This function defines an estimation task allowing to specify the estimation algorithm, estimation options, and whether standard errors should be obtained.
tsk_estimation(algorithm = "foce", se = FALSE, target_options = list())
| algorithm | The estimation algorithm to use for the task ("foce", "foce-inter", "foce-no-inter", "fo", "imp", "saem") |
|---|---|
| se | Whether to calculate parameter uncertainties |
| target_options | List of additional options that should be passed to NONMEM |
A building block of type 'estimation_task'
Tasks are building blocks that allow to specify what a model should
“do”. Like other model building blocks, they can be combined using
the + operator. However, they should not be added to a model but
rather provided via the tasks= argument to the render function, e.g.,
render(m, tasks = tsk_estimation() + tsk_output_xpose4())
Estimation tasks provide details on the parameter estimation process, in terms of estimation algorithm, estimation options and whether standard errors should be obtained.
The algorithm argument allows to select the estimation algorithm among the following options:
| foce | First-order conditional estimation with interaction detection |
| foce-inter | First-order conditional estimation with interaction |
| foce-no-inter | First-order conditional estimation without interaction |
| fo | First-order estimation |
| imp | Importance sampling |
| saem | Stochastic approximation expectation maximization |
The default algorithm "foce" detects whether the observation model includes an epsilon-eta interaction and includes the INTERACTION option accordingly.
The foce-inter option forces the use of the INTERACTION argument independent of the residual error model, foce-no-inter enforces no interaction.
Each algorithm includes a set of default options that the package authors consider sensible defaults (for example MAXEVAL=999999 for FOCE). These defaults can be
overwritten using the target_options= argument which is described below.
The se= argument allows to request the calculation of parameter standard errors. When standard errors are requested (se=TRUE) it will results
in the inclusion of the $COVARIANCE record in the generated control stream.
The target_options= argument provides a mechanism to specify additional estimation options for the selected algorithm. The options should be provided
as a list, e.g.,
tsk_estimation(algorithm = "foce", target_options = list(mceta=100))
The provided options are passed verbatim to the target tool and not checked by assemblerr for correctness.
The target_options= argument
A sequence of estimation tasks can be specified in assemblerr by combining multiple estimations, for example
render(m, tasks = tsk_estimation("foce") + tsk_estimation("imp"))
will create model code that contains an FOCE as well as an importance sampling estimation step.
Other tasks:
tsk_output()
m <- model() + input_variable("dose") + prm_log_normal("emax", median = 10, var_log = 0.09) + prm_log_normal("ed50", median = 50, var_log = 0.09) + algebraic(effect~emax*dose/(ed50 + dose)) + obs_proportional(~effect, var_prop = 1) # add estimation task using importance sampling, covariance step # and user-defined ISAMPLE option render( model = m, tasks = tsk_estimation( algorithm = "imp", se = TRUE, target_options = list(isample=1000) ) )#> $PROBLEM #> $INPUT DOSE ID DV #> $DATA data.csv IGNORE=@ #> $PRED #> EMAX = THETA(1) * EXP(ETA(1)) #> ED50 = THETA(2) * EXP(ETA(2)) #> EFFECT = EMAX * DOSE/(ED50 + DOSE) #> Y = EFFECT + EFFECT * EPS(1) #> $ESTIMATION METHOD=IMP AUTO=1 ISAMPLE=1000 #> $COVARIANCE PRINT=E #> $THETA (0, 10, Inf) ; POP_EMAX #> $THETA (0, 50, Inf) ; POP_ED50 #> $OMEGA 0.09 ; IIV_EMAX #> $OMEGA 0.09 ; IIV_ED50 #> $SIGMA 1; RUV_PROP