Iteration printing / parameter history over a loaded likelihood

The standard nlmixr2est iteration table (the shared `scale.h` machinery behind every estimation method's printout) exposed for an EXTERNAL sampler driving [foceiLikLoad()]'s conditional likelihood – e.g. nlmixr2stan's `est="stan"`. `foceiLikIterPrintStart()` prints the header and arms the row entry (entry 8 of the FOCEi C API table, `nlmixr2FoceiIterPrintRow`), which the sampler's gradient evaluations call from C; every `every`-th call prints a row AND records it, so the returned history holds exactly the printed rows. `foceiLikIterPrintEnd()` prints the closing line and returns the history as a `parHistData`-style data frame.

Usage

foceiLikIterPrintStart(
  every,
  initPar,
  names,
  iterPrintControl = NULL,
  xform = NULL
)

foceiLikIterPrintEnd()

Arguments

every

print/record cadence in row-entry calls (0 arms nothing; the row entry then records nothing and returns immediately)

initPar

display vector at the starting point (defines the width)

names

column names, same length as `initPar`

iterPrintControl

optional list applied via the shared `scaleApplyIterPrintControl` (e.g. `list(useColor=FALSE, printNcol=6L)`)

xform

optional back-transform list (the `.iterPrintXParFromUi` shape) driving the `X` row

Value

`foceiLikIterPrintStart()`: invisibly `NULL`; `foceiLikIterPrintEnd()`: the recorded history data frame (or `NULL` when printing was never started)

Details

The display vector is CALLER-defined, not the internal focei parameter vector: pass natural-scale thetas plus (optionally) the sampler's current actual omega entries, named `om.<eta>` for variances and `cov.<eta1>.<eta2>` for covariances. (The internal vector's omega tail is `chol(Omega^-1)` in the `diagXform` parameterization, which a Bayesian sampler neither uses nor understands, so it is NOT printed.)

Author

Matthew L. Fidler