Release Notes for Pumas 2.6.1
Pumas
Features and significant changes
Bug fixes and minor improvements
- Fix formatting of
infer(..., SIR)
docstring - Fix the check for zero gradients in matrix domains used for variance-covariance matrices such that they successfully throw and report the correct index.
- Fix a bug where simulation of non-DiffEq models with covariate interpolation but without events would fail.
- Fix a bug that would cause an error to be thrown when fitting subjects that had ranges such as
0:1:24
provided as their obstimes. - Fix warning message for duplicated
BayesMCMC
options given in both the type and as keyword arguments tofit
. Previously, the options used was not correctly reflected in the warning. - Use provided parameters in
zero_randeffs
to construct the distributions of the random effects instead of the default values in the model. - Remove trailing whitespace in
show
of PumasEMModel - Remove trailing whitespace in
show
of PumasModel - Use
median
for initialization ofMvLogNormal
instead ofmean
for consistency withLogNormal
. - Use
mean
/mode
for initialization ofWishart
/InverseWishart
. - Fix
constantcoef
use incovariate_select
when passed asNamedTuple
. Previously, the input values would be ignored. Note, the preferred method is to input the parameters to be used in theparam
input and specify the constant parameters only as a tuple of symbols. - Allow for
MAP
infindinfluential
. Previously, the function would not allowMAP
due to a type restriction. - Fix a bug where
randeffs
input to functions would not allow for a different ordering of the random effects than the ordering found in the model. - Fix
DataFrame
constructor ofSimulatedObservations
when some subjects have events defined and others do not. - Fix a bug where the variables defined in
@vars
would not be in@observed
if@init
had been specified. - Fix
infer
withMarginalMCMC
andconstantcoef
. Previously, this would fail. - Fix
show
for likelihood approximations such asFOCE
,LaplaceI
etc. Previously, these would be way to verbose when showing the results offit
, when storing approximation information inDataFrame
s etc. - Error if variables defined in
@observed
and@derived
use names that are already used to define variables in other blocks. - Throw useful error when cholesky decomposition fails as part of
MvNormal
construction. For example, this could happen when callingsimobs
withinfer
input. - Add docstrings for
SimulatedPopulation
andSimulatedObservations
. - Fix
@delay
with integerobstimes
. - Fix
inspect
forJointMAP
. Previously, this would fail with a method error. - Fix
DataFrame
construction whenroute
is part of the model. Previously, theroute
would be filled in an inconsistent manner that caused output of ourDataFrame
constructors to produce content that could not be read byread_pumas
. - Fix CorrDomain constructors.
- Allow subjects to have different number of observations in
ByObservation
Bayesian cross-validation. - Fix a bug where parameters would not always be transformed back to the original parameter domain in some Bayesian diagnostics methods.
- Improve error message when no subject observations are in the model and fix some diagnostic functions when some observations are not in the model. This applies to situations where there are more observed variables in the subject than there are observed variables being modelled in the
@derived
block. - Fix parameter elimination in MTK systems with units.
- Fix
loglikelihood
for MAP and JointMAP to correctly exclude the contribution of the prior.
Optimal Design
- Fix specifying initial sampling times as a vector that doesn't contain vectors.
- Changed the optimization
verbose
option default totrue
. - Fixed the rendering of
design
docstring.
Features
Bug fixes and minor improvements
Pumas Utilities
Features
Bug fixes and minor improvements
Bioequivalence
Features
Bug fixes and minor improvements
- Fix rendering of markdown table in docstring of
pumas_be
. - Fix confidence interval in output table of reference scaled analyses
- Allow DataFrame inputs with abstractly typed columns