notes/diagnose_repfit.R

In bbolker/McMasterPandemic: Pandemic Model

library(McMasterPandemic)
## devtools::load_all("../../McMasterPandemic")
library(tidyverse)
library(furrr)

p1 <- fix_pars(read_params("ICU1.csv"))
summary(p1)
p2 <- update(p1, obs_disp=1, proc_disp=0, zeta=5)
set.seed(101)
r1 <- run_sim(p2, stoch=c(obs=TRUE, proc=TRUE), end_date="2020-05-31")
plot(r1, log=TRUE)

dd_r <- r1 %>% select(date,report) %>% pivot() %>% mutate(value = ifelse(is.na(value),0,value))
dd_rh <- r1 %>% select(date,report,hosp) %>% pivot() %>% mutate(value = ifelse(is.na(value),0,value))
opt_pars <- list(params=c(log_E0=4, log_beta0=-1), log_nb_disp=c(report=1, junk=1))
opt_pars <- list(params=c(log_E0=4, log_beta0=-1), log_nb_disp=c(report=1))
unlist(opt_pars)

## opt_pars2 <- list(params=c(log_E0=4, log_beta0=-1), log_nb_disp=c(report=1,1))
##  params.log_E0   params.log_beta0 log_nb_disp.report 
##          4                 -1                  1 

dd_r  <- r1 %>% select(date,report) %>% pivot() %>% na.omit()
dd_rh <- r1 %>% select(date,report,hosp) %>% pivot() %>% na.omit()

range(dd_r$date)
range(dd_rh$date) ## with hosp is longer: zero values

c_rh <- calibrate_comb(params=p2, use_phenomhet=FALSE, debug_plot=TRUE,
                       data=dd_rh, use_DEoptim=FALSE,
                       use_spline=FALSE)
## re-run calibrate() [NOT calibrate_comb]
c_r <- update(c_rh, data=dd_r)
## re-run calibrate_comb (only change is data)
c_r2 <- calibrate_comb(params=p2, use_phenomhet=FALSE, debug_plot=TRUE,
                       debug=TRUE,
                       data=dd_r, use_DEoptim=FALSE,
                       ## likelihood only makes sense
                       use_spline=FALSE, opt_pars=opt_pars)

## inspect components
t1 <- c_rh$forecast_args$time_args
t2 <- c_r2$forecast_args$time_args
cmp <- map2(t1, t2, ~all.equal(.x,.y))
cmp[!map_lgl(cmp,isTRUE)]
## start_date, time_args, opt_pars

debug(calibrate_comb)

range(t1$X_date)  ## starts 03-20
range(t2$X_date)  ## starts 03-29