Questions to Jim Chiang (paraphrasing): From Randall Smith: Q: Have you considered parallelizing the calculations to make the execution for the fitting faster? A: The present code hasn't been written with parallelization in mind, but this is something worth considering. Certain parts of the calculation have been factored out and precomputed outside of the parallelization. From Vinay Kashyap: Q: Why do you have an N_pred in your objective function? Where does it come from? A: It comes from the n_k! that you see in the Poisson likelihood (viz. p11)