Parse parameter labels from the model object. Currently this parses labels from comments in the control stream, However it will be extended to parse labels from the model YAML file as well. The syntax for the labeling is described in "Details" below.
param_labels(.mod, ...)
# S3 method for bbi_nonmem_model
param_labels(.mod, ...)
# S3 method for character
param_labels(.mod, ...)generic model input
arguments passed through (currently to nowhere)
Note that param_labels() will not return indices for the parameters, though they can be added with param_labels() %>% apply_indices().
See the apply_indices() documentation for more details.
The syntax for parsing labels from comments is inherited from the "Census" specification
that was also used in the tidynm and mrgtable packages. The syntax is as follows,
with everything after the comment character ; parsed into the label.
<nonmem_inits> ; [directive] <Parameter name <LOGD>>
Inits are the NONMEM initial values, following NONMEM specs
Directives include:
For fixed effects the directives are translated to units, i.e., L/h
For variance components (and only variance components) the directive is parsed into the type column
and can be used for post-processing transformations. Some common types include:
[A] is the assumed directive and it yields standard deviation / correlation. Typically used when the random component has a linear relationship to the typical value.
[P] yields Coefficient of Variation. Commonly transformed with \(CV_lognormal = sqrt( exp(omega^2)-1 ) \times 100\) for omegas, and the value of omega from the correlation matrix (i.e., the SD) for sigmas. Typically used with exponentiated random effects or proportional residual error.
[R] indicates that this is an item that should be read from the correlation matrix as opposed to the covariance matrix. I.e., standard deviation and correlation of the parameter on the raw scale will be reported.
[C] indicates that this item should be read from the covariance matrix, i.e., variance and covariance of the parameter on the raw scale will be reported.
Parameter name is parsed into the label column, and is commonly formatted to be run through a LaTex parser downstream.
LOGD is a flag to indicate that the parameter is modeled on the log
scale (i.e., \(e^\theta\)) and should be exponentiated for reporting.
If so, the standard error for the parameter should adjusted via the delta method.
NOTE: this is only a convention, such that downstream code can look for "LOGD" in the label column
and transform the estimates accordingly. None of that processing is done automatically by this function.
param_labels(bbi_nonmem_model): Takes a bbi_nonmem_model object
param_labels(character): Takes a character scalar of the raw control stream with newline character separating lines