From SeSAm-Wiki

Jump to: navigation, search

Experiments / Sensitivity Analysis / Search for appropriate parameter sets

Manuel Fehler developed a calibration and analysis tool that enables you to optimize your simulation, or better said: it searches the parameter constellation that is optimal with reference to a given objective function usually expressing similarity to given data, etc.

If you just want to scan through parameter space, experiments are also appropriate. You can use theforTimesorforElements-primitives to generate parameter values (one into the other, if you want do treat more than one) and use the following construction:

Dowith DECL = new SimulationDeclaration
Simulate (createSimulation DECL)

to specify the new simulation. Then "Edit Situation" and insert the iterator values from the loop constructs as parameter variables of the world, insert defined analysises and construct a speaking name for the generated simulation. For every simulation run, new analysis files will be generated. Be aware that you have to reconstruct parameter constellations afterwards - this can be done by either also writing parameter constellation to the file, or by using "%E" in the analysis file name (together with a speaking simulation name).

The tutorial also contains a page about DefiningExperiments.

Personal tools