Euclid

Euclid is a lightweight, educational deep learning framework built from the ground up using only NumPy or CuPy. Its goal is to provide an intuitive, easy-to-use API while exposing the inner workings of modern neural networks.

Unlike large machine learning frameworks, Euclid is designed to be understandable, extensible, and easy to experiment with.

Features

• Automatic differentiation (Autograd)
• CPU (NumPy) and GPU (CuPy) backends
• Sequential model API
• Common neural network layers
  • Dense
  • Conv2D
  • MaxPool2D
  • Flatten
  • Dropout
• Activation functions
  • ReLU
  • Sigmoid
  • Tanh
  • Softmax
• Loss functions
  • Cross Entropy
  • Mean Squared Error (MSE)
  • Huber Loss
• Optimizers
  • SGD
  • RMSProp
  • Adam
• Built entirely in Python using only NumPy/CuPy

Philosophy

Euclid aims to bridge the gap between educational neural network implementations and production machine learning libraries.

The project focuses on:

• Readable source code
• Minimal dependencies
• Performance where possible
• Learning how deep learning frameworks work internally

Whether you're interested in understanding backpropagation, building custom layers, or experimenting with new neural network architectures, Euclid provides a simple and easily editable foundation.

Installation

pip install euclid-ml

Example

import euclid as Euclid

model = Euclid.Sequential([
    Euclid.Dense(784, 128),
    Euclid.ReLU(),
    Euclid.Dense(128, 10),
    Euclid.Softmax(),
])

optimizer = Euclid.Adam(
    model.parameters(),
    learning_rate=1e-3
)

loss = Euclid.CrossEntropy()

GPU Support

Euclid can run on either CPU or GPU depending on your configuration.

• CPU: NumPy
• GPU: CuPy

Simply configure the desired backend and the same code runs on either device.

Why Euclid?

Euclid was created to answer a simple question:

How do modern deep learning frameworks actually work under the hood?

Instead of hiding implementation details behind millions of lines of optimized C++ and CUDA, Euclid implements everything in Python using NumPy and CuPy, making it easy to explore, modify, and extend.

Roadmap

• Automatic differentiation
• Dense layers
• Convolutional neural networks
• GPU backend
• Optimizers
• Recurrent neural networks
• Transformers
• Mixed precision training
• Model serialization
• Distributed training

Contributing

Contributions, bug reports, feature requests, and discussions are welcome.

If you'd like to contribute, feel free to open an issue or submit a pull request.

License

This project is licensed under the MIT License.