Install error on R 3.4, worked on 3.3 about projpred HOT 10 CLOSED

It seems you removed the detailed description so I'm copy-pasting it here for reference:

"I get an install error on R 3.4 but it installed fine with R 3.3 a couple of days earlier.
devtools::install_github('stan-dev/projpred', build_vignettes = TRUE)
....
This is projpred version 0.4.2
error occurred during calling the sampler; sampling not done
Quitting from lines 36-47 (quickstart.Rmd)
Error: processing vignette 'quickstart.Rmd' failed with diagnostics:
Invalid stanfit object produced please report bug
Execution halted
Error: Command failed (1)

Session info :
setting value
version R version 3.4.0 (2017-04-21)
system x86_64, darwin15.6.0
ui RStudio (1.0.143)"

We tested the installation again with the new R version 3.4 (on Linux) and the vignette builds correctly. Based on the error message, it seems that the rstanarm is crashing and not our software. Which rstanarm version are you using? If you are running old version of rstanarm, you could try updating it and then installing projpred again.

from projpred.

Thank you. I keep the software up-to-date and I am using rstanarm 2.15.2 with Stan 2.15.1. I am on Mac OS Sierra 10.12.3 (up to date).

from projpred.

Hi,
unfortunately we are not able to reproduce the error. However, we made some changes to the package, so you could simply try to install the package again.

If that does not work, at least you should be able to install the package without building the vignettes (ie. just devtools::install_github('stan-dev/projpred', build_vignettes = FALSE)). If that works, it would be helpful if you could try to run the following code and see if it produces an error (it is the same code that seems to produce the crash that you observe):

library(rstanarm)
library(projpred)
library(ggplot2)
data('df_gaussian', package = 'projpred')
n <- nrow(df_gaussian$x) # 100
D <- ncol(df_gaussian$x) # 20
p0 <- 5 # prior guess for the number of relevant variables
tau0 <- p0/(D-p0) * 1/sqrt(n) # scale for tau (notice that stan_glm will automatically scale this by sigma)
prior_coeff <- hs(df=1, global_df=1, global_scale=tau0) # horseshoe prior
fit <- stan_glm(y ~ x, family=gaussian(), data=df_gaussian, prior=prior_coeff,
            # to make this vignette build fast, we use only 2 chains and
            # 800 draws. In practice, more conservative values, eg. 4 chains
            # and 2000 draws might be required for reliable inference.
            seed=1, adapt_delta=0.999, chains=2, iter=800) 
fit <- varsel(fit, method='L1')
fit$varsel$chosen

from projpred.

projpred loaded. Thank you.

After the first fit statement I received the following message:
SAMPLING FOR MODEL 'continuous' NOW (CHAIN 1).
[1] "Error in sampler$call_sampler(args_list[[i]]) : "
[2] " empty_nested() must be true before calling recover_memory()"
error occurred during calling the sampler; sampling not done
Error in check_stanfit(stanfit) :
Invalid stanfit object produced please report bug

I am running with Mac OS Sierra 10.12.4 with the latest command line tools and also installed the latest llvm with homebrew. /usr/local/Cellar/llvm/4.0.0_1

clang version 4.0.0 (tags/RELEASE_400/final)
Target: x86_64-apple-darwin16.5.0
Thread model: posix
InstalledDir: /usr/local/opt/llvm/bin

from projpred.

@jgabry : do you have any idea about what might cause this kind of error message when fitting rstanarm-model ('Invalid stanfit object produced please report bug')? Looks like it comes from rstanarm but I really have no clue about the possible cause.

from projpred.

There was an update to rstanarm that fixed the problem.

from projpred.

from projpred.

@dfumento : so did I understand correctly that the issue is now resolved and can be closed?

from projpred.

from projpred.

Great, thanks!

from projpred.