Do pre-fit calculations and checks
# S3 method for pk
do_prefit(obj, ...)The same `pk` object, but with a new element `prefit`, containing the results of pre-fit calculations and checks for each model and for the error model.
A `pk` object
Additional arguments. Not in use currently.
Caroline Ring
This function does the following:
Based on the error group in `pk_groups` and the pre-processed data, determines the number of residual standard deviations ("sigmas") hyperparameters to be estimated.
Determines which "sigma" hyperparameter corresponds to each observation in the data.
Calculates lower/upper bounds and starting guesses for each "sigma" hyperparameter
For each model in `stat_model`, calls its `params_fun`, the function that, based on the data, determines whether to optimize each model parameter, and calculates lower/upper bounds and starting guesses for each model parameter to be optimized. Only non-excluded observations are passed to each model's `params_fun`.
Lower bounds for each "sigma" hyperparameter are set to `sqrt(.Machine$double_eps)`.
Upper bounds for each "sigma" hyperparameter are calculated as the standard deviation of observations in the corresponding error SD group (see [combined_sd()]), with any specified transformations applied (dose-normalization and/or log10-transformation). If the combined SD is non-finite or less than the sigma lower bound, then the maximum concentration is used as an upper bound; if this still returns a non-finite value or a value less than the lower bound, then a constant value of 1000 is substituted.
The starting guess for each "sigma" hyperparameter is one-tenth of the upper bound.
If there are less detected observations than timepoints, or if there are parameters necessary for model fitting that have missing values, these models will not be fit.