GeoTorch now is fully compatible with torch.nn.utils.parametrize and it uses it whenever possible.

More flexible initialisation

The current release implements a number of important changes

• Now it's possible to initialise a constraint to a fixed matrix by simply assigning to the constrained weight
• Complete rewrite of the logic that allowed to heavily simplify it
• Updated to be compatible with torch.nn.utils.parametrize, which will be part of PyTorch 1.9
• Added many new spaces
• More thorough testing. All the possible constructions are now tested for every manifold

Initial release

Implements the following manifolds:

• Rn(n): Rⁿ. Unrestricted optimization
• Sym(n): Vector space of symmetric matrices
• Skew(n): Vector space of skew-symmetric matrices
• Sphere(n): Sphere in Rⁿ. It is Sⁿ⁻¹ = { x ∈ Rⁿ | ||x|| = 1 }
• SO(n): Manifold of n×n orthogonal matrices
• Stiefel(n,k): Manifold of n×k matrices with orthonormal columns
• Grassmannian(n,k): Manifold of k-dimensional subspaces in Rⁿ
• LowRank(n,k,r): Variety of n×k matrices of rank r or less

And the following constructions:

• Manifold: Manifold that supports Riemannian Gradient Descent and trivializations
• Fibration: Fibred space π : E → M, constructed from a Manifold E, a submersion π and local sections of dπ
• ProductManifold: M₁ × ... × Mₖ