Minimum Enclosing Ball and Anomaly Detection using variants of Frank-Wolfe

Table of contents

• General Info
• Group Members
• Project Structure
• Setup
• Run

General Info

This is a final project for Optimization for Data Science course in University of Padua.

Task

Minimum Enclosing Ball and Anomaly Detection

• analyze in depth the papers and the theoretical results
• implement the Away-Step Frank Wolfe [Lacoste-Julien et al]
• implement PFW algorithm described in [Tsuji et al., Algorithm 1]
• implement the first algorithm in the paper [Yildirim, 2008].
• choose two datasets and use the MEB defined over them to find anomalies (new points that are out of the MEB)

Group Members:

• Dejan Dichoski
• Marija Cveevska
• Suleyman Erim

Project Structure

├── configurations
│   ├── experiment1_Uniform
│   ├── experiment2_Gaussian
│   ├── experiment6_CustomerChurn
│   ├── experiment9_BreastCancer
├── datasets
│   ├── CustomerChurn
│   └── BreastCancer
├── runs (results will be saved here)
│   ├── graphs
│   ├── output.yaml
│   └── log files
├── papers
│   ├── Lacoste_Julien et al.
│   ├── Tsuji et al.
│   ├── Yıldırım
├── src (source codes)
│   ├── FrankWolfeVariants.py
│   ├── data_generation.py
│   ├── execution.py
│   ├── logger.py
│   ├── plotting.py
│   ├── utils.py
│   └── partials/template
├── main.py
├── setup.py
├── requirements.txt
├── Report.pdf
├── Presentation.pptx
├── README.md
└── .gitignore

Setup

$ conda update conda -y
$ conda create -p venv python==3.10 -y
$ conda activate venv/

Go to requirements.txt file and uncomment symbol (#) before "-e ." Then,

$ pip install -r requirements.txt

Run

Choose a configuration file(.yaml) to run an experiment configs/"experiment.yaml"

$ python main.py --cfg "exp0_Default.yaml"

Results will be saved in runs/experiment folder