nlr and wnl5 no longer fail with "cannot change value of locked binding" when the user's workspace happens to contain an object named e.
nlr recomputes the predictions at the final estimates, so that Prediction, Residual, the run test and dx no longer use the values left behind by the last hessian evaluation.
nlr no longer stops with an uninformative error when the estimated error variance is very small; a non-positive variance is now penalised instead of producing NaN.
nlr with Error="NOSG" works; the count of free error variance parameters was computed from a reversed sequence. Error="NOSG" is now documented.
ExpandDH no longer overwrites the CMT of other dose records that fall inside an infusion interval, which sent a bolus into the wrong compartment.
ExpandDH copies ID and covariates into the row it inserts at the end of an infusion, instead of setting every column to zero. The returned table is always sorted by TIME.
hSkew finds the predictors from the model data rather than from terms, which failed or silently used same-named objects from the workspace.
nComp handles Ka equal to an eigenvalue, which gave NaN.
cmpChi returns 1 rather than 0 for two models of equal fit, and for a larger model that fits worse.
pComp returns quietly when there is no rate to draw, and detects elimination by compartment number instead of compartment count.
SolComp3 is robust to round-off near repeated roots.
pProf draws the likelihood interval labels slightly above the curve minimum, so that they no longer overlap the profile curve.
numDeriv moved from Depends to Imports, since only hessian is used. It is no longer attached to the search path by library(wnl); load it yourself if your own code calls it.
Contact field added for bug reports.
Minor bugs/errata fix and refactored using Claude Code.
Default k value is adjusted to be more conservative.
A bug in ExpandDH is fixed.
logk in nlr adjusted
g2inv utility function bug with p=0 fixed.
g2inv utility function is revised to handle non-square matrix.
Default k value in nlr() is adjusted.
If hessian matrix is singular, g2inv() is used instead of solve().
If k option in nlr() is missing, recommended k for conf.level is used.
pProf() improved and exposed
EnvObj() introduced
pProf() bug fixed for NaNs
hSkew() function added
nlr function returns 1/8 likelihood interval which is similar to 95 percent CI.
pProf() internal function plots likelihood profile and interval for the last nlr() run.
Minor beautification of the script - more spaces
dx made compatible for rmarkdown; it further checks the default graphic parameter.
pComp option Col for compartment color added.
pComp option asp=1 should be provided if you need.
pComp for plotting compartment model diagram is added.
qr.solve is used instead of solve.
Documents enhanced.
nlr function gained Error="S" (scale) option. This enables GLS.
DAT table changed.
Analytical solutions of compartmental models added.
Combined error with fixed additive variance model added.
Errata in manual corrected.
dx for simplest diagnostic plot is added.
nlr and wnl5 returns more.
nlr checks arguments more.
Better p-value from cmpChi.
nlr returns more information.
cmpChi is newly introduced.
Method argument added.
Secondary argument added.
wnl5 added.
Changed interface.
First version