Reuters news categorization with Pegasos and Perceptron
Project 1 for the Machine Learning exam.
In this project we address a classification problem on a dataset of news taken from Reuters. We will implement the Pegasos algorithm from scratch and we evaluate the performance with different values of lambda and T in a one-vs-all encoding. Then we evaluate accuracy through external cross validation. After this analysis, we will also implement another algorithm from scratch: the Perceptron. We consider the Perceptron's predictor that averages all the models for different numbers of epochs and we will compare it against Pegasos.
Be aware that you will need to extract the zip file containing the data since it exceeded the size allowed by GitHub.