Hi, I've been working with the ALPS classical monte carlo program, and according to the ALPS website/wiki, it should be possible to take a finished simulation, edit number of SWEEPS in the simulation_name.task#.out.xml files, set the status of the task from "finished" to "running" in the simulation_name.out.xml file and then do "spinmc simulation_name.out.xml". However, the program immediately throws an exception saying "Cannot change parameter SWEEPS".
I determined that the exception is being thrown from line 300 of worker.C in alps-1.3.1/src/alps/scheduler/. Is this a feature that just never got implemented or am I going about the procedure incorrectly?
Also, could ALPS (or does ALPS already) include a feature that allows it to run in a mode where it automatically advances to the next task if the current task has converged in all measurements, even if the full range of sweeps has not been performed? I think this would be an incredibly userful feature!
I have just tried it myself and it works for me. Can you send me your simulation files which show the problem?
Matthias
On 29 May 2008, at 03:10, Miles Stoudenmire wrote:
Hi, I've been working with the ALPS classical monte carlo program, and according to the ALPS website/wiki, it should be possible to take a finished simulation, edit number of SWEEPS in the simulation_name.task#.out.xml files, set the status of the task from "finished" to "running" in the simulation_name.out.xml file and then do "spinmc simulation_name.out.xml". However, the program immediately throws an exception saying "Cannot change parameter SWEEPS".
I determined that the exception is being thrown from line 300 of worker.C in alps-1.3.1/src/alps/scheduler/. Is this a feature that just never got implemented or am I going about the procedure incorrectly?
Also, could ALPS (or does ALPS already) include a feature that allows it to run in a mode where it automatically advances to the next task if the current task has converged in all measurements, even if the full range of sweeps has not been performed? I think this would be an incredibly userful feature!
-- -=Miles Stoudenmire=- miles@physics.ucsb.edu miles.stoudenmire@gmail.com http://www.crazytheory.com/
Hi Matthias - I've attached the relevant files for a little test simulation that I've been using to understand the workings of the ALPS spinmc program. I just tried increasing the sweeps in the 'energy_test.task1.out.xml' file and also in the 'energy_test' parameter file (even though I figure the program doesn't actually read that, but just to be double sure) and it immediately complains and quits upon my running of "spinmc energy_test.out.xml".
Do you think it matters that I'm using alps version 1.3b6?
On Thu, May 29, 2008 at 10:03 AM, Matthias Troyer troyer@phys.ethz.ch wrote:
I have just tried it myself and it works for me. Can you send me your simulation files which show the problem?
Matthias
On 29 May 2008, at 03:10, Miles Stoudenmire wrote:
Hi, I've been working with the ALPS classical monte carlo program, and according to the ALPS website/wiki, it should be possible to take a finished simulation, edit number of SWEEPS in the simulation_name.task#.out.xml files, set the status of the task from "finished" to "running" in the simulation_name.out.xml file and then do "spinmc simulation_name.out.xml". However, the program immediately throws an exception saying "Cannot change parameter SWEEPS".
I determined that the exception is being thrown from line 300 of worker.C in alps-1.3.1/src/alps/scheduler/. Is this a feature that just never got implemented or am I going about the procedure incorrectly?
Also, could ALPS (or does ALPS already) include a feature that allows it to run in a mode where it automatically advances to the next task if the current task has converged in all measurements, even if the full range of sweeps has not been performed? I think this would be an incredibly userful feature!
-- -=Miles Stoudenmire=- miles@physics.ucsb.edu miles.stoudenmire@gmail.com http://www.crazytheory.com/
On 29 May 2008, at 03:10, Miles Stoudenmire wrote:
Also, could ALPS (or does ALPS already) include a feature that allows it to run in a mode where it automatically advances to the next task if the current task has converged in all measurements, even if the full range of sweeps has not been performed? I think this would be an incredibly userful feature!
yes, that is possible. You need to edit the work_done() function of the simulations. For the spinmc codes this exist. You need to specify two variables:
ERROR_VARIABLE
ERROR_LIMIT
The first is the name of the observable, and the second the error you wish to achieve
Matthias
comp-phys-alps-users@lists.phys.ethz.ch