error with sdmTMB_cv when going back to models that previously ran

[jennybigman]

I'm getting an error with sdmTMB_cv(): 'Error: object does not inherit from class sdmTMB'

I was going back to some code from a few weeks ago that had no issues at the time and now it is throwing an error. I suspect this is due to package updates (similar to the error with s() - 'object of type 'symbol' is not subsettable' that I saw posted here and I had myself). I have tried installing the package from CRAN and from github with no success (this is what fixed the issue with using a smoother). In addition to my own code, I've tried the example code from the cross validation vignette (pasted below). Any ideas? Thanks so much!

data(pcod)
mesh <- make_mesh(pcod, c("X", "Y"), cutoff = 25)
pcod$fyear <- as.factor(pcod$year)

m_cv <- sdmTMB_cv(
density ~ 0 + s(depth_scaled) + fyear,
data = pcod,
mesh = mesh,
family = tweedie(link = "log"),
k_folds = 4
)

sanity(m_cv)

sessionInfo()
R version 4.3.1 (2023-06-16)
Platform: x86_64-apple-darwin20 (64-bit)
Running under: macOS Ventura 13.5.1

Matrix products: default
BLAS: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib
LAPACK: /Library/Frameworks/R.framework/Versions/4.3-x86_64/Resources/lib/libRlapack.dylib; LAPACK version 3.11.0

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

time zone: America/Los_Angeles
tzcode source: internal

attached base packages:
[1] stats graphics grDevices utils datasets methods
[7] base

other attached packages:
[1] sdmTMB_0.4.1

loaded via a namespace (and not attached):
[1] sandwich_3.0-2 utf8_1.2.4 future_1.33.0
[4] generics_0.1.3 class_7.3-22 fmesher_0.1.4
[7] KernSmooth_2.23-22 lattice_0.21-8 lme4_1.1-35
[10] listenv_0.9.0 digest_0.6.33 magrittr_2.0.3
[13] grid_4.3.1 estimability_1.4.1 mvtnorm_1.2-3
[16] Matrix_1.6-1 e1071_1.7-13 DBI_1.1.3
[19] survival_3.5-7 multcomp_1.4-25 mgcv_1.9-0
[22] fansi_1.0.5 TH.data_1.1-2 codetools_0.2-19
[25] cli_3.6.1 rlang_1.1.2 units_0.8-4
[28] parallelly_1.36.0 future.apply_1.11.0 splines_4.3.1
[31] parallel_4.3.1 tools_4.3.1 nloptr_2.0.3
[34] coda_0.19-4 minqa_1.2.6 dplyr_1.1.3
[37] boot_1.3-28.1 globals_0.16.2 assertthat_0.2.1
[40] vctrs_0.6.4 R6_2.5.1 zoo_1.8-12
[43] proxy_0.4-27 lifecycle_1.0.3 emmeans_1.8.9
[46] classInt_0.4-10 MASS_7.3-60 pkgconfig_2.0.3
[49] pillar_1.9.0 glue_1.6.2 Rcpp_1.0.11
[52] sf_1.0-14 tibble_3.2.1 tidyselect_1.2.0
[55] rstudioapi_0.15.0 xtable_1.8-4 nlme_3.1-163
[58] TMB_1.9.6 compiler_4.3.1 sp_2.1-1

[seananderson]

I'm not sure what's happening. Your initial post had:

as(, "dgTMatrix") is deprecated since Matrix 1.5-0; do as(., "TsparseMatrix") instead

which is curious, because at least in sdmTMB as(, "dgTMatrix") hasn't been used for a very long time.

Your Matrix package is one minor version behind: Matrix_1.6-1 vs. Matrix_1.6-1.1 and I know there was some breaking change in Matrix with respect to TMB recently. Can you try updating that, restarting your R session, and trying again?

Do you get the same error with a non-cross-validation example?

library(sdmTMB)
fit <- sdmTMB(
  density ~ 1,
  data = pcod_2011, spatial = "off",
  family = tweedie(link = "log")
)

Also, if the above doesn't work, can you try installing sdmTMB (and possibly TMB) from source?

install.packages("TMB", type = "source")
install.packages("sdmTMB", type = "source")

You'll need compilers installed first. If you don't have those, you can run xcode-select --install on a Mac in the Terminal first.

Then, again, restart R, and try the example.

[jennybigman]

@seananderson Thanks so much for the quick reply. I think uninstalling and reinstalling TMB may have done it? I tried uninstalling and reinstalling sdmTMB yesterday (both from source and CRAN) but it didn't work. This makes me think it was reinstalling TMB but I couldn't install it from source (tried and had zero-exit status) so installed it from CRAN.

Is this just a result of package versions? I want to forestall this in the future. Thanks!

[seananderson]

I assume you fell into a rare situation where your installed Matrix package version was different from the Matrix package version used to build the version of TMB you had installed and there was a breaking change in the Matrix API that affected TMB. The as(, "dgTMatrix") error might have been coming from your installed version of TMB itself.

[jennybigman]

Gotcha! Thanks again.