Modeling the following discrete-time stochastic epidemic SIR model

For several parameters where the cases have long periods of inactivity between epidemics, the calculation of the last_epi_day selects the wrong point.

Currently the calculation lives in a a Jupyter notebook. Move to its own script.

To compare the methods before the changepoint we consider two dates that can be computed heuristically:

1. For parameter scenarios where there's a long period of inactivity between two epidemics choose the date RIGHT before the start of the second epidemic.
2. For parameter scenarios where there isn't an inactive period between both epidemics, use the day right before the sigmoid curve starts changing.

Edit this in compute_days.py and then rerun for all param scenarios. The expectation is that the metrics should not be that different between methods during this early date, but should be quite different after.

Do create some pytests in \tests

Create functions for better likelihood visualization.

Question: Is there an optimal date where the calibration is optimal (in the sense that x% of data live in the x% CrI).

Rerun for only 100 parameter runs since it takes FOREVER to do the full set. Rerun for full set if there's anything interesting.

Manually write module for MCMC sampling with Metropolis-Hastings to help debug skinny posteriors with PyMC.