GitHubHelp

fcampelo/Imbalanced_epitope_prediction_2025

A pipeline that allows to test different imbalanced classification approaches systematically (initially developed by Summer intern Rodrigo Buzarco)

★ 0Forks 0GitHub ↗Compare

README

2025 Summer Internship

ML Pipeline that allows to test different imbalance classification approaches systematically.

Overview

Stage Script Purpose
Load CSV src/data_loader.py Streams the semicolon-delimited CSV in 50 k-row chunks; keeps every feat_ column; casts features to float32 and the label (Class) to int8.
Group split src/split_data.py GroupShuffleSplit (80 / 20, random_state) so each Info_group appears only in train or test.
Pipeline src/pipeline.py sampler → StandardScaler → PCA → classifier; PCA keeps 50 components (~85 % variance). Evaluates accuracy, precision, recall, F1, ROC-AUC. Saves plots.
Entry point main.py CLI / .env front-end. Lets you choose sampler, classifier, random seed and dataset path.

Set up the environment

Before utilizing this pipeline, ensure that you have activated a Python environment with the specified requirements:

python -m venv venv
source venv/bin/activate        # Windows - .\venv\Scripts\activate
pip install -r requirements.txt

Run the Pipeline

To execute this pipeline, a single entry point has been provided main.py which can be executed in two different ways:

python main.py

or:

python main.py -default_values

If you wish to use the -default_values flag, you must first populate the .env with the required parameters:

NAME=<pipeline name>        # Used to identify the run.
FILE_PATH=<csv_file_path>
CLASSIFIER=<classifier>
IMBALANCE_HANDLER=<sampler>
RANDOM_STATE=<random_state>

Results

The pipeline will return the following information:

  1. Accuracy
  2. Precision
  3. Recall
  4. F1-Score
  5. ROC-AUC
  6. Graphs:
    • confusion_matrix_<pipeline_name>.png
    • precision_recall_curve_<pipeline_name>.png
    • roc_curve_<pipeline_name>.png

Issues