Parallelize computations – prof-batchtune¶prof-batchtune assists you to run prof-interpolate or prof-tune in parallel, e.g. on a batch cluster, by filling job-script template files. This is done by splitting the MC run combinations into smaller sets. These sets can then be processed in parallel. The data locations must be given on the command line (e.g. --datadir or --weights), and the MC, reference and interpolation directories must all exist. The paths are converted to absolute paths. The number of scripts that are produced is controlled with -N NUMSCRIPTS. Additional arguments can be passed through to the called professor script with -o ADDITIONALOPTIONS. Paths in ADDITIONALOPTIONS must be absolute. The output of the scripts is stored in numbered directories under JOBBASEOUTDIR given by -j A simple algorithm is used to distribute the number of run combinations equally over the job scripts. If the number of scripts is not a divisor of the number of run combinations, the last script will contain less than the other scripts. In some cases less than NUMSCRIPTS scripts will be created. Note At the moment the Cheetah templating library is required. Templates¶Template files are distributed together with the professor source code:
Examples¶Create 20 job scripts for prof-interpolate to compute the coefficients of the polynomial parameterization. Use cubic polynomials: prof-batchtune -N 20 --weights weights --datadir . -R runs-cubic -o "--ipol-method=cubic" /PATH/TO/PROFESSOR/SOURCE/templates/prof-interpolate.template
The scripts are created in scripts_interpolate_weights/. The output directory for the coefficient files (./ipol/) must exist! Create 10 job scripts for prof-tune using quadratic polynomials and the default initial point method: prof-batchtune -N 10 --weights my.weights -R my.combinations --datadir my/data /PATH/TO/PROFESSOR/SOURCE/templates/prof-tune.template
The scripts are created in scripts_tune_my.weights/. The result files will be put in jobout/xxx/results.pkl. Create 20 job scripts for prof-tune with a fixed parameter and three initial points per run combination using cubic polynomials: prof-batchtune -N 20 --weights weights --datadir . -R runs.cubic -o "--ipol-method=cubic --spmethods='random,random,center' --fixed-params='Par1=3.1'" /PATH/TO/PROFESSOR/SOURCE/templates/prof-tune.template
Command-line options¶
Input data¶
|