STROH

Yes, the ticket can be postponed.

Virtual leave-one-out formulae were implemented: https://sourceforge.net/p/kriging/hg/ci/517cadbebb01df3ce0094c42feab60d759a01449/ Remark: the current function uses stk_make_matcov to compute response covariance matrix. This will have to be changed by stk_covmat_response in the "covmat" branch.

In this revision https://sourceforge.net/p/kriging/hg/ci/c36bcb988f6aaef8e9c01fbbfd46721ea6b562ac , the "norminv" function was changed by an expression using "erfcinv" function. Is it better ?

Changes are posted, see STK revision: https://sourceforge.net/p/kriging/hg/ci/f9da32feb81b6843f5e34bcab0417886f0d62be7/

Here is the patch without generating random data. It follows the previously attached patch.

You find attached a correcting patch, which replaces the function integral by the use of the function trapz, introduced in Matlab before version R2006a. The new test works on Matlab R2013a and Octave version 4.2.1. Remark: this ticket is linked to ticket 71.

OK, I do not know what changed, but now, it works on Matlab version 8.1.0 and Octave version 4.2.1. Problem solved.

@stk_sampcrit_eqi\stk_sampcrit_eqi does not pass tests.