Fix a subtle bug in DBSCAN noise marking #525
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
aprokop
added
bug
Situation: a core point is connected to only boundary points. However, those boundary are connected to other core points. So it may happen that the boundary points are assigned to other clusters, resulting in a cluster with a single point (core). This test makes sure that we don't accidentally label it as noise, as its label is likely to be itself.
dalg24
reviewed
May 27, 2021
src/ArborX_DBSCAN.hpp
Outdated
Comment on lines
208
to
227
| if (is_special_case) | ||
| { | ||
| // Cannot use CCSCore as it always returns true | ||
| Kokkos::parallel_for("ArborX::dbscan::mark_noise", | ||
| Kokkos::RangePolicy<ExecutionSpace>(exec_space, 0, n), | ||
| KOKKOS_LAMBDA(int const i) { | ||
| if (cluster_sizes(labels(i)) == 1) | ||
| labels(i) = -1; | ||
| }); | ||
| } | ||
| else | ||
| { | ||
| DBSCAN::DBSCANCorePoints<MemorySpace> is_core{num_neigh, core_min_size}; | ||
| Kokkos::parallel_for("ArborX::dbscan::mark_noise", | ||
| Kokkos::RangePolicy<ExecutionSpace>(exec_space, 0, n), | ||
| KOKKOS_LAMBDA(int const i) { | ||
| if (cluster_sizes(labels(i)) == 1 && !is_core(i)) | ||
| labels(i) = -1; | ||
| }); | ||
| } |
There was a problem hiding this comment.
I reverted that patch out of curiosity and it looks like the unit tests are still passing.
There was a problem hiding this comment.
Yeah, I did the same. I think some permutation of the points in the test may trigger the failure in Serial, but it's really hard to get to it. It did fix the failure when verifying 500K subset of HACC points, run as
$ ./ArborX_DBSCAN.exe --binary --filename hacc.arborx --print-dbscan-timers --core-min-size 5 --cluster-min-size 1 --eps 0.042 --impl fdbscan-densebox --max-num-points 500000 --verify
and it also fixed divergence in the number of clusters and number of points in clusters when running HACC problem using different algorithms (as long as --cluster_min_size=1 was also specified).
|
CI failure is not a failure, but a warning from a container build. |
dalg24
reviewed
May 28, 2021
dalg24
approved these changes
May 28, 2021
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Situation: a core point is connected to only boundary points. However,
those boundary are connected to other core points. So it may happen that
the boundary points are assigned to other clusters, resulting in a
cluster with a single point (core).
The bug was that those single core clusters were marked as noise.
The fix is ugly as heck, but I don't know how to do better.