Rstan 2.17.4 from CRAN does not improve over 2.17.3, wait for 2.18


#1

Do NOT install rstan 2.17.4 from CRAN. It is no better than the rstan 2.17.3 that is currently on CRAN. Wait until rstan 2.18.1 is on CRAN to upgrade.


#2

Is there something wrong with the RStan on CRAN?


#3

No, rstan 2.17.3 which is still what there are binaries for, is fine.


#4

The binaries are 2.17.4 now. Can you clarify if there is something wrong with 2.17.4 ? After erroring out due to generated quantities I checked for updates and installed hopefully :)


#5

2.17.4 is not going to overcome any errors that you were getting with 2.17.3 because it only changes one line.


#6

Thanks, that wasn’t the question but clarifies anyway :) :)


#7

Our users need to know if you mean:

  • 2.17.4 is broken and shouldn’t be installed, or

  • there’s nothing worth upgrading for in 2.17.4 over 2.17.3.

The former means they have to figure out how to get old versions and the latter just means they can avoid a useless upgrade.


#8

It is more the latter, although the toolchain configuration may be slightly different in 2.17.4 than 2.17.3. The rstan 2.17.4 package only exists to be compatible with both StanHeaders 2.17.1 and StanHeaders 2.18.0 in order for CRAN to push the latter out (which they have not done yet).


#9

If I understand correctly, rstan 2.17.4 :

  • is not incorrect ;
  • allows CRAN to push StanHeaders 2.18.0 (which is useful “CRAN-wise”) ;
  • entails recompilation of rstan (which is not light, granted…).

Its only “problem” is that it doesn’t do anything useful “user-wise”.

Many (most) R users have dozens (hundreds) CRAN packages, and have more or less automated their maintenance by (semi-)automatic upgrade from CRAN ; this usually turns out safe and useful. An alert about a flawed CRAN package involves manual work to either prevent or revert such an “un-useful” upgrade requires manual work. Which turns out not to be useful “user-wise”.

Could one avoid to shout “Wolf” when there’s but the shade of a chihuahua ?

And, by the way, what’s the problem with rstan and 2.18 ?

Thanks in advance,


#10

There isn’t a problem with rstan 2.18 other than CRAN hasn’t approved it. 2.17.4 can be made to work the same as 2.17.3 but requires C++14, which is what we were trying to avoid for 2.17.x.


#11

There isn’t a problem with rstan 2.18 other than CRAN hasn’t approved it. 2.17.4 can be made to work the same as 2.17.3 but requires C++14, which is what we were trying to avoid for 2.17.x.

Thanks a lot !

C++ 14 may indeed be a problem, at least on Windows (I understand tjat the C/C++ tools requirements for building R are quite peculiar…).

Did R Core emit recommendations about rstan 2.18 ?

And, BTW, is the (future) rstan 2.18 accessible somewhere for ®ash Linux users ?


#12

2.18 is going to work on Windows. CRAN has not said anything yet about rstan 2.18 because they haven’t gotten past StanHeaders 2.18, but both can be installed from GitHub.


#13

This is getting silly long so I changed the title to something that sounds less alarming otherwise you’ll be reassuring scared users till 2.18 comes out.


#14

I tried that and found it a bit problematic :

  • these instructions are specific to an available package ;

  • the build instructions seem to work, but concern only rstan,

  • but not StanHeaders, for which I didn’t find any up-o-date source (the mirror seems to be stuck at 2.17.2, nine months ago).

In consequence, the rstan_2.18.1.tar.gz package doesn’t install for lack of StanHeaders 2.18.

Advice ? Hints ?


#15

My advice was to stick with 2.17.x until 2.18.x is approved on CRAN. If you are going to go your own way, then you would need a StanHeaders 2.18.x , which you can get from the incoming/pending/ directory of the CRAN ftp server or from GitHub.


#16

I wanted to peek at novelties in a secondary R installation (Sage’s, BTW). But this sounds a bit too intricate for me (I’m not even able to find the relevant StanHeaders in Github…). So I’ll wait, I guess…


#17

Thank You…

https://discourse.mc-stan.org/