We (Jonathan Lindblum and Jaime Sevilla) have written a tutorial about how to use PyMC3 to model a record progression over time.
The basic setup is that the data observed in each timestep corresponds to the best attempt seen so far at a sport / videogame / etc. From there we want to infer the parameters of the distribution of each attempt.
We show how we can use DensityDist
in PyMC3 to model this scenario, fit the model with NUTS, sample the posterior predictive distribution and forecast future records.
Feeback would be welcome!