Parallel Genetic Fuzzy k-Modes Clustering Algorithm

This is based on the paper "A genetic fuzzy k-Modes algorithm for clustering categorical data" (https://www.sciencedirect.com/science/article/pii/S0957417407005957).

This project parallelizes the algorithm using CuPy and mpi4py. Most of the calculations are vectorized using CuPy. mpi4py is used to run multiple groups of chromosomes in parallel.

The results show significant speedup against the NumPy implementation.

The branches are organized as follow:

1. main branch: NumPy implmentation.
2. parallel branch: CuPy implementation
3. MPI branch: mpi4py + CuPy implementation

Running the Code

Run scripts/main.py with the following command for main and parallel branch.

python3 main.py

For MPI branch,

mpirun -n N python3 main.py

Note that the MPI built and used needs to be CUDA-aware. https://www.open-mpi.org/faq/?category=buildcuda

Reference

Gan, G., Wu, J., & Yang, Z. (2009). A genetic fuzzy k-Modes algorithm for clustering categorical data. Expert Systems with Applications, 36(2), 1615-1620.