Machine Learning from scratch in Odin.
This is a personal side project I work on Sundays only while pursuing my MSc in AI.

The goal:

• Learn ML internals by implementing algorithms from the ground up
• Explore Odin as a low-level systems language for numerical computing
• Document the journey for others who want to go beyond Python/NumPy
• Explore both supervised and unsupervised learning algorithms over time

• Vector type + basic ops (add, sub, scalar mul, dot product)
• Matrix type + basic ops (add, mul, transpose)
• Small numeric tests

• Implement y = Wx + b
• Mean Squared Error loss
• Gradient Descent optimizer
• Train/test on toy dataset

• Sigmoid activation
• Binary classification
• Cross-entropy loss

• Softmax function
• Multi-class classification

• Feedforward (1 hidden layer)
• Backpropagation
• Train on XOR

• Momentum optimizer
• Mini-batch training
• Matrix multiplication optimizations

• CSV loader
• Multilayer perceptron (MLP)
• MNIST (or subset)

odin-ml/
├── main.odin           # entry point (demo runner)
├── core/               # basic building blocks
│   ├── vector.odin
│   ├── matrix.odin
│   ├── math_utils.odin
│   └── notes.md
├── models/             # ML algorithms
│   ├── linear_reg.odin
│   ├── logistic_reg.odin
│   ├── nn.odin
│   └── notes.md
├── datasets/           # toy datasets & loaders
│   ├── csv_loader.odin
│   └── notes.md
├── optim/              # optimization algorithms
│   ├── gradient_descent.odin
│   ├── momentum.odin
│   └── notes.md
└── tests/              # unit tests
    ├── test_matrix.odin
    └── notes.md

Each directory includes a notes.md file where I jot down design decisions, bugs, and future ideas for that module.

Clone the repo and run:

odin run .

I only code on Sundays, so progress will be slow but steady.

The purpose is learning, not building a production-grade ML framework.

Expect bugs, incomplete features, and lots of iteration.

notes.md files track my thought process inside each module.

I don't know 🤷🏿‍♂️

I just want to learn Odin

Rarely used in ML → great opportunity to explore.

Beyond the roadmap above, I plan to explore:

• Unsupervised learning algorithms (clustering, PCA, dimensionality reduction)
• Reinforcement learning experiments
• Small-scale neural network experiments beyond one hidden layer
• Performance optimizations and possible GPU/parallelism exploration(Maybe I am overthinking but I am willing to explore this.)

MIT – free to use, learn from, and contribute.