POT

Why we included this project

POT is the standard Python toolkit when your problem is genuinely about optimal transport, whether that means Wasserstein distances between distributions, aligning point clouds, or adapting a model across domains. It bundles a large family of solvers behind one clean API: exact linear programs, entropic Sinkhorn iterations, Gromov-Wasserstein, and unbalanced or partial OT, plus fast one-dimensional and circular variants and barycenter routines. Because the solvers are differentiable and accept PyTorch, JAX, TensorFlow, NumPy, and CuPy arrays, you can drop them straight into a training loop as loss functions instead of treating transport as a post-processing step. That combination of breadth and framework support makes it a practical fit both for researchers prototyping new alignment methods and for engineers who would rather use a battle-tested implementation than build their own. The documentation and example gallery are detailed enough to move from a concrete problem, say matching two point sets, to a working call without much friction.