Incremental Affinity Propagation based on Cluster Consolidation and Stratification

To reproduce our results:

• unzip datasets.zip and samples.zip: unzip -q filename.zip
• comment the 3rd, 5th, and 15th lines in the bash files iris.sh, car.sh, wine.sh, kddcup.sh. By commenting you will perform exactly the same experiments as ours. Otherwise, you will randomly resample the datasets.
• run all the bash script: bash dataset.sh
• run the python script results.py specifying the stats file and the metric to be analysed: e.g., python3 results.py --path stats/kddcup/s1904_mg200.dill --metric purity returns the purity for the KDD-CUP'99 dataset in the incremental setting.
• run the python script plot_argmedian.py specifying the stats file, the dataset, the metric to be analysed, and where you want to save the plot as image: e.g., python3 plot_argmedian.py --input stats/kddcup/app_s53_ml0_mg0_abl.dill --dataset kddcup --output ablation_study.png --metric purity returns the purity trend for the KDD-CUP'99 ablation setting.

Check our paper for further details on the experiments.