Project 1: Kernelized Linear Classification

This project is part of the Machine Learning course of the DSE program at the University of Milan (UNIMI).
Detailed instructions for this project can be found here.
Student: Stefano Chiesa
Professor: Nicolò Cesa Bianchi

Abstract

This report presents an analysis of kernelized linear classification. I implemented several machine learning algorithms from scratch, including the Perceptron, Support Vector Machines (SVMs) using the Pegasos algorithm, and Regularized Logistic Regression. I analysed the impact of polynomial feature expansion and kernel methods using Gaussian and polynomial kernels. The performance of each model was evaluated based on $\ell_{0-1}$ loss, variance and runtime.

Contents

Kernelized_Linear_Classification.pdf: report on the implementation and performance of the algorithms /Code

• functions.py: This script contains the main functions used for the Kernelized Pegasos algorithm implementation
• test.ipynb: A Jupyter Notebook that shows the cleaning of the Data and the training and testing procedure of the algorithms

/Data:

• your_dataset.csv:the dataset used in test.ipynb to demonstrate the performance of the Kernelized Pegasos algorithm