Skip to content

Latest commit

 

History

History
59 lines (47 loc) · 3.12 KB

README_BATCH.md

File metadata and controls

59 lines (47 loc) · 3.12 KB
Raw

Running ensemble simulations

The script runylmo is designed to prepare and run one Yelmo simulation with parameters loaded from a specified parameter file, and potentially updated on the command-line via -p. To perform an ensemble of Yelmo simulations with various combinations of parameter values, runylmo can be called via the Python runner module. See below for installation instructions.

Simple ensemble

To run a simple ensemble, call runylmo via the jobrun command. For example, to run an ensemble of benchmark simulations with different resolutions of 30.0 and 50.0 km, you can run the following command:

jobrun ./runylmo -r -e benchmarks -n par-gmd/yelmo_HALFAR.nml -- -o output/run -p eismint.dx=30.0,50.0

Everything between jobrun and -- corresponds to the normal call to runylmo, with one key difference. The output directory argument -o output/run is specified after the -- as this is now the ensemble output directory. Individual simulation directories will appear inside this main ensemble directory.

Everything after the -- corresponds to the options of the jobrun command, namely you need to specify the ensemble output directory via -o and the parameter combinations to run via -p KEY=VAL [KEY=VAL...]. In this case, -o output/run sets the ensemble output directory location (individual simulations will appear as numbered folder 0,1,2,... inside of output/run). -p is the option to specify parameter options to modify, in this case setting the parameter values of eismint.dx (group.parameter) to 30.0 and 50.0.

So, the above command would make an ensemble of two simulations with different values of eismint.dx in the output location output/run.

It's also possible to acheive the above in two steps, first sampling the parameter values and storing them in a file, then loading that file to generate the ensemble:

job product eismint.dx=30.0,50.0 > ensemble1.txt 
jobrun ./runylmo -r -e benchmarks -n par-gmd/yelmo_HALFAR.nml -- -o output/run -i ensemble1.txt

Parameter sampling

Using the command job sample, it is possible to generate an ensemble drawing from different parameter distributions instead:

job sample eismint.dx=U?30.0,50.0 --size 10 > ensemble2.txt
jobrun ./runylmo -r -e benchmarks -n par-gmd/yelmo_HALFAR.nml -- -o output/run -i ensemble2.txt

The above command samples 10 values of eismint.dx from a uniform distribution ranging from 30.0 to 50.0 following a Latin-Hypercube sampling method. It is also possible to sample from a normal distribution, e.g., with mean and standard deviation of (40.0,10.0):

job sample eismint.dx=N?40.0,10.0 --size 10 > ensemble3.txt

Installing runner

  1. Download or clone the runner package from the repository: https://github.com/alex-robinson/runner

  2. From inside the main directory, install the package via pip:

cd runner
pip install ./

Once completed successfully, the system commands job, job sample, job run and jobrun should be available at the command line. Check via job -h for example.