Closed dcjones closed 8 years ago
@@ master #22 diff @@
==========================================
Files 3 3
Lines 644 683 +39
Methods 0 0
Messages 0 0
Branches 0 0
==========================================
+ Hits 600 639 +39
Misses 44 44
Partials 0 0
Powered by Codecov. Last update 085f48b...93a15ee
I saw https://github.com/BioJulia/Bio.jl/issues/340 and it prompted me try to figure out why IntervalTrees does so poorly there.
These changes help make it competitive. But there is still a big issue. The intervals tested in that benchmark contain entries that span entire chromosomes.
This is the worst case for the intersect algorithm in IntervalTrees, causing it to become basically a linear search. I'm not sure what do about that. I assumed, maybe incorrectly, when writing this that extremely long intervals were rare. This is maybe an inherent shortcoming of IntervalTrees that suggests we should use NCList.
Feel free to leave this open and I'll add commits if I have any epiphanies.