Higgs Boson Machine Learning Challenge

Machine Learning - Project 1

Tshtsh_club: Marie Anselmet, Sofia Dandjee, Héloïse Monnet

Run the project

1. Make sure that Python >= 3.7 and NumPy >= 1.16 are installed
2. Download the train and test data sets from Kaggle competition dataset, and put train.csv and test.csv into a data\ folder.
3. Go to script\ folder and run run.py. You will get submission.csv for Kaggle in the submission\ folder.

cd script
python run.py

Script files

proj1_helpers.py

• sigmoid: Sigmoid function.
• load_csv_data: Loads data from a csv file.
• compute_f1_score, predict_accuracy: Computes the accuracy and the F1 score of a prediction.
• predict_labels: Generates class predictions for a linear or a logistic regression.
• build_k_indices, cross_validation : Generate the training and validation data for cross-validation.
• classify: Converts the (-1,1) of a label vector into (0,1), to use for the logistic regression.
• batch_iter: Generates a mini-batch for a dataset.
• create_csv_submission: Creates a csv output file for submission to Kaggle.

cost.py

• compute_loss: Computes the loss by mse for linear regression.
• logistic_loss: Compute the loss by negative log likelihood for the logistic regression.
• reg_logistic_loss : Compute the regularized logistic loss by negative log likelihood.

compute_gradient.py

• compute_gradient: Computes the gradient for the linear gradient descent.
• logistic_gradient: Compute the gradient for the logistic gradient descent.
• reg_logistic_gradient: Compute the gradient for the regularized logistic gradient descent.

data_helpers.py

• get_jet_samples: Divides the input data depending of their jet values.
• clean_data, standardize : Standardizes data, removes undefined values and features with a null standard deviation.
• augment_data, build_model_data,build_poly_all_features: Augment the data by building polynomial features.

implementations.py

• least_squares_GD: Linear regression using gradient descent.
• least_squares_SGD: Linear regression using stochastic gradient descent.
• least_squares: Least squares regression using normal equations.
• ridge_regression: Ridge regression using normal equations.
• logistic_regression: Logistic regression using stochastic gradient.
• reg_logistic_regression: Regularized logistic regression using stochastic gradient descent.

run.py

Script to produce the same .csv predictions used in the best submission on the Kaggle platform.

report.pdf

The complete description of the project and the interpretation and discussion of the results can be found in this file.