search

signaturescience / focustools

Forecasting COVID-19 in the US
https://signaturescience.github.io/focustools/
GNU General Public License v3.0
0 stars 0 forks source link

incident deaths lower than it should be / get hospitalization data #29

Closed stephenturner closed 3 years ago

stephenturner commented 3 years ago

Looking at most recent submissions incident deaths is lower than it should be. Perhaps an issue with lagged predictor.

Is this an issue from lagging cases around the holidays? We could test this by going back before thanksgiving/christmas/ny interruptions, running forecast pipelines and code as in #21, see if we get this same phenomenon occuring.

Is this something we could fix with other covariates? I took a stab at this in a very quick and dirty way as discussed here to use hospitalization data as a covariate but the problem was exacerbated. (this was a hack, could be cleaned up / tested again. This revealed lots of problems with hard-coded functionality throughout all the ts_ functions, in addition to what's mentioned in https://github.com/signaturescience/focustools/issues/26#issuecomment-760963472

image

related #7

stephenturner commented 3 years ago

note to self, more hosp data: https://cmu-delphi.github.io/delphi-epidata/api/covid_hosp.html

stephenturner commented 3 years ago

closing this for now, similar stuff tracked in #5 #26 (get_hosp should have been in #3)