Demo usage of paramix in flepiMoP context

[pearsonca]

The discussion around {paramix} seemed generally positive, and indeed possibly to dovetail with some current efforts in flepiMoP oriented analyses. For folks that missed that conversation, briefly: {paramix} takes functionally-defined parameters (e.g. infection fatality ratio as a function of age) and creates the appropriate static parameters for aggregated compartments (e.g. age-blocks stratifying a population into pre-school, school age, working adults, and retirees). It also does the reverse, and imputes finer resolution outcomes (e.g. deaths by year, rather than block) from gross results.

But: talk is cheap - what are some concrete options to stress test the package for some real problems people are interested in?

• There's work in progress on the CMMID end to do a more thorough demo, but that'll continue to be a relatively lightweight model(l)ing application. If there are particular demonstrations people would expect to make the story convincing (more theory oriented than software-engineering), would love to hear about those.
• Is anyone interested in talking against an overall workflow diagram for flepiMoP and identifying points to consider inserting?
• Are there some active analyses that we can try out how it would fit in an otherwise existing pipeline?
• Any proverbial guinea pigs willing to try out the framework and then provide feedback on interface advice, sticky points in documentation, etc?

Recommend folks bounce some ideas around here, and then we can coordinate something specific via slack (and back-brief plan / outcomes here).

[pearsonca]

Also, thinking ahead to more generic use: #271 - suggests that paramix might want multi-way aggregation / disaggregation.

[twallema]

Hi Carl, I think that would indeed be a great addition in light of future generic use. I might want to give paramix a shot, I remember from the COVID-19 modeling days I had to a lot of static aggregated parameter to static aggregated parameter. F.i. I found a good literature estimate for the estimated fraction of asymptomatic COVID-19 infections,

f_a = [0.95, 0.90, 0.60] # asymptomatic stratified: 0-20, 20-60, 60+

and then I had to translate to my model's age groups, which were 0-6, 6-12, 12-18, 18-25, 25-35, ..., 75-85, 85+. Does paradox do that? I think @MacdonaldJoshuaCaleb is looking into defining contact matrices functionally, so cutting that 2D functional relationship into a matrix might be a good application of multi-way aggregation?