Implement a periodic tree that maps points to "mirrors" #193
Conversation
I think this is because some points may be far from the query point, represented by some point in the canonical image, but close to a shifted one version of the query point. So you may need to query for the canonical version and the shifted one, and out of those find the nearest neighbors. |
|
Aha, I misunderstood the link I think. What I thought they did was to duplicate the input points to all the mirrors (which is what this implementation does) but instead they map the input points to a single "canonical image". In the first case you only need to query once but in the latter you might need to query multiple times (for example if the ball around the query point is outside the image). That is probably a better idea than what I did here. |
KristofferC
force-pushed
the
kc/periodic
branch
3 times, most recently
from
c57c643 to
37889e0
Compare
KristofferC
force-pushed
the
kc/periodic
branch
3 times, most recently
from
9e7266b to
465e86f
Compare
KristofferC
changed the title
- Fix periodic tree implementation with proper boundary validation and mixed dimension support - Add mirror box pruning optimization for KNN searches to skip irrelevant periodic images - Implement comprehensive tests covering mixed periodic/non-periodic dimensions and boundary cases - Add extensive periodic tree benchmarks to benchmark suite for performance tracking - Add detailed PeriodicTree documentation with examples and usage patterns to README The periodic tree now properly handles: - Mixed periodic/non-periodic dimensions (using Inf for non-periodic bounds) - Boundary validation ensuring all data points are within the periodic box - Performance optimization by pruning irrelevant mirror boxes during searches - Comprehensive error handling for invalid box configurations 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
2 participants
Similar to what is described in https://namdanalyzer.readthedocs.io/en/latest/kdTree/periodic_kdtree.html.
I don't understand why this is needed..
TODO:
Fixes #133