Issue post-processing model with with Cox family type

[colinorourke]

I've been working with a model fitted using the "cox" family type. It works great, but most functions such are predict, log_lik, conditional_effects, etc. don't seem to work with models of this type. I'm not sure, but this seems to be because the bhaz_basis_matrix function is checking for objects of class mSpline, but the class of an object create by the mSpline function from the splines2 package seems actually to be MSpline.

A reprex for this error follows:

library(brms)
#> Loading required package: Rcpp
#> Loading 'brms' package (version 2.20.4). Useful instructions
#> can be found by typing help('brms'). A more detailed introduction
#> to the package is available through vignette('brms_overview').
#> 
#> Attaching package: 'brms'
#> The following object is masked from 'package:stats':
#> 
#>     ar

# load data from survival package
dt = survival::bladder1

# fit survival model in brms
surv_mod = brm(
  stop | cens(status == 0) ~ size + number,
  data = dt,
  family = cox(link = "log"))

# try to get pointwise log-likelihoods, which result in error 
log_lik(surv_mod)
#> Error in bhaz_basis_matrix(y, args, basis = bs): inherits(basis, "mSpline") is not TRUE

^{Created on 2024-01-06 with reprex v2.0.2}

[paul-buerkner]

Strange. Which version of the splines2 package are you using? Perhaps they have changed the object class names?

[paul-buerkner]

Okay, I can confirm that problem starts with splines2 version > 0.5.0. In the previous version I had (0.4.7) things were just working. Changing object class names is really not a good thing to do in any package... I will fix it in brms soon.

[paul-buerkner]

Should now be fixed.

[nstucke]

Hi @paul-buerkner! First, thank you for your work on this!

I have run into an issue similar to what's been described above. I am trying to calculate a Bayes Factor comparing the likelihood of the data under a null vs. alternative hypothesis using 2 Cox (survival) models. Using the bayes_factor() function, I am seeing the same error message described by @colinorourke above:

Error in bhaz_basis_matrix(y, args, basis = bs) :
inherits(basis, "mSpline") is not TRUE

A similar reprex to the above reproduces this error:

# load data from survival package
data <- survival::bladder1

# fit survival model in brms
surv_null_mod <- brm(
  stop | cens(status == 0) ~ size + number,
  data = data,
  family = cox(link = "log"))

surv_alt_mod <- brm(
  stop | cens(status == 0) ~ number,
  data = data,
  family = cox(link = "log"))

# try to calculate bayes factor...error 
bayes_factor(surv_alt_mod, surv_null_mod)

All of my packages/software are up to date:
splines2: version 0.5.1
brms: version 2.20.4
R: version 4.3.3 ("Angel Food Cake")

Could this be an object class names issue?

[paul-buerkner]

Update to the latest github version of brms

[nstucke]

Updated to brms: version 2.20.15 using:

if (!requireNamespace("remotes")) {
install.packages("remotes")
}
remotes::install_github("paul-buerkner/brms")

Re-ran all models above, and the error persists...

[paul-buerkner]

Did you restart R?

[nstucke]

Ooooof! That worked! Thank you so much, @paul-buerkner! Appreciate your time!