giustogianni / ML_Higgbozoncompetition19

A ML competition on various classification tasks with Higg Bozon data from the CERN.

Machine Learning (CS-433)

---

Project 1, October 2019

Yann Mentha, Maxime Epars, Gianni Giusto

Dataset

The dataset is divided into a training and a testing set composed of 250’000 and 568’238 samples respectively and both having 30 features. The training set is paired with labels where each sample is associated to a category (−1 for background noise and 1 for the presence of a Higgs Boson).

Code architecture

The code is separated into 2 distinct files:

1. run.py
2. implementations.py

The run.py file contains the main and can be run in a terminal.

The implementations.py file contains all useful functions and is divided into 5 sections:

• "IMPLEMENTATIONS" contains the 6 working methods from the labs, that is: least_squares_GD, least_squares_SGD, least_squares, ridge_regression, logistic_regression and reg_logistic_regression.
• "UTILITARIES" contains functions that are called by the 6 methods from "IMPLEMENTATIONS" (e.g. gradient computation) and other various handy methods
• "DATA PROCESSING" contains all the functions that are used to process the data, perform feature engineering, ...
• "DATA VISUALIZATION" contains all the functions that are made to display the results in figures.
• "HELPERS" contains the function provided by the teaching team. They allow to load the data, predict the labels and create a
  submission file in .csv format.

Data Pipeline

Code execution

Run the following command line in the terminal : python3 run.py