PolicyGrid doesn't have schools_open = TRUE with r_distribution = "lnorm"

[wfmackey]

Because of the %implies% conditions, PolicyGrid doesn't allow for schools to be open with a non-Cauchy distribution:

https://github.com/grattan/covid19.model.sa2/blob/e4f174c83c60ff3582d08a5f7ab1264ddceb2492/results/2020-05-07/runner28.R#L46-L69

We need this in the Shiny app given people's aversion to Cauchy.

[wfmackey]

> PolicyGrid %>% 
+   filter(schools_open == FALSE,
+          r_distribution == "lnorm") %>% 
+   nrow()
[1] 0

[HughParsonage]

Damn that's a blunder. Consequence of a hacksaw approach to reducing model output runtime. We do have school_days_per_wk == 1 and r_distribution == lnorm as a second-best salvage.

[wfmackey]

lol yeah it's the consequence of a chain of like three %implies% functions.

Possible to get in the next couple of days?

[HughParsonage]

Options are:

1. Run about 100 policy scenarios done manually under the current ('flawed first day') model.
2. Do a quick fix for a 'attack rate' scenario model and run as many as possible.
3. Defer the model development and just run the skipped policies

[HughParsonage]

(I think in my defence that I thought almost all states were going to have at least one day a week back to school + wanted to exhibit cauchy distribution, but definitely didn't intend for so few lnorm distros.)

[wfmackey]

I'm not 100% sure what's in the pipeline -- can you list out what has been/will be sent to unimelb for running?

[HughParsonage]

First step is to reduce the size of the repo: that's underway on my computer.

Then I need to develop the package to change the infection parameters to reflect the attack rates for children/workplaces/households.

But the cluster can't do both versions at once, the new version should be done after the old one. and the new results won't be comparable to the old.

I haven't decided what grid to send to unimelb once the new version is working.