Coverage for jetgp/full_degp_sparse/__init__.py: 100%

1#!/usr/bin/env python3 

2# -*- coding: utf-8 -*- 

3""" 

4Sparse Cholesky variant of the DEGP model. 

5 

6Replaces the dense Cholesky factorisation in the NLML with a sparse 

7inverse-Cholesky factor U built using the paper's geometric sparsity 

8criterion: dist(x_P(i), x_P(j)) <= rho * l(j), where P is the MMD 

9(maximin) ordering and l(j) is the fill-distance at step j. 

10 

11The sparsity pattern is computed once at model initialisation and is 

12independent of the kernel hyperparameters. 

13""" 

14