uhg 0.1.0

Universal Hyperbolic Geometry Library for PyTorch

Universal Hyperbolic Geometry (UHG) Library

A PyTorch library for hyperbolic deep learning using Universal Hyperbolic Geometry principles. All operations are performed directly in hyperbolic space without tangent space mappings.

Installation

Basic Installation

pip install uhg

With GPU Support

pip install uhg[gpu]

CPU-Only Version

pip install uhg[cpu]

Development Version

pip install uhg[dev]

Documentation Tools

pip install uhg[docs]

Quick Start

import uhg
import torch

# Create hyperbolic tensors
manifold = uhg.LorentzManifold()
x = uhg.HyperbolicTensor([1.0, 0.0, 0.0], manifold=manifold)
y = uhg.HyperbolicTensor([0.0, 1.0, 0.0], manifold=manifold)

# Compute hyperbolic distance
dist = uhg.distance(x, y)

# Create a hyperbolic neural network
model = uhg.nn.layers.HyperbolicGraphConv(
    manifold=manifold,
    in_features=10,
    out_features=5
)

# Use hyperbolic optimizer
optimizer = uhg.optim.HyperbolicAdam(
    model.parameters(),
    manifold=manifold,
    lr=0.01
)

Features

• Pure UHG implementation without tangent space operations
• Hyperbolic neural network layers and models
• Hyperbolic optimizers (Adam, SGD)
• Hyperbolic samplers (HMC, Langevin)
• Graph neural networks in hyperbolic space
• Comprehensive documentation and examples

Platform Support

• Linux (all major distributions)
• macOS (including Apple Silicon)
• Windows
• Docker containers
• Splunk environments

Documentation

Full documentation is available at https://uhg.readthedocs.io/

License

This project is licensed under the MIT License - see the LICENSE file for details.

Citation

If you use UHG in your research, please cite:

@software{uhg2023,
  title = {UHG: Universal Hyperbolic Geometry Library},
  author = {Bovaird, Zach},
  year = {2023},
  url = {https://github.com/zachbovaird/UHG-Library}
}