NMsim

New Features

• The carry.out argument is better implemented in NMsim and NMreadSim(). carry.out is used to specify what columns from the input data set to include in the simulation results (while table.vars is used to control what variables to include in Nonmem $TABLE). This can be supplied to NMsim(), and NMreadSim() reapplies carry.out. It is useful when performing very large simulations where a wide input data set us memory demanding to merge onto the simulation results.
• NMreadSim() has a new argument, reread.tmp, which can be used to force a re-read of the Nonmem outputs (in contrast to the compressed simulation results). This is especially useful if changing carry.out.
• sampleCovs() can now sample with or without replacement - see new argument replace.
• sampleCovs() supports grouping for successful (parallel) simulations. This is normally only used if sampleCovs() is used in combination with NMsim_EBE(), so if sampling from observed subjects to simulate with their empirical Bayes' estimates. This will likely lead to duplicated ID values. sampleCovs() now returns a column that can be used for splitting (using say data.table::split()) the data set into data sets without duplicated ID values. sampleCovs() also offers the arguments col.idcgrp to name the returned column and idcgrp.redist to redistribute the split to obtain (close to) equally sized splits.
• The qsub execution path can now be specified using NMdata::NMdataConf(), like this:

``` r

NMdata::NMdataConf(path.qsub="/path/to/qsub")

```

This should normally not be necessary, but in some setups, qsub is not in the path to executables available to the shell that executes Nonmem, using qsub if sge=TRUE. On linux, you can most often find the qsub path by opening a terminal and executing

``` shell

which qsub