create_posterior_func: Posterior function pointer

In seroanalytics/serosolver: Inference Framework for Serological Data

Posterior function pointer

Description

Takes all of the input data/parameters and returns a function pointer. This function finds the posterior for a given set of input parameters (theta) and infection histories without needing to pass the data set back and forth. No example is provided for function_type=2, as this should only be called within serosolver

Usage

create_posterior_func(
  par_tab,
  antibody_data,
  antigenic_map = NULL,
  possible_exposure_times = NULL,
  prior_version = 2,
  solve_likelihood = TRUE,
  age_mask = NULL,
  measurement_bias = NULL,
  n_alive = NULL,
  function_type = 1,
  antibody_level_before_infection = FALSE,
  data_type = 1,
  biomarker_groups_weights = 1,
  start_level = "other",
  start_level_randomize = FALSE,
  demographics = NULL,
  demographic_groups = NULL,
  fixed_inf_hists = NULL,
  verbose = FALSE,
  ...
)

Arguments

par_tab	the parameter table controlling information such as bounds, initial values etc. See example_par_tab
antibody_data	the data frame of data to be fitted. Must have columns: group (index of group); individual (integer ID of individual); samples (numeric time of sample taken); virus (numeric time of when the virus was circulating); biomarker_group (integer of the observation group type, using a unique value for each distinctive type of observation underpinned by the same generative model); titre (integer of titre value against the given virus at that sampling time). See example_antibody_data
antigenic_map	(optional) a data frame of antigenic x and y coordinates. Must have column names: x_coord; y_coord; inf_times. See example_antigenic_map
possible_exposure_times	(optional) if no antigenic map is specified, this argument gives the vector of times at which individuals can be infected
prior_version	which infection history assumption version to use? See describe_priors for options. Can be 1, 2, 3 or 4
solve_likelihood	usually set to TRUE. If FALSE, does not solve the likelihood and instead just samples/solves based on the model prior
age_mask	see create_age_mask - a vector with one entry for each individual specifying the first epoch of circulation in which an individual could have been exposed
measurement_bias	if not NULL, then use these indices to specify which measurement bias parameter index corresponds to which time
n_alive	if not NULL, uses this as the number alive in a given year rather than calculating from the ages. This is needed if the number of alive individuals is known, but individual birth dates are not
function_type	integer specifying which version of this function to use. Specify 1 to give a posterior solving function; 2 to give the gibbs sampler for infection history proposals; otherwise just solves the titre model and returns predicted titres. NOTE that this is not the same as the attack rate prior argument, version!
data_type	integer or vector, with an entry for each unique data type in 'antibody_data'. Set to 1 for discrete data (e.g., fold dilution) or 2 for continuous (e.g., ELISA optical density).
biomarker_groups_weights	integer or vector, giving a factor to multiply the log-likelihood contribution of this data type towards the overall likelihood.
start_level	character, to tell the model how to treat initial antibody levels. This uses the observed data to either select starting values for each unique 'individual', 'biomarker_id' and 'biomarker_group' combination. See create_start_level_data. One of "min", "max", "mean", "median", or "full_random". Any other entry assumes all antibody starting levels are set to 0. Can also pass a tibble or data frame of starting levels matching the output of create_start_level_data.
start_level_randomize	if TRUE, and data is discretized, then sets the starting antibody level to a random value between floor(x) and floor(x) + 1. Does nothing if using continuous data.
demographics	if not NULL, then a tibble for each individual (1:n_indiv) giving demographic variable entries. Most importantly must include "birth" as the birth time. This is used if, for example, you have a stratification grouping in 'par_tab'
verbose	if TRUE, prints warning messages
...	other arguments to pass to the posterior solving function
titre_before_infection	TRUE/FALSE value. If TRUE, solves titre predictions, but gives the predicted titre at a given time point BEFORE any infection during that time occurs.

Value

a single function pointer that takes only pars and infection_histories as unnamed arguments. This function goes on to return a vector of posterior values for each individual

Examples

## Not run: 
data(example_par_tab)
data(example_antibody_data)
data(example_antigenic_map)
data(example_inf_hist)

## Simple model solving code. Output matches entries of example_antibody_data
model_func <- create_posterior_func(example_par_tab, example_antibody_data, example_antigenic_map, function_type = 3)
y <- model_func(example_par_tab$values, example_inf_hist)

## Solve likelihood
par_tab <- example_par_tab[example_par_tab$names != "phi",]
likelihood_func <- create_posterior_func(par_tab, example_antibody_data, example_antigenic_map, function_type = 1, prior_version = 2)
liks <- likelihood_func(par_tab$values, example_inf_hist)

## End(Not run)

seroanalytics/serosolver documentation built on Dec. 14, 2024, 5:33 a.m.