A bit swamped now but when I have the time I can try to install g++, @bgoodri. Take your pick:
It might be interesting to see if PyStan on Catalina faces the same problems as RStan.
Iām currently running simulated based calibration in pystan on Catalina MacBook Pro and I see no issues. Catalina however forced me to reinstall entire python.
Apple has extended support for Mojave until Sept 2021. No reason to upgrade with the minimal updates and all the issues until a later date.
2 Likes
I updated to Catalina (sorry! I didnāt realize I wasnāt supposed to do this; I was under the impression that for security purposes I should always do every update as soon as possible). I then did that thing to open Xcode. 8 schools compiles and runs on my Mac now. But I havenāt tried anything more complicated.
Hehehe, well @andrewgelman, itās nice to see that this can happen to everyoneā¦ What happens when you run foo.R (1.8 KB)?
what was the reinstall procedure you used?
what version of Python is default on Catalina?
Nothing fancy - I used anaconda to have all dependencies, and after installing Catalina I had to install it again. From my understanding Catalina forbids you from writing outside of user/username folder, and things like anaconda suffer.
Good news everyone! The next version of rstan might avoid the problems we have had with exception handling when using clang++. However, Stan programs execute 300 times slower (on Linux; g++ seems to produce the expected speed).
2 Likes
I ran foo.R. Most of it ran ok but I did run into some errors at the end; see below.
I presume itās the sampling error (exception)?
which version of Rstan? is this just for Catalina?
hi Ben, just trying to follow this discussion - if I understand correctly:
on Linux, using clang++, unreleased Rstan is 300 times slower
on Linux, using g++, unreleased Rstan runs with expected speed.
is this correct?
on Linux, is there a default / preferred c++ compiler, if so, is it clang++ or g++ or?
Yes, that is all correct. I was just using clang++ to see if I could get something that would also work on Catalina and noticed the slowdown. The only caveat to what you said is that on RStudio Cloud, we need to be able to use clang++ on Linux because using g++ to compile the model exceeds the 1 GB limit on RAM. But you-know-who will complain if the runtime goes from 0.01 seconds to 3 seconds.
1 Like
Hello,
I updated my R, RStudio and my MacOS system recently. Today, I noticed that Rstan would not return me any error message if something were wrong in the code. For example, I declared the wrong dimensionality for one of the parameters, yet RStudio still executed the code without showing any errors. I only realized that my Stan code was wrong when I found out that my result object was empty. Is there a way that I can fix this issue?
Here is my system info:
R: 3.6.1
Rstudio: 1.2.5
Rstan: 2.19.2
MacOS: Catalina 10.15
Folks the thread is super long so iām not clear - whats the current best to install Stan under Catalina. Iām reunited with 12" Macbook I use for dabbling around with stuff like this. Planning to put Catalina on it and willing to give Stan a go too.
Current best practice is to not use Catalina.
2 Likes
haha thats why Iām using my dabbling machine and not the one I use to to real work on!
There is a way, but it causes Stan to use an unlimited amount of RAM. Iāll make another thread for that soon.
2 Likes
Just to clarify for anyone else reading this: I think the problem right now is just installing and running via RStan but not installing or running Stan in general. As far as I know you should be able to still install and use the other interfaces, including the new CmdStanR if you want to stick with running Stan from R. RStan is officially released and has more features than CmdStanR so itās not a full replacement for RStan, but just wanted to clarify that as far as I know all of Stan isnāt broken by Catalina.
Anyway, thanks to everyone trying to help with this and sorry for the inconvenience while this gets sorted out!
2 Likes