Manually setting SCE optimization parameters & CPU usage of mHM ? #103
-
Hi! I want to optimize mHM using the shuffled complex evolution approach (SCE). When I run it, i get this error message: Run mHM optimization When i manually set the parameters sce_nps and sce_npg to -9 in my mhm.nml file (which, according to the source code, should be the default value), it runs without any problem. My question is: is it ok to set these parameters manually or does this error indicate that something is off with my setup ? And my next question is: is somewhere hardcoded how many CPUs mHM will use during optimization? Thanks a lot for your help, |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments
-
Hi Malve, Setting the npg manually should be fine, but there should be at least 3 points per complex. Complexes in SCE cannot be lines or points (hence the error message). However, the default value provides a reasonable choice that we also often use. The number of CPUs used during running mHM (including the application during optimization) is not hard-coded. You can set the number of OPEN_MP threads to be used in the command line before you execute mHM via export OMP_NUM_THREADS=#number_of_cores# Make sure that you actually compiled mHM with the OPEN_MP switch turned on. If you use CMAKE for installation, then you can use this script for compilation https://git.ufz.de/mhm/mhm/-/blob/develop/CI-scripts/compile_OpenMP Best, |
Beta Was this translation helpful? Give feedback.
-
Can you tell me how to do the calibration? Thanks a lot. |
Beta Was this translation helpful? Give feedback.
Hi Malve,
Setting the npg manually should be fine, but there should be at least 3 points per complex. Complexes in SCE cannot be lines or points (hence the error message). However, the default value provides a reasonable choice that we also often use.
The number of CPUs used during running mHM (including the application during optimization) is not hard-coded. You can set the number of OPEN_MP threads to be used in the command line before you execute mHM via
export OMP_NUM_THREADS=#number_of_cores#
Make sure that you actually compiled mHM with the OPEN_MP switch turned on. If you use CMAKE for installation, then you can use this script for compilation https://git.ufz.de/mhm/mhm/-/blob/deve…