Context 🔗Simulation models can be abstracted as a program that transforms a set of inputs into a set of outputs.
Even though any type of data can be used as inputs for simulation models (parameters , pictures, CSV files, DB connections, etc.), simulation experiments revolves mostly around parameters space exploration (e.g. sensitivity analysis). Likewise, considered outputs in simulation experiments are most of times measures computed on the dynamics produced by a simulation run (e.g. fitness or error functions)
Questioning your model 🔗There are four types of general questions that can be addressed by OpenMOLE methods.
- What are all the possible inputs producing a given output ?
- What is the effect of an input variation on the output ?
- How do the inputs participate to produce the outputs ?
Are every parameter compulsory to produce outputs ?
What are the robustness intervals of the inputs that lead to a desired output ?
- What are the possible outputs of the model ?
Question 1 is referred as a calibration or optimization problem and solved via Genetic Algorithms.
Question 2 is addressed by performing a sensitivity analysis via custom Exploration tasks.
Question 3 is addressed using Calibration Profiles Algorithm, that extends the results of a sensitivity analysis.
Question 4 is addressed by using the Pattern Space Exploration (PSE) method. Calibration methods make an extensive use of Genetic Algorithms.
Specific task are available to handle the stochasticity of your model as well as distribution schemes specifically designed for distributed computing environment.
Methods characteristics 🔗Each of OpenMOLE methods comes with an image summarizing the performance and characteristics regarding criterion they have in common e.g. coverage rate of input/out of sensitivity to dimensionality.
These performance scores are defined relatively to each other and are not absolute values.