Project: Addressing Bugs in nlme

[hturner]

Discussed in #4

^{Originally posted by giscus[bot] June 26, 2023}

R Project Sprint 2023 - Addressing Bugs in nlme

Addressing open bugs on the recommended package nlme for (non)linear mixed effect models.

https://contributor.r-project.org/r-project-sprint-2023/projects/nlme-bugs/

[MetaEntropy]

I am interested to contribute to this project.

[aitap]

I'll be happy to help with bugs that involve compiled C/Fortran code.

[aitap]

Bug ID	Topic	Outcome
18505	behaviour changing with order of group variable
17884	new confint.lme method for fixed effects only
17988	what to return for corARMA with p = q = 0
18472	correct behaviour of gls
17712	conditions when algorithm should use decomposition
18285	how to handle rank deficiency
17961	appropriate df to use in tests
17866	Initializing corStruct outside gls call gives different results
16890	intervals.lmList does not work for with pool = FALSE
18192, 17227	crash in gnls when using spatial correlation structures	Traced by Mikko Korpela and Benjamin Tyner to originate in nlme:::Dim.corStruct where val[["start"]] is calculated to contain offsets into c(loss_gradient, residuals) that result in corStruct_recalc → mult_mat overflowing a buffer on the heap. It seems that the right values would have been c(0, cumsum(val[["len"]])[-val[["M"]]]), or maybe even the original value of val[['start'] = match(levels(groups), groups) - 1L), but the code has been there since the oldest version of nlme present on CRAN, so it's hard to argue in favour of changing it.
16864	lme fails to warn when data are near-redundant
16975	REML estimation incorrect with fixed sigma
17236	Correct df for intercept in lme
18433	infinite loop dependent on random seed
16806	method fails for individuals with single observations	patch available as of 7th of October