Sample model parameters using `mvrnorm` or the `simpar` package
Source:R/samplePars.R
samplePars.RdSample model parameters using `mvrnorm` or the `simpar` package
Arguments
- file.mod
Path to model control stream. Will be used for both `NMreadExt()` and `NMreadCov()`, and extension will automatically be replaced by `.ext` and `.cov`.
- nsims
Number of sets of parameter values to generate. Passed to `simpar`.
- method
The sampling method. Options are "mvrnorm" and "simpar". Each have pros and cons. Notice that both methods are fully automated as long as ".ext" and ".cov" files are available from model estimation.
- seed.R
seed value passed to set.seed().
- format
The returned data set format "ext" (default) or "wide". "ext" is a long-format, similar to what `NMdata::NMreadExt()` returns.
- as.fun
The default is to return data as a data.frame. Pass a function (say `tibble::as_tibble`) in as.fun to convert to something else. If data.tables are wanted, use as.fun="data.table". The default can be configured using NMdataConf.
Details
samplePars() uses internal methods to sample using mvrnorm or simpar. Also be aware of NMsim_NWPRI which is based on the Nonmem-internal NWPRI subroutine. NMsim_NWPRI is much faster to execute. Simulation with paramater uncertainty on variance components (`OMEGA` and `SIGMA`) is only reliable starting from Nonmem 7.6.0.
mvrorm: The multivariate normal distribution does not ensure non-negative variances. Negative variances are not allowed and can not be simulated. To avoid this, `method=mvrnorm` truncates negative variance diagonal elements at zero.
simpar: simpar must be installed.
Please refer to publications and vignettes for more information on sampling methods.