covidseir
covidseir copied to clipboard
seananderson
Re #7 Add VoC component to seir fit
- Implemented changes to introduce VoC into fitting and projections.
- Updated version number and news to reflect changes
- Ran
R CMD checkwithout errors
Awesome. Thanks for putting this together. I'm wondering whether it would be more flexible to just have fit_seir() take a vector of relative reproduction ratios with one value per day (defaulting to a vector of 1s) rather than constructing the ramp within the Stan code. Then the user can make a ramp of any shape if needed and also accommodate more than one ramp if needed in the future. We could have a little helper function in R code to construct a linear (and/or logistic?) ramp that takes your named arguments. I could be convinced otherwise though.
i think with the helper function this would be really good, and it would have the advantage that if we knew we were going to be able to, for example, step up contact tracing and we knew something about the impact, we could model that as a change in transmission (to fudge doing some broader intervention).
i have found getting the lengths right for projections really frustrating at times though (and basically I never get it right so I always have to find some earlier call to the function to copy from.. duh... ) so i hope the helper function would sort out the length of the vector for me.
On Mon, 22 Mar 2021 at 21:25, Sean Anderson @.***> wrote:
Awesome. Thanks for putting this together. I'm wondering whether it would be more flexible to just have fit_seir() take a vector of relative reproduction ratios with one value per day (defaulting to a vector of 1s) rather than constructing the ramp within the Stan code. Then the user can make a ramp of any shape if needed and also accommodate more than one ramp if needed in the future. We could have a little helper function in R code to construct a linear (and/or logistic?) ramp that takes your named arguments. I could be convinced otherwise though.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/seananderson/covidseir/pull/9#issuecomment-804606838, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADD6EX6WLUIRJGPKMHXXPLLTFAJ2JANCNFSM4ZT7QHIA .
I know the projection vector stuff can be a pain and it’s partly a legacy of the fixed f values needing to start at some future date rather than the multipliers. Would it help if we just cut off the vector internally so you could pad it with extra values and not worry about it in the case of a constant multiplier after some date?
yes that'd be brilliant, then when I inevitably have that problem I can stick 10 or 100 values on the end and have done with it :)
On Mon, 22 Mar 2021 at 22:38, Sean Anderson @.***> wrote:
I know the projection vector stuff can be a pain and it’s partly a legacy of the fixed f values needing to start at some future date rather than the multipliers. Would it help if we just cut off the vector internally so you could pad it with extra values and not worry about it in the case of a constant multiplier after some date?
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/seananderson/covidseir/pull/9#issuecomment-804632241, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADD6EX6KOO42MZCT2OL3WQLTFASPHANCNFSM4ZT7QHIA .
Agree I like the approach of including as a vector for more flexibility and incorporating helper functions for the f-segments and sampling fractions. I had previously created some code to do this so can recycle it. The other advantage is that we could potentially fit the transmissibility vector to external data on VoC, convert that into a vector and then apply to the model. I'm imagining we would want a couple of functions something like:
create_segments_vector(start_date,segments,total_days,values=NULL)
where if values are specified (like for the sample fraction) then each segment will take that value otherwise it will number each segment incrementally.
The other function could be:
create_ramp(start_date,total_days,start_ramp,ramp_length,ramp_max)