Nina Miolane

Refactor/clean module of discrete curves

12

There are a few tasks that remain to be done to clean-up or refactor the module of discrete curves: - [x] Address the comments from @Florent-Michel 's PR #1078 to...

### Describe the bug This bug is reported in PR #1622 . The geodesic unit-tests are skipped for TestElasticMetric even if: - no skip_ flag is put, - there are...

### Describe the bug The Frechet mean gives "Segmentation fault: 11" when computed with the SRV metric on a large dataset of curves, with many sampling points. The code to...

### Where should the content be modified? https://github.com/geomstats/geomstats/blob/master/geomstats/geometry/product_manifold.py https://github.com/geomstats/geomstats/blob/master/geomstats/geometry/product_riemannian_metric.py ### What should be modified? NFoldManifold should inherit from ProductManifold. NFoldMetric should inherit from ProductMetric. ProductManifold and ProductMetric should allow the...

### Where should the content be modified? _No response_ ### What should be modified? The sample code: ``` import geomstats.backend as gs from geomstats.geometry.discrete_curves import DiscreteCurves, R2 dim = 2...

### Where should the content be modified? https://github.com/geomstats/geomstats/blob/6f260f767fc06e2752a4a801557147004049943b/geomstats/geometry/discrete_curves.py#L224 ### What should be modified? The function `pointwise_inner_products` computes the inner-products of two "tangent vectors" at one base curve, where the tangent...

## What? Create a new notebook that explains how to use the Riemannian geodesic (squared) distance to create loss functions for regression on Riemannian manifolds. ## Where? The notebook should...

Adapt k-NN on manifolds with data points of ndim=2 (e.g. manifolds of matrices)

1

## What? The algorithm k-nearest neighbors on manifolds has a bug, see code snippet below. This prevents its use for manifolds whose points have ndim=2, e.g. manifolds of matrices. This...

Np, tf and torch initialize arrays with different default dtype

3

The following code: ``` import numpy as np import tensorflow as tf import torch a = np.array([1.]) b = tf.constant([1.]) c = torch.tensor([1.]) print(a.dtype, b.dtype, c.dtype) ``` gives: ``` float64...

The `einsum` function in the pytorch and tf backends needs to be generalized to handle these cases: - x, y not of the same dtype, - when there is no...