Skip to contents

Generic function for running a pharmpy tool, like bootstrap, or modelsearch. A separate function is available for fit()

call_pharmpy_tool.Rd

Generic function for running a pharmpy tool, like bootstrap, or modelsearch. A separate function is available for fit()

Usage

call_pharmpy_tool(
  id,
  model = NULL,
  results = NULL,
  tool = NULL,
  folder = NULL,
  clean = TRUE,
  verbose = TRUE,
  force = FALSE,
  options = list(),
  remove_tables = TRUE,
  keep = NULL,
  uppercase_mfl = TRUE
)

Arguments

id

model id. Optional. If not specified, will generate random modelfit id. The id will be used to create the run folder.

model

Pharmpy model object, preferably created using create_model().

results

TODO

tool

TODO

folder

TODO

clean

if one or more run folders exists for the tool, do we want to remove them first?

verbose

verbose output?

force

TODO

options

list of arguments pass on to tool as argument. Documentation for available arguments for each Pharmpy tool can be found here: https://pharmpy.github.io/latest/mfl.html. For tool = "structsearch" with type = "tmdd", the pharmr.extra-specific kd element (target dissociation constant seed, in concentration units) seeds the QSS candidates' target parameters via seed_tmdd_results() and is not forwarded to Pharmpy.

remove_tables

if TRUE (default), removes all $TABLE records from the model before passing it to the Pharmpy tool.

keep

NONMEM only: a folder to keep a record of the search in. When set, the run folder's base fit (run.mod, run.lst, final.mod, stdout, stderr and the .ext / .shk / .cor / .cov files at its root), the tool's own results.csv / results.json summaries, the final_<tool>.mod written here, and the candidate the search settled on (<tool>N/models/final, or models/sim for tool = "simulation") are copied there at the same relative path once the tool has returned, whether it succeeded or aborted. The run folder is then removed with everything else in it: the datasets and the one folder per candidate fit a search leaves behind. Relative paths resolve against the working directory; the folder may already exist, and only files of the same name are overwritten. A run that aborts before the tool writes anything leaves an already existing run folder in place, base fit and all. NULL (default) leaves the run folder in place. Ignored, with a warning, for nlmixr-format models, whose candidate fits are not a NONMEM record.

uppercase_mfl

if TRUE (default), uppercases the model's $INPUT / datainfo / dataset column names and the options$search_space string before calling the Pharmpy tool. Works around Pharmpy's MFL parser, which unconditionally uppercases every identifier in a search_space and then fails the case-sensitive lookup against datainfo (see https://github.com/pharmpy/pharmpy/issues/4576). Set to FALSE to disable.

Value

fit object

Examples

if (FALSE) { # \dontrun{
# Run 200 bootstrap samples on a fitted model
bs <- call_pharmpy_tool(
  id      = "run1",
  model   = model,
  results = results,
  tool    = "bootstrap",
  options = list(samples = 200)
)

# Inspect parameter estimates (one row per sample)
head(as.data.frame(bs$parameter_estimates))

# Plot distributions and overlay original estimates
orig <- setNames(results$parameter_estimates$estimates,
                 results$parameter_estimates$parameter)
plot_bootstrap(bs, original_estimates = orig)
} # }