CS-433 Machine Learning - Project 1

Team : machine-learners

Team members:

• Madeleine HUEBER
• Adrian MARTINEZ LOPEZ
• Duru BEKTAS

Getting started

Project description

This project is part of the course CS-433 Machine Learning at EPFL. The goal of the project is to predict if someone will have a cardiovascular disease based on a set of answers to a medical questionnaire. The dataset used comes from the Behavioral Risk Factor Surveillance System (BRFSS) and contains for each person the answers to 330 questions. The target variable is binary and indicates if the person has a cardiovascular disease or not. For this project we first did some data preprocessing to clean the data and handle missing values. Then we did some feature selection to reduce the number of features. Finally, we trained a logistic regression model to predict the target variable.

More details about the project can be found in the project paper project1_report.pdf.

Repository structure

data/: Directory containing datasets.

• processed_x_test.npz: Processed test dataset in compressed NumPy format.
• processed_x_train.npz: Processed train dataset in compressed NumPy format.
• processed_y_train.npz: Processed train labels in compressed NumPy format.
• test_dataset.npz: Test dataset in compressed NumPy format.
• train_dataset.npz : Training dataset in compressed NumPy format.
• train_targets.npz : Labels for the training dataset in compressed NumPy format.

dataset_text_analysis/: Directory containing auxiliar code, not used in the run script, related with the text processing of the dataset. It is not necessary to run these files, we provide them to give visibility on the data processing steps.

• codebook15_llcp.txt: Dataset Report describing every feature and their possible value labels.
• dataset_text_analysis.py: Script that reads the dataset report and outputs a dictionary with weird values for every feature together with their right replacements (77 -> Nan). This dictionary is stored as a json file called var_abnormal_values.json
• var_abnormal_values.json: Example of the json file generated by the script. This json is then manually modified to remove mistakes in the data processing and stored as a variable (config.ABNORMAL_FEATURE_VALUES) in the src/config.py file

src/ : Main directory for source code modules.

• config.py: Configuration file containing paths, settings, and helper data structures for the project.
• data_cleaning.py: Module for cleaning data and handling outliers or missing values.
• data_preprocessing.py: Functions for preprocessing data before training, such as normalization or encoding.
• evaluation.py: Module to evaluate model predictions using various metrics.
• feature_engineering.py: Contains functions to create and transform features.
• feature_type_detection.py: Detects data types of features and helps with automated preprocessing.
• helpers.py: Utility functions for tasks like saving files, managing logs, etc.
• model.py: Contains functions for training, validating, and predicting with the model.

Project Root:

• .gitattributes and .gitignore: Git configuration files for version control, specifying files to include or exclude.
• README.md: Project documentation with instructions on setup, usage, and structure.
• implementations.py : Functions from the part 1 of the project
• run.py: Main script to execute the full pipeline, from data loading and preprocessing to model training and evaluation. If the processed dataset is already in the data folder, the preprocessing step will be skipped. Othwerwise, the original dataset will be loaded and preprocessed. Both datasets are present in the data folder.
• hyperparameter_selection.py: Script to perform hyperparameter selection with 5-fold cross-validation.
• project1_report.pdf : The project paper explaining our process

Installation

To use our model on this project, you will first need to clone the repository

git clone https://github.com/CS-433/ml-project-1-machine-learners/

The code required to have these packages installed :

• numpy
• matplotlib

Running the model

To run our model, you will need to run the following command in the terminal:

python run.py --seed 42 --gamma 0.1 --max_iters 1000 --lambda_ 0 --undersampling_ratio 0.2

It will preprocess the data, train the model and output the predictions in the form of a csv file. The predictions will be saved as submission.csv in the root folder.

Arguments

You can specify the following arguments when running the model:

Argument	Description	Type	Default
--seed	Set the seed for deterministic results	int	42
--gamma	Learning rate for training	float	0.1
--max_iters	Maximum number of iterations	int	1000
--lambda_	Regularization parameter	float	0
--undersampling_ratio	Undersampling ratio to balance the classes	float	0.2

Parameters exploration

To do some parameters exploration, you can run the following command in the terminal:

python hyperparameter_selection.py

It will test different values for the hyperparameters listed above and ouput which values of each hyperparameter give the best F-1 score using a 5-fold cross-validation.