How linusjuni/mathematical-modelling-linear-classification works
Project type and purpose
• This is a research‐oriented data‐science pipeline for binary classification (pneumonia vs. healthy) on chest X-ray images.
• There is no client‐server or deployed web service; instead, it’s a modular set of scripts and utilities to preprocess data, extract features, train/evaluate models, and persist results.
File structure
• Top-level directories:
• data/ – raw MedMNIST training and test image folders
• models/ – serialized model files (*.joblib)
• src/ – all source code modules and experiment scripts
• In src/:
• part_1.py, part_2.py, part_3.py (and “_own_data_split” variants) orchestrate three experimental pipelines (raw‐pixels, Sobel‐gradients, HOG)
• model_utils.py, data_utils.py, evaluation_utils.py, visualization_utils.py – reusable functions
• individual model implementations (baseline.py, random_forest.py, ada_boost.py, gradient_boost.py, multi_layer_perceptron.py)
• helper notebooks (cat_dog_example.ipynb) and comparison scripts
• Config/build/deployment: none (experiments run via command‐line Python)
README for additional insights
• The README describes three “parts” focusing on different feature strategies, how to run each part, where models are saved, and how to load them via model_utils.py
• Emphasizes PCA as an optional dimensionality reduction step, visualizations, and hyperparameter tuning
Detail
• Both training and inference (loading models for new data) flows
• Multiple experiment entry points (scripts vs. comparison modules)