This contains the NSGA-II swarm-algorithm evolutionary script that is used to evaluate and compare the multi-objective performance of such hybrid algorithms. This work was supported by National Science Foundation Award (#1749635).
Requires Python 3.7 (or later) alongside numpy and scipy. matplotlib may also be required if the visualization script is used.
To run the model selection script, change the parameters within the script and use
python3 evaluate.py [Model] [output_filename]
There are two models available, Weibull and Covariate. Including -p in the launch options can be used to evaluate pre-defined chromosome patterns in order to cross-validate data, and -h can be used to view objectives per-iteration in order to output to a file to plot histograms. NSGA-II parameters can be changed within the script.
The models.py script contains objective functions for optimization, so a user may opt to modify or add their own swarm-applicable model. The swarm-hybrid algorithms will attempt to minimize whichever model it is given. This script is used by the evaluate script, and should not be ran on its own.
python3 visualize.py [input_filename] [Model]
may also be used to plot various data (histograms, objectives, i.e. figures 5-9 in manuscript). Some slight editing of the script may be required to modify plot parameters. Included in the repository is the cross-validation data for the Covariate model testing, for both the histogram and test plotting.
Algorithms involved were adapted from R implementations of Brownlee's Clever Algorithms (Link), most notably the NSGA-II-based overhead management, or Nature-inspired metaheuristic algorithms (Yang, Xin-She) in the case of the swarm algorithms involved.
Accurate comparison between results depends on the performance of the computer running the script. The NSGA-II optimization is ran single-threaded in order to verify the integrity of all separate algorithm runs. As such, the script will take some time to run provided somewhat-large NSGA parameters. Algorithm tuning was performed using an Intel i9-9900K using stock clocks.