Closed bkloppenborg closed 11 years ago
There also appears to be a problem with the regular convex Chi approximation method. This happened around the time when getoifits was replaced by ccoifits and we switched to the non-convex approximation method. Commits around or before d9320d80ff04f61324716796e809c1b55153c7e4 probably caused this.
Since the non-convex chi2 kernels were introduced the log likelihood values appear to have increased dramatically. This needs to be addressed before version 1.0 is released.