Evaluate

[ywteh]

Evaluate method.

• 1. evaluate short-term forecasting, based on historical data. Compare against:

• https://epiforecasts.io approach (Cori et al, with additional data preprocessing)
• https://arxiv.org/abs/2006.16487
• Look into covid forecasting methods from US (see van der Schaar paper ref)

Evaluate based on predictive probability of future counts.

• 2. evaluate based on inference of Rt. Use simulator here to evaluate:
  https://www.medrxiv.org/content/10.1101/2020.06.18.20134858v2.full.pdf

[MJHutchinson]

Use simulator here to evaluate:
https://www.medrxiv.org/content/10.1101/2020.06.18.20134858v2.full.pdf

[ywteh]

Added code to compute predictive probs and forecasting here: 45d3a68

Note that main file for this change is now run_cori_eval.r and cori-gp-immi-eval.stan Sorry for the terrible version management. It involves also some refactoring so didn't want to override cori-gp-immi.stan.

Sorry for trouble.

[ywteh]

Added code, following options and stan functions approach of @MJHutchinson to allow options for GP kernels, cross-coupled metapopulation models, and likelihood models.

@shehzaidi is it easy enough to set up ziz runs now?

[MJHutchinson]

Switching simulation methods to generate data from our assumed model, rather than from an SEIR model. Update in progress.

[brynhayder]

Switching simulation methods to generate data from our assumed model, rather than from an SEIR model. Update in progress.

Are we no longer using the metapopulation simulation with changing Rt?

[MJHutchinson]

We are using metapopulation effects, just not an SEIR model. Instead, we're generating data from the model we're proposing to check it can recover its own parameters.

[brynhayder]

ok cool, I was under the impression that we would be doing both.