Skip to content

Quadratic time complexity with bins using toeplitz covairance matrices

Anarya Ray requested to merge scalable_inference into main

Created by: AnaryaRay1

Reparametrizing the GP's dependent variable into bin-indices makes the covariance matrices take a Toeplitz form. The Schur algorithm Stewart, 1997, for Cholesky factorizing a Toeplitz matrix reduces the net complexity of the correlated inference to O\left(\sum_{\theta}n_{bins,\theta}^2\right). This algorithm is now implemented in gppop as an experimental module.

Merge request reports