moeasensitivity

Sensitivty analysis on MOEA parameters using MOEAFramework and Python.

Analysis procedes in this order:

• submit does a massive number of optimization runs
• ref is for reference set computation
• hv is for hypervolume calculation
• statistics is for statistical summaries of the hypervolume data
• Make plots
  • controlmaps uses statistics to make colored contour plots of MOEA hypervolume performance in parameter space.
  • sobol uses statistics to make radial convergence plots showing Sobol' global sensitivity indices for variance in MOEA performance across the parameter space. Also make first/total order bar charts.
  • cdf does shaded bar / dot plots based on the hypervolume data
  • parallel does parallel coordinate plots and makes input data files for AeroVis scatter plotting

Bonus:

• contour shows joint probability density functions for MOEA performance, but it is less informative than I had hoped. These plots are not used in the paper.

What you have to do by hand:

• Arrange the Pareto approximation sets files into a sensible folder structure after all of the optimization has been done. All of the analysis from ref on assumes a particular folder structure for the approximation sets data.
• Run the analysis scripts in each directory. I don't have a makefile that lets you do make researchpaper. Maybe for my next study!