Let sample Z be the zip of two samples X and Y. It is guaranteed that the order in which Z traverses the elements of X is monotonic. Not strict monotonic, however: if the points p of some element i of X are divided over n >= 2 elements of Y, then element i is evaluated n times, each with a different subset s_j of points p and weights w. If the subsets (s_j)_j are consecutive slices, max(s_j) < min(s_{j+1}), then everything works fine. If not, for the j-th evaluation of element i points p[s_j] is used together with weights w[t_j] instead of w[s_j], where (t_j)_j is a sequence of consecutive slices such that the sizes of the subsets s_j and t_j are equal. This patch fixes this problem and adds a test for this situation.
joostvanzwieten
commented
1 year ago
Let sample
Zbe the zip of two samplesXandY. It is guaranteed that the order in whichZtraverses the elements ofXis monotonic. Not strict monotonic, however: if the pointspof some elementiofXare divided overn >= 2elements ofY, then elementiis evaluatedntimes, each with a different subsets_jof pointspand weightsw. If the subsets(s_j)_jare consecutive slices,max(s_j) < min(s_{j+1}), then everything works fine. If not, for thej-th evaluation of elementipointsp[s_j]is used together with weightsw[t_j]instead ofw[s_j], where(t_j)_jis a sequence of consecutive slices such that the sizes of the subsetss_jandt_jare equal. This patch fixes this problem and adds a test for this situation.Fixes #791