Release Notes for Pumas 2.4.1


Bug fixes and minor improvements

  • Fix handling of the nnodes keyword in vpc with stratification such that nodes are chosen by stratum, and the requested number of nodes is always used unless it's larger than the number of unique values of the independent variable in the stratum.
  • Fix a bug where it was not possible to set variance information to zero to remove the presence of a given random effect.
  • Correctly pass down the ensemblealg information in infer, vcov and stderror. The information was previously ignored.
  • Fix some show methods when there is no dynamical model.
  • Fill dosing columns with 0s instead of missing when creating a DataFrame from Subjects and diagnostics.
  • Mention next observation carried backward (NOCB) and last observation carried forward (LOCF) when explaining how to set covariate interpolation direction in the read_pumas docstring.
  • Fix the check for a combination of ss and ii that is not allowed in read_pumas.
  • Throw an error if the dose control parameters used in the @dosecontrol do not match the names of the dynamical variables in @dynamics.
  • Throw useful errors when there are no observations for a given subject in various functions that have to calculate empirical bayes estimates.
  • Ignore CompositeException and TaskExceptions during a fit if they are due to numerical errors. This allows the the optimization procedures built into the algorithm to continue if they end up with a numerically unstable parameter proposal.
  • Fix a bug where fit would fail if the first subject of a population had all missing values for one of multiple dependent variables.
  • Fix a bug where @nca use in the @derived block would fail with a DimensionMismatch error.


Bug fixes and minor improvements

  • Switch to LN_COBYLA solver when fitting linear mixed effects model to avoid saddlepoint solutions for some models with missing data.