This repository contains the code to reproduce the experiments in $p$-value Adjustment for Monotonous, Unbiased, and Fast Clustering Comparison.
Install the dependencies with conda:
conda env create -f ./environment.yml
conda activate p_value_ri
Compile our Cython implementation of the standardized mutual information:
python setup.py build_ext --inplace
Clone the code for other clustering comparison indices
git submodule update --init --recursive
The Monte Carlo and normal approximation to the PMI are implemented in clustering_comparison_measures.py. The experiments can be run using:
# Synthetic experiments in Figures 1, 2, 3
python synthetic_experiments.py
# Clustering experiments
python real_experiment_clustering.py --download
# Figure 4a and 4b
python real_experiment_clustering.py -r 1000 -o ./results/clustering.csv -d olivetti digits -a kmeans
# Appendix E Figure 5
python real_experiment_clustering.py -r 1000 -s -o ./results/clustering_spectral.csv -d segment texture -a spectral
# Community detection experiment in Figure 4c
python real_experiment_community_detection.py -c 30 -r 100
# Examples for type II bias in Table 2
python test_biases.py