Models with smooth terms no longer run - "object of type 'symbol' is not subsettable"

[nilescbn]

The Error

I'm a very new user of sdmTMB but have successfully run a number of models on fisheries data. In going back to some models I was running successfully last week. I'm getting this error and have traced it to the inclusion of smooth terms:

Error in x[[3]] : object of type 'symbol' is not subsettable

I came back to the models because I was having convergence problems. I saw the note about the nlminb_loops. Increasing the number of loops to 2 worked on models without smooth terms.

I tried installing the development version of sdmTMB but that didn't solve it. I tried running the same model with smooth terms using gam() from mgcv and it runs. Same for brm() from brms.

Simulated Example

In terms of reproducible example, using this simulated data I get the same errror.

sim_data <- data.frame(X = runif(1000, 164.5, 415), Y = runif(1000, 4560, 5366), depth = runif(1000, 100, 300), haul_hours = runif(1000, 0.5, 5), catch = rnorm(1000, mean = 50, sd = 24) )

sim_mesh <- make_mesh(sim_data, xy_cols = c("X", "Y"), cutoff = 10)

This runs:

m1 <- sdmTMB(catch ~ depth + haul_hours, data = sim_data, mesh = sim_mesh, anisotropy = FALSE, spatial = "on")

This produces the error:

m1s <- sdmTMB(catch ~ s(depth) + s(haul_hours), data = sim_data, mesh = sim_mesh, anisotropy = FALSE, spatial = "on")

Traceback

This is from RStudio's Traceback feature.
7. deparse(x, 500L)
6. paste(deparse(x, 500L), collapse = collapse)
5. safe_deparse(x[[3]])
4. FUN(X[[i]], ...)
3. vapply(bars, function(x) safe_deparse(x[[3]]), "")
2. barnames(split_formula[[ii]]$reTrmFormulas)

1. sdmTMB(catch ~ s(depth) + s(haul_hours), data = sim_data, mesh = sim_mesh,
   anisotropy = FALSE, spatial = "on")

Package Updates

I updated a number of R packages since running the models successfully before, I wasn't paying close attention to which ones updated. I was getting a warning about a mismatch in versions of the Matrix package, but that was back when the models ran fine.

Session Info

Here is the session info from running the model on the simple, simulate data.

R version 4.2.3 (2023-03-15 ucrt)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 19044)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.utf8 LC_CTYPE=English_United States.utf8
[3] LC_MONETARY=English_United States.utf8 LC_NUMERIC=C
[5] LC_TIME=English_United States.utf8

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

other attached packages:
[1] sdmTMB_0.3.1.9000

loaded via a namespace (and not attached):
[1] Rcpp_1.0.11 INLA_23.04.24 nloptr_2.0.3 TMB_1.9.6
[5] glmmTMB_1.1.8 compiler_4.2.3 pillar_1.9.0 class_7.3-21
[9] tools_4.2.3 boot_1.3-28.1 lme4_1.1-34 tibble_3.2.1
[13] lifecycle_1.0.3 nlme_3.1-162 lattice_0.20-45 mgcv_1.8-42
[17] pkgconfig_2.0.3 rlang_1.1.1 Matrix_1.6-1.1 DBI_1.1.3
[21] cli_3.6.1 rstudioapi_0.15.0 parallel_4.2.3 mvtnorm_1.2-3
[25] xfun_0.40 e1071_1.7-13 coda_0.19-4 fmesher_0.1.2
[29] dplyr_1.1.3 knitr_1.44 generics_0.1.3 vctrs_0.6.3
[33] tidyselect_1.2.0 classInt_0.4-10 grid_4.2.3 glue_1.6.2
[37] sf_1.0-14 R6_2.5.1 fansi_1.0.5 sp_2.1-0
[41] minqa_1.2.6 magrittr_2.0.3 MASS_7.3-58.2 emmeans_1.8.8
[45] splines_4.2.3 units_0.8-4 assertthat_0.2.1 xtable_1.8-4
[49] numDeriv_2016.8-1.1 KernSmooth_2.23-20 utf8_1.2.3 estimability_1.4.1
[53] proxy_0.4-27

[seananderson]

Thanks for the reproducible example. That does run for me (other than the smoother penalization terms collapsing to zero, which is to be expected in this simple example). I've simplified it here to isolate the smoothers. Does this run for you? I believe I've seen that error before and it's from mgcv parsing the s() part, but I forget what aspect of the data or other loaded libraries that could be masking a function(?) was causing it. Try starting a fresh R session and running this:

library(sdmTMB)

set.seed(1)

sim_data <- data.frame(
  depth = runif(1000, 100, 300),
  haul_hours = runif(1000, 0.5, 5),
  catch = rnorm(1000, mean = 50, sd = 24)
)

m1 <- sdmTMB(
  catch ~ depth + haul_hours,
  data = sim_data,
  spatial = "off"
)
m1
#> Model fit by ML ['sdmTMB']
#> Formula: catch ~ depth + haul_hours
#> Mesh: NULL (isotropic covariance)
#> Data: sim_data
#> Family: gaussian(link = 'identity')
#>  
#>             coef.est coef.se
#> (Intercept)    49.54    3.30
#> depth          -0.01    0.01
#> haul_hours      0.86    0.59
#> 
#> Dispersion parameter: 24.91
#> ML criterion at convergence: 4634.250
#> 
#> See ?tidy.sdmTMB to extract these values as a data frame.

m1s <- sdmTMB(
  catch ~ s(depth) + s(haul_hours),
  data = sim_data,
  spatial = "off"
)

m1s
#> Model fit by ML ['sdmTMB']
#> Formula: catch ~ s(depth) + s(haul_hours)
#> Mesh: NULL (isotropic covariance)
#> Data: sim_data
#> Family: gaussian(link = 'identity')
#>  
#>             coef.est coef.se
#> (Intercept)    49.61    0.79
#> sdepth          0.65    0.79
#> shaul_hours    -1.15    0.79
#> 
#> Smooth terms:
#>                 Std. Dev.
#> sds(depth)              0
#> sds(haul_hours)         0
#> 
#> Dispersion parameter: 24.91
#> ML criterion at convergence: 4634.250
#> 
#> See ?tidy.sdmTMB to extract these values as a data frame.
#> 
#> **Possible issues detected! Check output of sanity().**


packageVersion("mgcv")
#> [1] '1.9.0'
packageVersion("sdmTMB")
#> [1] '0.3.1.9000'

^{Created on 2023-10-10 with reprex v2.0.2}

[nilescbn]

Thank you for the quick reply, much appreciated.

Unfortunately, I get the same error when trying to run the "m1s" model.

I noticed my version of mgcv was lower than yours, and it wasn't showing up as having a new version available. But manually upgraded to v.1.9.0 and it didn't help.

I'm guessing this is related to the package updates I did in the last few days. I will keep researching it.

Thanks again.

[seananderson]

Ah, it's because of the glmmTMB update to 1.1.8 and some change to glmmTMB::splitForm(). I'll figure this out and push a fix. Thanks for flagging it!

[nilescbn]

Thank you for tracking that down so quickly!

[seananderson]

Should be fixed now with the latest commit on GitHub. Let me know if not.

[seananderson]

Sorry, my fix broke other things. I hope the latest will pass all checks.