Skip to Content.
Sympa Menu

opal - Re: [Opal] Multi objective optimization

opal AT lists.psi.ch

Subject: The OPAL Discussion Forum

List archive

Re: [Opal] Multi objective optimization


Chronological Thread 
  • From: Anastasiya Bershanska <nastya AT mit.edu>
  • To: Achim Gsell <achim.gsell AT psi.ch>
  • Cc: Anastasiya Bershanska <nastya AT mit.edu>, Daniel Winklehner <winklehn AT mit.edu>, "opal AT lists.psi.ch" <opal AT lists.psi.ch>
  • Subject: Re: [Opal] Multi objective optimization
  • Date: Fri, 29 Jun 2018 12:42:03 +0000
  • Accept-language: en-US

Here’re the input files. I’ve noticed that the error occurs because optimizer seems to start recursive optimization jobs and never actually run the simulation. Also, note that I’m using runOPAL.py to schedule jobs on the cluster, which might be the reason the recursive jobs occur.

- Anastasiya

On Jun 29, 2018, at 5:26 AM, Achim Gsell <achim.gsell AT psi.ch> wrote:


On 29 Jun 2018, at 06:03, Anastasiya Bershanska <nastya AT mit.edu> wrote:

UDP: I’ve figured out how to run the optimizer and I think I have encountered a bug. I get following error, which is probably related to a memory allocation issue:
*** Error in `opal': free(): invalid next size (fast): 0x0000000002e2a360 ***
I’ve tried increasing RAM (up to 20GB for single particle simulations) and this issue always arrises.

Can you please provide input files so that we can reproduce the error?

Thanks

Achim

- Anastasiya

On Jun 28, 2018, at 5:08 PM, Daniel Winklehner <winklehn AT mit.edu> wrote:

Hi All,

The information Anastasiya needs from the stat file is the z coordinate of the reference particle (seems like only s is given, which probably stems from the mismatch between OPAL-T and OPAL-cycl coordinate systems). Maybe we were looking in the wrong place, but if not it might warrant an additional column in the stat file rather than making this complicated through use of SYSTEM.

Cheers,
  Daniel

-----Original Message-----
From: opal-request AT lists.psi.ch <opal-request AT lists.psi.ch> On Behalf Of Anastasiya Bershanska
Sent: Thursday, June 28, 2018 4:59 PM
To: Jochem Snuverink <jochem.snuverink AT psi.ch>
Cc: Anastasiya Bershanska <nastya AT psfc.mit.edu>; opal AT lists.psi.ch
Subject: Re: [Opal] Multi objective optimization

Hello Jochem,

The calculation is not complicated and I can do math using built-in functions, but I need information from *-trackOrbit.dat. (I will have to use SYSTEM command.) How can I tell what is the correct output file for a specific run that is a part of optimization job?

Also, do you have an example of an optimization job that works? If not, can someone update the documentation to specify how such optimization should be setup? (i.e. is optimization command and the actual simulation setup are in one file or in two separate ones? what is the folder structure in case it’s two different files?)

Btw, runOPAL.py no longer recognizes —optPilot (at least when I tried). Could you please update/fix that on the wiki?

Thank you,
Anastasiya

On Jun 28, 2018, at 4:59 AM, Jochem Snuverink <jochem.snuverink AT psi.ch> wrote:

Hi Anastasiya,

Is the calculation of your variable very complicated? If not you can calculate it perhaps in the input file directly? You have a few basic functions available in the EXPR part (sqrt, pow, exp, log, fabs, sin, etc., see https://gitlab.psi.ch/OPAL/Manual-2.0/wikis/optimiser) that you can use on the stat file variables.

The general case is however not possible at the moment. If it is a meaningful variable, we can possibly add it to the .stat file.

Alternatively, you can try to use the SYSTEM command (https://gitlab.psi.ch/OPAL/Manual-2.0/wikis/control) at the end of each run. This executes a system call, and if you do your calculation in a separate program and write it to file, you can read in this file again with "fromFile". This is untested, but I believe it should work.

hope this helps,
Jochem

On 27/06/18 21:19, Anastasiya Bershanska wrote:
Hello,

I have a questions on the optimizer. According to the documentation, EXPR in OBJECTIVE and CONSTRAINT can be specified using a few built-in functions that take data from .stat file and variables defined in the temple file. However, I would like to be able to optimize with respect to a variable that I calculate using .stat and/or .dat files (i.e. specify OBJECTIVE as a new function of input files). Is it possible to specify an new function that reads from output files within OPAL input temple file?

Thank you,
Anastasiya



<IsoDAR_opt.o11629666>


Paul Scherrer Institut
WHGA/U136
CH-5232 Villigen PSI
+41 56 310 36 74


Attachment: optimizer.zip
Description: optimizer.zip




Archive powered by MHonArc 2.6.19.

Top of Page