Homemade Machine Learning: Python Implementations of Popular Machine Learning Algorithms with Jupyter Notebook Demos

Introduction

The Homemade Machine Learning repository ( https://github.com/trekhleb/homemade-machine-learning) provides pure‑Python implementations of classic machine‑learning algorithms together with Jupyter Notebook demonstrations that illustrate the underlying mathematics, model training, parameter tuning, and result visualization.

Supervised Learning

Regression

Regression predicts continuous values by fitting a line, plane, or hyper‑plane to the training data. The project includes:

Linear Regression – mathematical derivation, reference implementation, and three notebooks: single‑variable regression (GDP → happiness index), multivariate regression (GDP + freedom index), and nonlinear regression using polynomial and sinusoidal features.

Classification

Classification assigns discrete labels to inputs. The project provides:

Logistic Regression – theory, reference code, and four notebooks: binary linear‑boundary classification (Iris), binary non‑linear classification (micro‑chip effectiveness), multi‑class classification on MNIST handwritten digits, and multi‑class classification on Fashion‑MNIST.

Unsupervised Learning

Clustering

Clustering discovers structure in unlabeled data. The repository implements the K‑means algorithm with:

mathematical background, reference implementation, and a notebook demonstrating K‑means clustering on the Iris dataset.

Anomaly Detection

Anomaly (outlier) detection identifies rare events by modeling the main data distribution. The project includes a Gaussian‑distribution‑based detector with:

theoretical explanation, reference code, and a notebook that detects server‑operation anomalies such as latency spikes.

Neural Networks

Multilayer Perceptron (MLP)

The MLP section treats neural networks as a framework for handling complex inputs. It provides:

mathematical overview, reference implementation, and two notebooks: MLP on MNIST handwritten digit recognition and MLP on Fashion‑MNIST clothing classification.

Learning Prerequisites

Install Python (version 3.7 or newer recommended).

Install required Python packages: pip install -r requirements.txt Launch Jupyter Notebook locally or on a remote server and open the desired notebook.

Datasets

All example datasets are stored in the repository’s data directory: https://github.com/trekhleb/homemade-machine-learning/tree/master/data

Illustrations