This repo contains the source code that extracts the Fitness Landscape measures used for characterizing QAP instances in the paper "On the use of fitness landscape features in meta-learning based algorithm selection for the quadratic assignment problem".
The sampling is performed by the Metropolis-Hasting algorithm, followed by a best improvement local search on each solution with the swap operator.
Running make will generate two binary files:
- bin/mh
- bin/ft
First you need to execute the sampling method:
$ ./bin/mh -f <path to instance file> -s <sample size> -o <sample file>
This will save the sample in the binary <sample file> and print the CPU execution time. Then, you run the feature extractor on the generated sample:
$ ./bin/ft -f <sample file> [-c <cropped sample size>]
Which will print the values of the extracted features in the following order:
- Optima Fitness Coefficient
- Average Descent
- Fitness Distance Correlation
- Average Distance to Optima
- Accumulated Escape Probability
- Dispersion Metric (base sample)
- Dispersion Metric (local optima sample)
If you pass the optional argument -c <cropped sample size>, the features will be computed considering a cropped subsample.