In this PR, I have adapted the approach that is taken to find the variance from covariance matrix and mean. Now instead of using likelihood sampling, we produce a range of random samples from the Gaussian process and apply GLS (in a vectorised fashion) to all of these samples. This has involved the introduction of the rtol parameter, which is associated with the tolerance of the matrix inversion for the construction of the Gaussian process. Additionally, the tests have been updated and the documentation (including methodology) modified to reflect this change.
In this PR, I have adapted the approach that is taken to find the variance from covariance matrix and mean. Now instead of using likelihood sampling, we produce a range of random samples from the Gaussian process and apply GLS (in a vectorised fashion) to all of these samples. This has involved the introduction of the
rtol
parameter, which is associated with the tolerance of the matrix inversion for the construction of the Gaussian process. Additionally, the tests have been updated and the documentation (including methodology) modified to reflect this change.