Optimal Transport Tools
A Python toolbox for solving optimal transport problems with JAX-powered computational efficiency.
What is Optimal Transport Tools?
OTT is a Python toolbox for solving optimal transport problems, which involve finding optimal ways to move mass between probability distributions. It provides efficient implementations of algorithms like Sinkhorn and Greenkhorn, with JAX acceleration for computational performance. The library enables applications in machine learning, computer vision, and data science where measuring distances between distributions is essential.
Machine learning researchers and practitioners working with distributional data, computational geometry problems, or applications requiring Wasserstein distances. Data scientists and engineers needing efficient optimal transport implementations for large-scale problems.
Developers choose OTT for its JAX-powered computational efficiency, clean API design, and focus on differentiability—making it ideal for integration with modern machine learning pipelines. It offers a balance between research flexibility and production-ready performance.
Overview
OTT is a Python library that provides efficient implementations of algorithms for solving optimal transport problems. It focuses on computational performance through JAX acceleration, making it suitable for large-scale applications in machine learning and data science.
Key Features
- JAX-powered computations — Leverages JAX for automatic differentiation, vectorization, and GPU/TPU acceleration
- Comprehensive algorithm suite — Implements Sinkhorn, Greenkhorn, and other optimal transport solvers
- Differentiable transport — Supports end-to-end differentiable pipelines for integration with neural networks
- Geometric tools — Provides utilities for Wasserstein distances, barycenters, and Gromov-Wasserstein alignment
- Modular design — Clean API with separable components for custom algorithm development
Philosophy
OTT prioritizes computational efficiency and differentiability while maintaining a clean, research-friendly API that balances flexibility with performance.
Use Cases
Best For
- Computing Wasserstein distances between high-dimensional distributions
- Implementing differentiable optimal transport layers in neural networks
- Solving large-scale optimal transport problems with GPU acceleration
- Research on optimal transport algorithms and applications
- Calculating barycenters and Gromov-Wasserstein alignments
- Benchmarking optimal transport solvers in machine learning pipelines
Not Ideal For
- Projects not using JAX or requiring seamless integration with PyTorch/TensorFlow ecosystems
- Applications where optimal transport problems are small-scale and computational overhead isn't justified
- Teams seeking long-term support from an actively maintained repository without migration hassles
Pros & Cons
Pros
Leverages JAX for automatic differentiation and GPU/TPU acceleration, enabling efficient large-scale computations as highlighted in its focus on computational performance.
Supports end-to-end differentiable pipelines, allowing integration with neural networks for machine learning applications, which is a core feature for modern ML workflows.
Implements Sinkhorn, Greenkhorn, and other solvers with geometric tools for Wasserstein distances, providing a wide range of options for optimal transport problems.
Cons
Development stopped on the main GitHub branch in 2022, with migration to ott-jax, causing potential confusion and extra steps for users to access updated versions.
Requires JAX, which adds setup complexity and may not align with projects using other deep learning frameworks, limiting flexibility for some teams.
As a research-focused tool, it has fewer community resources and documentation compared to established libraries like POT, which can hinder adoption and troubleshooting.
Frequently Asked Questions
Related Projects
ALXGoogle Research
OryxProbabilistic reasoning and statistical analysis in TensorFlow
BRAXMassively parallel rigidbody physics simulation on accelerator hardware.
MctxMonte Carlo tree search in JAX
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.