Optionally resetting the random number generator stream in Rockstar
Hi all, I've just pushed a change to the fork of Rockstar I have on bitbucket that adds a default-off compile-time setting that governs whether or not the random stream is reset following the call to "random_sleep" that governs how the clients connect to the server. https://bitbucket.org/MatthewTurk/rockstar/commits/cf88fc3f9c1c9f0768005be5a... This addresses something that's come up on this list before. I think it should make runs completely deterministic (it does in my case) and will fix the halo finding and halo analysis tests. They are now, in my tests, bitwise identical between runs whereas before I was getting definitely non-bitwise results. (This is noted in the paper.) I wanted to make this a parameter, but I wasn't able to figure out how to do it without breaking API compatibility. So eventually, this should make its way upstream, but for now, it should (if compiled correctly!) fix the tests. If you're building rockstar at home: RNG_FLAGS=-DRESET_RNG_STREAM make clean lib should get it for you. -Matt _______________________________________________ yt-dev mailing list yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
Good detective work, Matt. Thanks for this! Just to clarify, the change you made was in the version of Rockstar that gets automatically built/installed with yt if you ask the install_script.sh to install Rockstar? So this will be fixed for people who install Rockstar from here on out, and for those of us who already have Rockstar installed, we just need to make the compile time flag you suggest in the Makefile and rebuild rockstar? Thanks again! Cameron On Fri, Dec 19, 2014 at 4:32 PM, Matthew Turk <matthewturk@gmail.com> wrote:
Hi all,
I've just pushed a change to the fork of Rockstar I have on bitbucket that adds a default-off compile-time setting that governs whether or not the random stream is reset following the call to "random_sleep" that governs how the clients connect to the server.
https://bitbucket.org/MatthewTurk/rockstar/commits/cf88fc3f9c1c9f0768005be5a...
This addresses something that's come up on this list before. I think it should make runs completely deterministic (it does in my case) and will fix the halo finding and halo analysis tests. They are now, in my tests, bitwise identical between runs whereas before I was getting definitely non-bitwise results. (This is noted in the paper.)
I wanted to make this a parameter, but I wasn't able to figure out how to do it without breaking API compatibility. So eventually, this should make its way upstream, but for now, it should (if compiled correctly!) fix the tests.
If you're building rockstar at home:
RNG_FLAGS=-DRESET_RNG_STREAM make clean lib
should get it for you.
-Matt
_______________________________________________ yt-dev mailing list yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
-- Cameron Hummels Postdoctoral Researcher Steward Observatory University of Arizona http://chummels.org _______________________________________________ yt-dev mailing list yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
Hi Cameron, Sort of. Behavior will be unchanged unless you specifically rebuild from my version. My version is the one that the install script grabs. On Fri, Dec 19, 2014, 4:30 PM Cameron Hummels <chummels@gmail.com> wrote:
Good detective work, Matt. Thanks for this!
Just to clarify, the change you made was in the version of Rockstar that gets automatically built/installed with yt if you ask the install_script.sh to install Rockstar? So this will be fixed for people who install Rockstar from here on out, and for those of us who already have Rockstar installed, we just need to make the compile time flag you suggest in the Makefile and rebuild rockstar?
Thanks again!
Cameron
On Fri, Dec 19, 2014 at 4:32 PM, Matthew Turk <matthewturk@gmail.com> wrote:
Hi all,
I've just pushed a change to the fork of Rockstar I have on bitbucket that adds a default-off compile-time setting that governs whether or not the random stream is reset following the call to "random_sleep" that governs how the clients connect to the server.
https://bitbucket.org/MatthewTurk/rockstar/commits/cf88fc3f9c1c9f0768005be5a...
This addresses something that's come up on this list before. I think it should make runs completely deterministic (it does in my case) and will fix the halo finding and halo analysis tests. They are now, in my tests, bitwise identical between runs whereas before I was getting definitely non-bitwise results. (This is noted in the paper.)
I wanted to make this a parameter, but I wasn't able to figure out how to do it without breaking API compatibility. So eventually, this should make its way upstream, but for now, it should (if compiled correctly!) fix the tests.
If you're building rockstar at home:
RNG_FLAGS=-DRESET_RNG_STREAM make clean lib
should get it for you.
-Matt
_______________________________________________ yt-dev mailing list yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
-- Cameron Hummels Postdoctoral Researcher Steward Observatory University of Arizona http://chummels.org _______________________________________________ yt-dev mailing list yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
_______________________________________________ yt-dev mailing list yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
participants (2)
-
Cameron Hummels -
Matthew Turk