Here I would like to open my previous question to a broader understanding of the PYMC3 library.
I would like to know is the following: once one has defined which family distribution to use (in this post case, the Poisson distribution), should the likely function be defined in terms of the inverse of the linking-function, or itself?
For instance, for the current problem (the Poisson distribution), I wonder whether one should use the Exponent of the dot product of matrices B and X (betas and feature matrix, respectively); or the Log of this dot product.
Please, let me know your thoughts in this regard.
Sincerely