Demystifying the black box of Machine Learning and Core Algorithms.
This repository documents my transition from traditional backend development (Golang/Node.js) into AI, serving as a practical library for core algorithms and low-level derivations.
Facing the current AI landscape where "theory is obscure, but coding is just calling APIs," I decided to return to the code itself, building a "Progressive" practical system for every core algorithm:
- Demystifying the Black Box (Pure NumPy): Ditching all AI frameworks to manually derive matrix calculus and reverse-engineer the math with low-level code, proving that complex formulas are just paper tigers.
- Industrial Benchmarking (Modern AI Frameworks): After mastering the low-level mechanics, I provide equivalent implementations using PyTorch and other industrial-grade frameworks to ensure robust performance in real-world production environments.
Whether you are:
- 🎓 A student struggling with obscure textbooks, looking for "plain English" derivations.
- 💻 A software engineer looking to pivot into AI and needing to solidify your algorithmic foundation.
- 🚀 A geek tired of just "importing libraries", who wants to truly understand the core mechanics and build real-world applications.
Welcome to your AI core algorithm practice sandbox.
Quick Teaser: You can thoroughly master the low-level backpropagation of neural networks via the Formula Derivation Guide, and get your hands dirty by running the pure NumPy neural network following the Quick Start Guide. Training 5 Epochs takes only 0.6 seconds, with accuracy soaring past 96%!
This project embraces modern Python toolchains, utilizing the blazing-fast uv for package and project management.
# 1. Clone the repository and navigate to the directory
git clone git@github.com:smiletrl/machine_learning.git
cd machine_learning
# 2. Sync and install all dependencies (uv will automatically create a .venv virtual environment for you)
uv sync
# 3. Activate the current project's virtual environment
source .venv/bin/activate| No. | Project Name | Core Technologies | Status | Video Tutorial |
|---|---|---|---|---|
| 01 | Neural Network From Scratch (MNIST) | MNIST, NumPy Backprop | 🟢 Done | Coming Soon |
| 02 | Classic Machine Learning Core Algorithms | Decision Trees, SVM, K-Means, PCA | 🟡 Planned | - |
| 03 | Convex Optimization in Practice | Gradient Descent, Duality, Loss Surfaces | 🟡 Planned | - |
| 04 | Transformer Architecture Demystified | Self-Attention, Positional Encoding | 🟡 Planned | - |
Follow my journey from traditional engineering to hardcore AI algorithms. Give it a ⭐️ if it inspires you!