iris).

For some values of distribution family, such as InverseGaussian, I even get null exceptions from GlmAlgorithm (state is null).

protected TerminationEvaluationMethod shouldTerminate = (state, iteration) -> { if (state.improved() && state.improvement() >= this.mTol) { return iteration >= this.maxIters; } else { return false; } };

Help appreciated.
Thanks again
Ben.