Pre-compute exposure probabilities

Checks if the provided exposure model can be solved in advance of the main simulation in fewer function calls than would be expected in the main simulation. Note that there is some overhead to checking for pre-computation, so this may not be quicker and may actually slow down the simulation. However, in some cases where many individuals share identical exposure probabilities, running this function can lead to massive speed ups.

Usage

precomputation_checks(
  N,
  times,
  exposure_ids,
  groups,
  exposure_model,
  foe_pars,
  demography,
  VERBOSE,
  check_correct = FALSE,
  ...
)

Arguments

N

integer number of individuals in the simulation

times

vector of time periods in the simulation

exposure_ids

integer vector of unique exposure IDs matching the biomarker_map

groups

integer vector of unique groups matching entries in demography

exposure_model

function pointer to the desired exposure model

foe_pars

object (usually an array, but may be a list or other class) corresponding to the exposure_model inputs

demography

tibble of demography variables used for the simulation

VERBOSE

(optional) if specified, an integer giving the frequency at which updates should be printed, as well as dictating whether progress messages are printed. Defaults to NULL

check_correct

if TRUE, computes the entire exposure probability array as would be done in runserosim with no pre-computation. This is usually quite slow, but can be used to check that the pre-computed exposure probability array is correct.

...

other inputs to the exposure_model

Value

a list containing: 1) a boolean set to TRUE if precomputation was successful; 2) a 3D array matching the dimensions of N, times and exposure_ids giving the individual probability of exposure in each time period

Examples

times <- seq(1,100,by=1)
N <- 100
n_exposure_ids <- 2
n_groups <- 2
demography <- generate_pop_demography(N=N, times=times,prob_removal=0,
aux=list(Group=list(name="group",options=c(1,2),proportion=c(0.5,0.5))))
#> Joining with `by = join_by(i)`
foe_pars <- array(runif(n_exposure_ids*length(times)*n_groups),dim=c(n_groups,
length(times),n_exposure_ids))
res <- precomputation_checks(N, times=times, exposure_ids=1:2,groups=1:2,
exposure_model_simple_FOE,foe_pars=foe_pars, demography=demography,VERBOSE=10,check_correct=TRUE)
#> Run time can be reduced by pre-computation! Pre-computation would require a maximum of  501  calls to exposure_model as opposed to  20000 
#> 
#> Checking if exposure model can be vectorized...
#> 
#> Solving the model for one individual without vectorization would take:  9.822845e-05  seconds
#> 
#> Solving the model for one individual with vectorization would take:  1.096725e-05  seconds
#> 
#> Exposure model can be vectorized!
#> 
#> Precomputing exposure probabilities...
#> 
#> Checking if pre-computed exposure probabilities are the same as calculating individually...
#> 
#> Individual:  10 
#> 
#> Individual:  20 
#> 
#> Individual:  30 
#> 
#> Individual:  40 
#> 
#> Individual:  50 
#> 
#> Individual:  60 
#> 
#> Individual:  70 
#> 
#> Individual:  80 
#> 
#> Individual:  90 
#> 
#> Individual:  100 
#> 
#> Pre-computed exposure probabilities are all correct
#>