Baeysian PCA on Poisson data

Hi,

I'm trying to build a PCA model to fit poisson data. My use case is nearly the same as the PCA tutorial, except that my observed data is Poisson distributed.

For example instead of `Variable.GaussianFromMeanAndPrecision` in
https://github.com/dotnet/infer/blob/master/src/Tutorials/BayesianPCA.cs#L102
my case would contain something like
`data[observation, feature] = Variable.Poisson(events * T[observation, feature])`.

My problem is that Variable.Poisson expects mean rates in Log space, so the above does not compile.
The error code is saying `Gaussian is not assignable from Gamma for result of method PoissonOp.MeanAverageLogarithm` (interesting thing regarding this error is that my code doesn't contain any gamma distribution - they are all gaussian - yet error report complains about gamma).

On the other hand, if I transform my principal components into log space and do a
`data[observation, feature] = Variable.Poisson(events * Variable.Exp( T[observation, feature] ))` then the code compiles, but I can't get correct results because MatrixMultiply doesn't do the right thing -- in log-space, weights and factors should be added rather than multiplied.

I'd be very thankful for any hints on how to approach this.