Some families with smooth and bs='gp' crash R

[daskandalis]

I am trying to analyse non-Gaussian data with smooths. In the following example, the binomial and Poisson models crash R but the beta model does not. The Binomial and Poisson models run fine without bs='gp' (I am also curious as to whether a Gaussian process is appropriate with such distributions). Binomial and Poisson models can be rescued by running them together, with or without beta. In my data the beta model crashes and the binomial is fine, but the beta can similarly be rescued by running with the binomial.

require(brms)

N <- 100

n <- sample( 1:N, N, replace=TRUE )

binom <- rbinom(N, n, 0.6)
pois <- rpois(N, 1)
beta <- rbeta(N, 1,1)
t=seq(0,10,length.out=N)

dat <- data.frame(t=t, binom=binom, n=n, pois=pois, beta=beta)

pois.bf <- bf(pois~s(t,bs='gp'), family=poisson())
binom.bf <- bf(binom|trials(n)~s(t,bs='gp'), family=binomial())
beta.bf <- bf(beta~s(t,bs='gp'),family=zero_one_inflated_beta())

# crashes session
pois.mod <- brm( pois.bf, seed=1, chains=1, iter=100, data=dat )

# crashes session
binom.mod <- brm( binom.bf, seed=1, chains=1, iter=100, data=dat )

beta.mod <- brm( beta.bf, seed=1, chains=1, iter=100, data=dat )

mv.mod <- brm( binom.bf + pois.bf + beta.bf, seed=1, chains=1, iter=100, data=dat )

---

sessionInfo()
R version 3.5.1 (2018-07-02)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)
Matrix products: default
locale:
[1] LC_COLLATE=English_United States.1252 LC_CTYPE=English_United States.1252 LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C LC_TIME=English_United States.1252
attached base packages:
[1] stats graphics grDevices utils datasets methods base
loaded via a namespace (and not attached):
[1] compiler_3.5.1 tools_3.5.1 yaml_2.2.0

[paul-buerkner]

Thanks for providing a reproducible example. Unfortunately, it runs for me just fine (on Windows 10, R 3.6, Rtools 3.5). I can't tell you what causes a problem but it is likely related to Stan itself (if at all) but not to brms, since the latter merely writes the Stan code.

Here is my session Info

sessionInfo()
R version 3.6.0 (2019-04-26)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 17763)

Matrix products: default

Random number generation:
 RNG:     Mersenne-Twister 
 Normal:  Inversion 
 Sample:  Rounding 
 
locale:
[1] LC_COLLATE=German_Germany.1252  LC_CTYPE=German_Germany.1252   
[3] LC_MONETARY=German_Germany.1252 LC_NUMERIC=C                   
[5] LC_TIME=German_Germany.1252    

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

other attached packages:
[1] brms_2.8.9 Rcpp_1.0.1

loaded via a namespace (and not attached):
 [1] Brobdingnag_1.2-6    splines_3.6.0        gtools_3.8.1        
 [4] StanHeaders_2.18.0   threejs_0.3.1        shiny_1.1.0         
 [7] assertthat_0.2.1     stats4_3.6.0         pillar_1.3.1        
[10] backports_1.1.4      lattice_0.20-38      glue_1.3.1          
[13] digest_0.6.18        promises_1.0.1       colorspace_1.4-1    
[16] htmltools_0.3.6      httpuv_1.4.5         Matrix_1.2-17       
[19] plyr_1.8.4           dygraphs_1.1.1.6     pkgconfig_2.0.2     
[22] rstan_2.18.2         purrr_0.3.2          xtable_1.8-3        
[25] mvtnorm_1.0-8        scales_1.0.0         processx_3.2.0      
[28] later_0.7.5          tibble_2.1.1         mgcv_1.8-28         
[31] bayesplot_1.6.0      ggplot2_3.1.1        DT_0.4              
[34] withr_2.1.2          shinyjs_1.0          lazyeval_0.2.2      
[37] cli_1.1.0            magrittr_1.5         crayon_1.3.4        
[40] mime_0.6             ps_1.2.0             nlme_3.1-139        
[43] xts_0.11-1           pkgbuild_1.0.2       colourpicker_1.0    
[46] rsconnect_0.8.8      tools_3.6.0          loo_2.1.0           
[49] prettyunits_1.0.2    matrixStats_0.54.0   stringr_1.4.0       
[52] munsell_0.5.0        callr_3.0.0          packrat_0.4.9-3     
[55] compiler_3.6.0       rlang_0.3.4          grid_3.6.0          
[58] ggridges_0.5.1       rstudioapi_0.8       htmlwidgets_1.3     
[61] crosstalk_1.0.0      igraph_1.2.2         miniUI_0.1.1.1      
[64] base64enc_0.1-3      codetools_0.2-16     gtable_0.3.0        
[67] inline_0.3.15        abind_1.4-5          markdown_0.9        
[70] reshape2_1.4.3       R6_2.4.0             gridExtra_2.3       
[73] rstantools_1.5.1     zoo_1.8-4            bridgesampling_0.6-0
[76] dplyr_0.8.0.1        shinystan_2.5.0      shinythemes_1.1.1   
[79] stringi_1.4.3        parallel_3.6.0       tidyselect_0.2.5    
[82] coda_0.19-2         

[daskandalis]

Switched to R 3.6, Rtools 3.5, Rstudio 1.2.x and both models are now compiling.

[paul-buerkner]

Glad it works now!