Postprocessing meta

[athowes]

Tagging @seabbs @kgostic @parksw3 for feedback:

Background

The existing issues we have about postprocessing are as follows:

• Help user pass newdata as sensible things (e.g. all strata) #213
• Posterior prediction for lognormal #214
• Create delay_summary function built on predict_delay_parameters #215

We have already implemented some postprocessing functionality, including including

• The add_mean_sd function Issue 169: Natural scale postprocessing #170
• The predict_delay_parameters function Issue 188: Add delay samples functionality #210
• (Not merged) the all_strata_newdata function which would feed into predict_data_parameters https://github.com/epinowcast/epidist/pull/231/files

By extension of the all_strata_newdata function I can see a path to achieving the postprocessing functionality that we intially agreed for 0.1.0 of the package. However, it'd be better if we don't have to write this functionality outselves. Therefore:

Plan

My plan going forward is to follow Plan A until it seems like there are roadblocks that make it not possible or too difficult for some reason. At that point I'll follow Plan B.

Plan A

• Use Ebola vignette #48 to do DDD exploration of the possibility to use tidybayes (or other packages) for all post-processing
• Figure out the extent to which Posterior prediction for lognormal #214 is required
  • Split Posterior prediction for lognormal #214 into the easier and harder part Function for pointwise log likelihood #240
• I think this workflow will also cover "summary" type functions, so no need for Create delay_summary function built on predict_delay_parameters #215

Plan B

• Return to Issue 213: Create functionality for passing in newdata #231 in order to implement the functionality that we need for 0.1.0
• To be determined: after how much effort on Plan A should I decide it's not possible?
  • I have quite a strong expectation that Plan A is possible
  • Postprocessing in some ways blocks figure creation

[seabbs]

Agree with the proposed roadmap

[athowes]

I have done some initial light investigation into tidybayes and it has updated me further towards Plan A being possible and desirable.

[seabbs]

There is also https://marginaleffects.com/

[seabbs]

Can we safely delete sensible-newdata-input or does it contain work we need?

[athowes]

Can we safely delete sensible-newdata-input or does it contain work we need?

We can delete it. I will do so.

[athowes]

I think that with the Ebola vignette up we have now demonstrated how to do the postprocessing tasks that we had scope / previously talked about, and this can be closed as completed.