model_averaging() to make averaged predictions over
estimations performed from several models. Also exports
do_model_averaging() and compute_weights() for
low-level implementations.mapbayr_vpc() to make prediction-corrected visual
predictive checks (pcvpc) from a given model and dataset.
Control the independent variable (idv), and the
stratification on a numeric variable in the dataset
(stratify_on).hist() method (hist.mapbayests()) now
shows the values of eta-shrinkage in multiple subjects setting. New
argument shk to control the definition of shrinkage, either
based on the standard deviation ("sd") or on the variance
("var") (@LauraMvn, #192).mapbayr_plot() in order to plot results from tables
(data.frame). This is the function now called by
plot.mapbayests() internally. Can plot the results of
multiple estimation object (informed in the column “MODEL”), useful when
model averaging is performed. Argument MODEL_color to force
the color of a model on the plot.do_mapbayr_sims() as an engine to simulate from
estimation results. Experimental. Now mostly useful for internal or
programmatic uses, but might be extended in the future.augment.mapbayests(). Now easier to debug and
much more faster, especially when uncertainty on predictions is
required.$PARAM, however the definition
on “ETA” in $PARAM remains mandatory.$PARAM to avoid hazardous behaviours (@jbwoillard, #187).augment() now simulates at least 200 points
per individual. Fix a bug where delta was miscalculated and
strange-looking plots were sometimes generated (@LauraMvn, #191).mapbay_tab now has the same number of rows as
original data especially if it did not have observation rows (@LauraMvn, #193)..datehour column is updated
after realize_addl is being called (@LauraMvn, #194).etasrc specification.New adm_rows() and obs_rows() replace
and improve adm_lines() and obs_lines(),
respectively. See ?data_helpers for a comprehensive
documentation (#175).
In adm_rows(), obs_rows(), and
add_covariates(), the first argument x:
can be missing, which enables the creation of a new dataset from scratch.
accepts a data.frame, which enables the modification of a pre-existing dataset (#155). For example:
adm_rows(amt = 100, cmt = 1) %>%
obs_rows(time = 24, cmt = 2, DV = 0.123)still accepts an ‘mrgsolve’ model, which enables the modification of a dataset stored in the model arguments.
In adm_rows() and obs_rows(), new
argument .datehour in order to compute time as
function of date and hours provided as character. The value passed to
.datehour is parsed with parse_datehour() into
a date-time value (“POSIXct”). For example:
obs_rows(.datehour = c("2023/02/01 12:00", "2023/02/01 12:34"), DV = c(0.123, 0.456), cmt = 1)Data helpers now always rearrange data to fulfill the NM-TRAN compatibility and readability: filling missing covariate values with the last observation carried forward rule, relocation of NM-TRAN variables in the first positions etc…
LLOQ (lower limit of quantification, e.g. 0.22
mg/L) and BLQ (below limit of quantification, e.g. 1 or 0)
are in the data (@pchelle, #182).LLOQ and BLQ to the
data by yourself.LLOQ to the data by yourself:
BLQ will automatically be inferred from LLOQ
and DV.mapbayest(lloq = ) to automatically add the
LLOQ and BLQ variables in the data.mapbayest(), new argument lloq in order
to add a variable LLOQ to the data. For example:
mapbayest(model, data, lloq = 0.22).In mapbayest(), new argument select_eta
in order to select the numbers of the ETAs to estimate. Default are ETAs
related to an OMEGA not equal to zero. Non-selected ETAs will not be
estimated and returned equal to zero. This can be useful in order to
ignore the estimation of ETAs not of interest, e.g. in case of
inter-occasion variability or non-identifiability. For example:
mapbayest(model, data, select_eta = c(1,3))
(#170).
In mapbayest(), new argument lambda in
order to modify the weight of the priors in the Bayesian estimation.
This could be useful in order to flatten the priors with the objective
to favor observed data instead of a priori information. For
example: mapbayest(model, data, lambda = 0.1) to decrease
the weight of priors of a ten-fold (#174).
adm_rows.data.frame(),
adm_rows.missing(), adm_rows.mrgmod(),
obs_rows.data.frame(), obs_rows.missing(),
obs_rows.mrgmod(), and
add_covariates.data.frame(), as new methods for data
helpers.parse_datehour(), used to parse arguments passed
to .datehour in adm_rows() and
obs_rows().filter.mrgmod(), a method, wrapper around
dplyr::filter() for dataset stored in mrgsolve model object
(‘mrgmod’).adm_lines() and obs_lines().
Stop exporting adm_lines.mrgmod() and
obs_lines.mrgmod().see_data(). Was deprecated since 0.4.
Use get_data() instead.lubridate.arg.optim$select_eta element,
arg.ofv.fix$omega_inv now has the dimensions of the number
of ETAs selected.select_eta argument. This condition is not tested anymore
with check_mapbayr_model().print.mapbayests(), now only shows the estimated
ETAs.hist(), new argument select_eta in
order to select the ETAs to plot. Default are ETAs estimated with
mapbayest() (#167).eta() properly sorts vectors of length superior to 9
(#159).use_posterior() works if covariates had not been
defined in data (#160).hist(), get_phi() and
plot_phi(), ETAs are now properly re-ordered if they are
more than 9 (#165).tidyverse packages
(#171).ggplot2 >= 3.4.0.add_covariates() accepts empty arguments.mapbayest(), new argument ... in order
to fix compatibility issues, not used yet.summarise_phi() and bar_phi()
summarizes the comparison of estimation of ‘mapbayr’ and ‘NONMEM’
(i.e. classifies it as Excellent/Acceptable/Discordant) and graphically
represents it as a bar plot.eta() generates numerical values named
ETA1, ETA2, ETA3..., either from scratch, from a
pre-existing vector or from a ‘mrgsolve’ model object.plot(), PREDICTION = c("IPRED", "PRED")
controls to plot either “PRED”, “IPRED” or both (#113).add_covariates(), covariates is
relocated in last position, in the favor of ... which now
accepts covariate values. Calling
add_covariates(list(BW = 90)) will still works (with a
warning) for the sake of compatibility but will be deprecated. Instead,
just use add_covariates(BW = 90) or explicitly call
add_covariates(covariates = list(BW = 90)) if you want to
pass covariate values as a list (#156).get_eta(), output = "num" returns a
matrix if multiple IDs are available instead of an error message
(#145).postprocess.optim() and
postprocess.output(). Removed due to refactoring of
internal post-processing.adm_0_cmt().mapbayest(), reset is now a numeric and
drives the maximum allowed reset during optimization.check_mapbayr_model() now returns an error if a check
fails instead of a table that summarized the errors.check_mapbayr_model() now only checks critical points
and not suggested features.check_mapbayr_model() now explicitly forbids
IPRED, PRED and ETA1, ETA2...
(#148).mapbayest(output = ) (#134).stats::optim() if
method is ‘L-BFGS-B’ and minqa::newuoa() if method is
‘newuoa’. These replace optimx::optimx() (#136).optimx package.tibble package from dependency to
suggestion.vs_nonmem() and get_phi() works even if
covariance was missing/failing in mapbayests object (#126).pred() does not generate NaN if small
negative concentrations were predicted after log-transformation
(#140).pred() does not propagate ‘mrgsolve’ error when lag
time is longer than inter-dose interval at steady-state (#142).NA values in DV if
mdv == 0 is checked (#131).check_mapbayr_model() is now called before any use the
model inside mapbayest() (#149).DV if
error is exponential (#150).obs_lines(), mdv will be 1 if
DV is set to NA (#147).qmod, omega_invand all_cmt now
replace mrgsolve_model, omega.inv and
obs_cmt. For individual-related elements, idDV
replaces DVobs, data is removed,
idvaliddata and idcmt are added. This can have
an impact for the user since these elements are reported in the standard
output. However, it does not change the behaviour of
get_data().mapbayest(verbose = TRUE)
now only displays the messages related to optimization reset, and not
the progression of ID being optimized which is now controlled by
mapbayest(progress = TRUE).derivatives(), now replaced by
mapbayr:::h().mbrlib() and associated models. See the
“Model examples” section below.The example models system was totally re-thought around a new
function: exmodel(). It now embeds several models that were
used in the validation study, with a small corresponding dataset that
can be loaded automatically (the default). They are used in multiple
places inside the package, especially in tests and examples. More models
could be added in the future.
exmodel(). See the list of available models in
the documentation.exdata(), to load data only.mapbayest(progress = TRUE) displays a
progress bar with the number of the ID being optimized. #118 #28mapbayest(output = "eta") returns
only estimated ETA in order to skip most of post-processing steps.
#106eta_descr() now always returns a non-NA value even if
description is missing. #87do_compute_ofv(), a wrapper around
do.call(compute_ofv, ...).{progress}.{testthat}, {minqa}.
#120R CMD check.plot_phi() now plots correct values on the x-axis.
#108NA.
#114This version of mapbayr introduces several features that aim to express uncertainty around the point estimate. Please note that the results of these functions were not validated vs a gold-standard software such as NONMEM. This is why they are referred as “experimental features” in the following subsections. They are exported with the objective to ease their future validation, and to provide a very rough idea of the estimation uncertainty.
data slot in estimation object. Use
get_data() instead. #64$model@args$data is now always NULL in
the estimation object. It was carried out if the data was initially
passed with data_set() or built with
adm_lines()/obs_lines(). #64augment). Also use
recsort=3 to deal with steady-state administrations.
#85plot() are now directly passed to
augment().collapse_omega() function. (thanks @kylebaron)mapbayest(hessian = ) is used to
compute the hessian with stats::optimHess() by default. The
variance-covariance matrix is returned in a covariance slot
in the estimation object, and can be accessed with
get_cov().use_posterior(update_omega = TRUE) update the OMEGA matrix
with the covariance matrix, in order to simulate with uncertainty and
derive confidence intervals.plot(ci = TRUE) displays
approximate confidence intervals on predicted concentrations. Parameter
uncertainty is approximated with the covariance matrix. Confidence
interval computation relies on the delta approximation
(ci_method = "delta"), but can also be computed thanks to
simulations (see augment() documentation).get_cov(): function to get the covariance matrix of
estimation. #43get_phi(), read_nmphi(),
merge_phi() and plot_phi(): functions to
compare the estimations vs NONMEM. #55est001: an example mapbayests estimation
object. #94update_omega, update_cov, and
update_eta arguments to control what to update..zero_re default behavior now depends on
update_ arguments values.mapbayest(verbose = ) now mutes the message that
indicates a reset during optimization. #96opt.value inherited from
optimx. #95get_data() can now return a list of individual data
sets with output = "list". #64@annotated tag in model code,
especially for $PARAM and $CMT blocks.
#73adm_lines() and
obs_lines() don’t need the [ADM] and [OBS] tags in model
code anymore (yet strongly recommended, otherwise it errors
cleanly).testthat bugs due to upgrade of R and French
translation of warnings.plot() legend, due to new version of
ggplot2 3.3.4 #82NEWS.md file to track changes to the
package.NEWS.md