Antoine Collas

Add Complex valued manifolds.

7

Hello, I have added the ComplexEuclidean and the HermitianPositiveDefinite manifolds. They are implemented with their corresponding unit tests. Do not hesitate if you have any remark!

Fix error produced by `jax.config` and complex valued dot product

1

This PR fixes several problems with `jax` to make the tests work again: - fixes the `jax.config` problem, see issue #260 - raises the minimum version of python from 3.7...

Tests are not running anymore because `config` cannot be imported from `jax.config` with the latest version of `jax`. This is likely due to https://github.com/google/jax/pull/19895 . Libraries relying on `pymanopt` cannot...

[WIP] Add numerics functions

12

WIP: Add numerics functions This PR follows https://github.com/pymanopt/pymanopt/pull/252

Center data when using Coral

2

Center data when using Coral algorithm. See https://arxiv.org/pdf/1612.01939

[ON_HOLD] Geodesic Flow Kernel

3

This PR is a WIP implementation of [Geodesic Flow Kernel](https://projet.liris.cnrs.fr/imagine/pub/proceedings/CVPR2012/data/papers/261_O2D-03.pdf). This method requires the [Generalized SVD](https://en.wikipedia.org/wiki/Generalized_singular_value_decomposition), and it is not implemented either in numpy or Scipy. The `transferlearning` library uses...

`skada.deep` does not support binary classification because `DomainAwareNet` inherits from `skorch.classifier.NeuralNetClassifier`. To do classification with skorch, it should inherits from `skorch.classifier.NeuralNetBinaryClassifier`. See: https://skorch.readthedocs.io/en/latest/classifier.html

`_check_y_masking` is used in some functions such as CircularValidation but not everywhere. Some adapters can be used with `y_target` not masked.

`DomainAwareDataset` is not exposed in the doc api