In the supportVectorRegression(...) function of LandmarkLBF.cpp there is a typo. lambda and _upperbound were declared as a size 1 array, but it was supposed to be a size l array! This can be verified by looking at the "original" code the algorithms was based of. And indeed, on github, it is almost impossible to tell the difference between l and 1. This typo results in some nan values and occasional crashes due to segfaults (not suprising, considering the array size is too small!).
In the supportVectorRegression(...) function of LandmarkLBF.cpp there is a typo. lambda and _upperbound were declared as a size 1 array, but it was supposed to be a size l array! This can be verified by looking at the "original" code the algorithms was based of. And indeed, on github, it is almost impossible to tell the difference between l and 1. This typo results in some nan values and occasional crashes due to segfaults (not suprising, considering the array size is too small!).
https://github.com/scikit-learn/scikit-learn/commit/2b60d329770206a32e1cb8d1bf78bf05141e3033#diff-bf60bef97c003384f56a1fd92fe102f6ac1eea40298e911dd95b4d28601bbc14 (line 830 in linear.cpp)