Coverage for src/cvxcla/operators/__init__.py: 100%

1"""Covariance backend abstraction for the Critical Line Algorithm. 

2 

3The turning-point loop of the CLA touches the covariance matrix through a 

4small number of operations: full matrix-vector products, solves against the 

5free-asset block, and cross-products between free and blocked assets. This 

6package defines that contract as the ``CovarianceOperator`` protocol (see 

7:mod:`cvxcla.operators._core`), together with several implementations: 

8 

9- ``DenseCovariance`` / ``IncrementalDenseCovariance`` (:mod:`cvxcla.operators.dense`): 

10 adapters that reproduce the behaviour of a plain ``numpy`` covariance matrix, 

11 the latter maintaining the free-block inverse across turning points. 

12- ``FactorCovariance`` (:mod:`cvxcla.operators.factor`): a diagonal-plus-low-rank 

13 covariance ``Sigma = diag(d) + U @ Delta @ U.T`` whose solves go through the 

14 Woodbury identity, so no ``n x n`` matrix is ever materialised. Memory and 

15 per-solve cost are ``O(n * k)`` instead of ``O(n^2)``. 

16- ``GramCovariance`` (:mod:`cvxcla.operators.gram`): a sample covariance backed by 

17 the ``(T, n)`` data matrix, never forming ``Sigma``. 

18 

19See https://github.com/cvxgrp/cvxcla/issues/646 for the roadmap. 

20""" 

21 

22from ._core import CovarianceOperator, QuadraticForm, bordered_solve 

23from .dense import DenseCovariance, IncrementalDenseCovariance 

24from .factor import FactorCovariance 

25from .gram import GramCovariance 

26 

27__all__ = [ 

28 "CovarianceOperator", 

29 "DenseCovariance", 

30 "FactorCovariance", 

31 "GramCovariance", 

32 "IncrementalDenseCovariance", 

33 "QuadraticForm", 

34 "bordered_solve", 

35]