DNA Binding Sites Prediction

Second Project of the Machine Learning course in collaboration with the EPFL Laboratory for Computation and Visualization in Mathematics and Mechanics.

File Structure

• code - Folder containing training utilities
  • metrics_helper.py - Utilities for performance measures.
  • preprocessing_helper.py - Helper for preprocessing the data.
  • sampling_helper.py - Implementation of the undersampling and oversampling experimented strategies.
  • Best Run Regression.ipynb - Notebook for reproducibility of regression results.
  • Best_Run_classification.ipynb - Notebook for reproducibility of classification results.
• report - Report folder
  • report.pdf - Report pdf file
• requirements.txt - Requirements text file

Dataset

The original dataset was provided by the EPFL Laboratory for Computation and Visualization in Mathematics and Mechanics.

Installation

To clone the following repository, please run:
git clone --recursive https://github.com/johnmavro/Machine-Learning-P2.git

Requirements

Requirements for the needed packages are available in requirements.txt. To install the needed packages, please run:
pip install -r requirements.txt

Reproducibility

Optimal Parameters (Regression)

To reproduce the results we obtained for the regression task, please use a XGBoost Regressor model with the following hyperparameters:

parameters = {'subsample' : 0.9,
              'n_estimators' : 500,
              'max_depth' : 20,
              'learning_rate' : 0.01,
              'colsample_bytree' : 0.8,
              'colsample_bylevel' : 0.6}

Optimal Parameters (Classification)

To reproduce the results we obtained for the classification task, please use a Random Forest Classifier with the following hyperparameters:

parameters = {'n_estimators': 400,
              'min_samples_split': 5,
              'min_samples_leaf': 1,
              'max_features': 'auto',
              'max_depth': 20,
              'bootstrap': False}

Report

The report can be found in pdf format in the report folder.

Authors

• Federico Betti
• Ioannis Mavrothalassitis
• Luca Rossi