ethen8181/machine-learning

What it solves

This repository is a personal learning archive that helps both the author and readers understand machine learning concepts deeply by balancing theory and practice. It solves the problem of scattered knowledge by providing curated, well-organized notebooks that cover everything from basic algorithms to advanced deep learning architectures, along with practical deployment and evaluation strategies.

How it works

The content is organized as Jupyter notebooks, each focusing on a specific topic. Many notebooks implement algorithms from scratch using Python's scientific stack (numpy, scipy, pandas, matplotlib) to illustrate the underlying math, then show how to use established libraries like scikit-learn, TensorFlow, PyTorch, and others. The repository is continuously updated, and each notebook is also available as an HTML page for easy viewing. Coverage is extensive, including deep learning (CNNs, RNNs, transformers, fine-tuning, contrastive learning), model deployment (FastAPI, Kubernetes, vLLM), A/B testing and causal inference, time series forecasting, recommendation systems, and more.

Who it's for

It's for data science and machine learning practitioners who want to deepen their understanding beyond just calling library functions. It's also useful for students and self-learners who appreciate a mix of mathematical derivations and hands-on code. The tutorials range from beginner to intermediate/advanced, so it's suitable for someone with basic Python knowledge who wants to systematically explore machine learning topics.

Highlights

• Extensive collection of Jupyter notebooks covering over 20 topic areas, from linear regression to state-of-the-art LLM fine-tuning (e.g., DPO, GRPO).
• Balance between mathematical notation and implementation from scratch using numpy, numba, scipy, etc., with usage of popular libraries like scikit-learn, PyTorch, TensorFlow.
• Includes practical end-to-end projects (Kaggle competitions) and deployment examples with FastAPI, Kubernetes, and cloud services.
• Covers modern topics like transformer models, contrastive learning, CLIP, and LLM alignment techniques.
• Each notebook is also available as static HTML, making it easy to browse without running code.