Skip to contents

Run a simulation based on supplied parameters estimates, and combine into proper format for VPC

create_vpc_data.Rd

This rewrite of create_vpc_data() removes the Pharmpy/pharmr dependency from every step that runs before run_nlme(). The only Pharmpy touch is reading model$code as a string when a Pharmpy model is supplied. All subsequent manipulation of the NONMEM code (parameter updates, record removal/insertion, $EST -> $SIM conversion) and the input dataset (sanitisation of string-typed columns) is done in pure R, so the Python/R <-> reticulate serialisation cannot reject a value that NONMEM itself would accept. run_nlme() is still used to actually invoke nmfe, and post-processing of obs/sim is identical to the original.

Usage

create_vpc_data(
  fit = NULL,
  model = NULL,
  data = NULL,
  parameters = NULL,
  keep_columns = c(),
  n = 100,
  verbose = FALSE,
  id = NULL,
  use_pharmpy = TRUE,
  fix_input_heuristic = TRUE,
  seed = NULL
)

Arguments

fit

fit object from pharmr::run_modelfit(). Optional; alternative to model + parameters.

model

Either a Pharmpy model object (its $code is read), a character vector of NONMEM model code, or a path to a .mod file.

data

Path to the NONMEM-ready CSV. Optional; if missing, the function reads $DATA from the model code.

parameters

Named list of parameter inits, e.g. list(THETA_1 = 0.23, OMEGA_1_1 = 0.097, SIGMA_1_1 = 1). Optional.

keep_columns

character vector of column names in the original dataset to keep in the output.

n

number of simulation iterations.

verbose

verbose output?

id

run id used as folder name. Defaults to a random name.

use_pharmpy

retained for backward compatibility; controls whether PRED/TAD are transferred from obs to sim during post-processing.

fix_input_heuristic

If TRUE (default), detect the common pharmr::set_dataset() side-effect that rewrites $INPUT to the CSV headers, and rebind TIME -> TAFD and (for log-transform-both-sides models) DV -> LNDV. Set to FALSE to leave $INPUT untouched. The LTBS detection only scans the $ERROR record (not the whole model) so a LOG() call in a covariate transform doesn't trigger a false swap.

seed

integer seed passed to the simulation step. Default NULL draws a random seed per call (so repeated create_vpc_data() calls in one session aren't pinned to identical draws); supply a value for bit-reproducible runs.

Value

list with obs and sim data frames.