f-dangel / curvlinops

PyTorch linear operators for curvature matrices (Hessian, Fisher/GGN, KFAC, ...)
https://curvlinops.readthedocs.io/en/latest/
MIT License
18 stars 8 forks source link
fisher ggn hessian linalg linear linearoperator operators pytorch scipy

Logo scipy linear operators of deep learning matrices in PyTorch

Python
3.9+ tests Coveralls

This library implements scipy.sparse.linalg.LinearOperators for deep learning matrices, such as

Matrix-vector products are carried out in PyTorch, i.e. potentially on a GPU. The library supports defining these matrices not only on a mini-batch, but on data sets (looping over batches during a matvec operation).

You can plug these linear operators into scipy, while carrying out the heavy lifting (matrix-vector multiplies) in PyTorch on GPU. My favorite example for such a routine is scipy.sparse.linalg.eigsh that lets you compute a subset of eigen-pairs.

The library also provides linear operator transformations, like taking the inverse (inverse matrix-vector product via conjugate gradients) or slicing out sub-matrices.

Finally, it offers functionality to probe properties of the represented matrices, like their spectral density, trace, or diagonal.

Installation

pip install curvlinops-for-pytorch

Examples

Future ideas

Other features that could be supported in the future include:

Logo mage credits