ott
ott copied to clipboard
ott-jax
Optimal transport tools implemented with the JAX framework, to get differentiable, parallel and jit-able computations.
Unbalanced Sinkhorn results in negative `reg_ot_cost`. ```python import jax from ott.core.sinkhorn import Sinkhorn from ott.geometry.pointcloud import PointCloud import jax.numpy as jnp from ott.core import LinearProblem n = 1000 dim =...
Addressing issues [# 110](https://github.com/ott-jax/ott/issues/110) and [#105](https://github.com/ott-jax/ott/pull/105), this is a small pull request adding more docstrings and comments to the existing code 👀.
Here are the implementations of the update for the lr-sinkhorn algorithm: - New init using kmeans - Rescaling of the gamma - New stopping criterion - New implementation for the...
https://github.com/ott-jax/ott/blob/710f807f17b15edc6bc941ad4f760656059e6beb/ott/tools/transport.py#L128 solver wrapper missing init args, silent error if passing init_dual_a, init_dual_b in kwargs
_TL;DR in this [colab](https://colab.research.google.com/drive/13uYiraKBxau5m4CTiFwFwPjJWjRJXkyj?usp=sharing) we provide an example for our failure in obtaining a valid mapping using `low-rank`._ **problem setup**: In this example data set we look into mapping spatial...
Better implementation of rescaling of transport matrix and epsilon for unbalanced gromov wasserstein. Fix for regulariser in cost_unbalanced_correction. This is a first commit towards fixing UGW in response to issue...
This refers to issue #156. This PR introduces a few changes: - old`Euclidean` is moved to `SqEuclidean` and continues wokring with naive **2 and vdot - `Euclidean` is now the...
**Describe the bug** We chose originally to implement the squared Euclidean distance as `jnp.sum(x**2,axis=-1) + jnp.sum(y**2,axis=-1) - 2 * jnp.vdot(x,y)`. Although this works with `power=2.0` (which leaves it unchanged) this...
Hi, Thanks for this amazing library, while exploring the API docs I noticed that the side panel is too full and not very helpful in the readability of docs. I...
