Code for DS 2020: Constrained Clustering Via Post-Processing

Introduction

This work proposed the novel approach by converting the output of many clustering algorithm to a cluster fractional allocation matrix (CFAM). Then, takes it as an optimization problem with requirement to satisfy (combined) different types of constraints.

In addition, we also shows practical application namely the attribute-level constraint and the (combined) fairness constraint.

Installation

Step 1: Clone the Code from Github

git clone git@github.com:dung321046/ConstrainedClusteringViaPostProcessing.git
cd ConstrainedClusteringViaPostProcessing/

Step 2: Install Requirements

Install conda

Install libs

conda install --yes --file requirements.txt 

Install gurobipy

conda config --add channels http://conda.anaconda.org/gurobi
conda install gurobi

Running pairwise, cluster-size and attribute-level costraints experiments

Step 0: Setup

export PYTHONPATH="./"

Download:

IDEC weights

More details for IDEC and DCC is here Deep Constrained Clustering.

Step 1: Generate dataset

Generate pairwise and cluster-size contraints

python generate_testsets/generate-pw-csize.py --data $DATA

Step 2: Run models

python models/encode-kmeans-post.py --data $DATA
python models/pw-csize-ilp.py --data $DATA --csize True
python models/attribute-level-ilp.py

Running Fairness Experiments

Step 1: Generate neighbor set for individual fairness constraints

python fairness/generate_neighbor.py 

Step 2: Run individual and/or group fairness

python fairness/run_individual_fairness.py
python fairness/run_group_fairness.py
python fairness/run_combine_fairness.py