@LeoEgidi , in this merge request, I have implemented for both
a) Diagonal Inflated Biv. Poisson (2003, Karlis-Ntzoufras)
b) Zero-Inflated Skellam (2009, Karlis - Ntzoufras).
the following work:
Added both fitted models within stan_foot.R using the corresponding names of "diag_infl_biv_pois", "zero_infl_skellam" using also 2 different ways of formulation for their corresponding log-likelihoods.
Added the new fitted models to the rapid_guide.rmd and corresponding html file.
The results of both newly fitted models are much similar to the simple bivariate Poisson as well as Skellam since they produce much similar LOO-IC and both in-sample, out-of-sample evaluation of those ones.
However, the generation of either replications or predictions (out-sample) are implemented for both fitted models as you already implement it through the generation of replications or predictions by the double Poisson distributions. This is an indirect way to generate those ones but I think that the correct way is to create a custom function for each fitted model (for example diag_infl_biv_pois_rng function in Stan).
Next Steps:
-Propose functions enabling the calculation of betting odds using our fitted models.
@LeoEgidi , in this merge request, I have implemented for both
a) Diagonal Inflated Biv. Poisson (2003, Karlis-Ntzoufras)
b) Zero-Inflated Skellam (2009, Karlis - Ntzoufras).
the following work:
Added both fitted models within stan_foot.R using the corresponding names of "diag_infl_biv_pois", "zero_infl_skellam" using also 2 different ways of formulation for their corresponding log-likelihoods.
Added the new fitted models to the rapid_guide.rmd and corresponding html file.
The results of both newly fitted models are much similar to the simple bivariate Poisson as well as Skellam since they produce much similar LOO-IC and both in-sample, out-of-sample evaluation of those ones.
However, the generation of either replications or predictions (out-sample) are implemented for both fitted models as you already implement it through the generation of replications or predictions by the double Poisson distributions. This is an indirect way to generate those ones but I think that the correct way is to create a custom function for each fitted model (for example diag_infl_biv_pois_rng function in Stan).
Next Steps: -Propose functions enabling the calculation of betting odds using our fitted models.