How lucirie/ml-regression-models works
Project type and purpose
• This is an open-source, educational ML toolkit demonstrating core regression algorithms (univariate linear, multivariate linear, polynomial, and logistic) implemented from scratch in Jupyter notebooks using only NumPy, Pandas, and Matplotlib.
• Its primary goal is to illustrate the math and mechanics of each algorithm, with data loading, training loops, and visualization tightly coupled in the notebooks.
File structure
• Top-level directories:
• unilinear/ (train.csv, test.csv, unilinear.ipynb)
• multilinear/ (Student_Performance.csv, multilinear.ipynb)
• nonlinear/ (icecream.csv, nonlinear.ipynb)
• logistic/ (wdbc.data, logistic.ipynb)
• Each directory bundles:
a) A dataset file (CSV or raw data)
b) A single notebook orchestrating data ingestion, model implementation, training, evaluation, and plotting
• No separate library modules—logic is embedded in notebooks.
README for additional insights
• Emphasizes implementing algorithms “from scratch” with core Python scientific libraries.
• No external ML frameworks; focus is on algorithmic intuition, gradient descent, cost visualization, decision boundaries.
Diagramming the architecture
a. Main components
• Data Sources (CSV/WDBC files)
• Data Loader & Preprocessor (in-notebook code for reading files, normalization)
• Model Implementations (functions or classes for each regression type)
• Training & Optimization (gradient descent loops, cost function computations)
• Evaluation & Visualization (plots of cost over iterations, decision boundary, fit lines)
• External Libraries (NumPy, Pandas, Matplotlib)
• Jupyter Notebook Runner (driver that ties everything together)
b. Relationships and data flow
2) Loader → Model Implementation (feeds preprocessed arrays)
3) Model Implementation ↔ Training & Optimization (iteration loop)
4) Training → Evaluation & Visualization (outputs metrics & plots)
5) Notebook Runner sits atop all components, invoking each step
6) Each component depends on the external libraries (NumPy, Pandas, Matplotlib)
c. Architectural patterns & design principles
• Layered pipeline pattern: data ingestion → preprocessing → model → evaluation
• Modular separation by algorithm type, each in its own directory
• Reuse of core utility routines (e.g., normalization, cost computation) across notebooks
d. Key technologies
• Python 3.x
• NumPy (array ops, linear algebra)
• Pandas (dataframes, CSV I/O)
• Matplotlib (plotting)
• Jupyter Notebook (interactive execution)