GeneticProgramming_k-means_Clustering

Abstract:

Using Genetic programming to optimize the distance function for Clustering with k-means. The algorithm present in this repository was able to improve the results by up to 458% (v_score) in relation to the commonly used Euclidean distance. The algorithm is capable of adapting to any dataset with little customization.

Execution:

Initial requirements:

• Pytho 3.8.2
• Venv (sudo apt install python3.8-venv)
• pip (sudo apt install python3-pip)

Initial setup:

1. Create enviroment: python3 -m venv ./code/venv
2. Activate enviroment: cd code && source venv/bin/activate
3. Install requirements: pip install -r requirements.txt

Execute:

The optimized parameters are at the beginning of the main file.

1. Activate enviroment: cd code && source venv/bin/activate
2. Execute: python3 main.py
3. Exit: deactivate

Exit:

When executing the algorithm, the function of the best individual will be printed using variables x, y and z, in addition to the individual v-scored on test of the chosen dataset. A training_v-score.json file will also be generated, where all training results per generation will be recorded with the following order: best value, the average and the worst value.

Datasets:

• Breast Cancer Coimbra
• Glass Identification