low dimension data (dimension 4) problem occur in "Parallel TpTree Partition begin" step

Open pclin1212 opened this issue 6 years ago • 2 comments

for now, the index build and index search go well in dataset which dimension is bigger than 8. However, when i use the random dataset 10M * 4, there is a "Segmentation fault". the detail position is as follow: Start to build BKTree 1 1 BKTree built, 10000001 10000000 build RNG graph! Parallel TpTree Partition begin Segmentation fault (here).

So, i wonder whether there exist a parameter that limits the input dataset dimension?

Dec 20 '19 08:12 pclin1212

or may be in the codes, there exist a illegally access to the 5th dimension.

Dec 20 '19 09:12 pclin1212

I found the problem in the parameter setting in BKT: DefineBKTParameter(m_pGraph.m_numTopDimensionTPTSplit, int, 5L, "NumTopDimensionTpTreeSplit"). I change the 5L->4L, and it works.

Dec 20 '19 09:12 pclin1212