Kahypar Versions Save

Skipping v1.3.4 to align the version with the python package version

• Don't print a warning for size one hyperedges #195
• Export kahypar_supress_output function #190
• fix tests that assume input validation is enabled #188
• Fix compiler warnings #187
• Update boost version to 1.83 #186
• Config for cut-net optimization with kKaHyPar-E #181
• Update pybind and remove STATIC linkage #179
• Update googletest #173
• Integrate Kahypar Shared Resources #176
• Fix Titan23 Instances #172
• Add input validation #167

• Input validation for main binary and interfaces (https://github.com/kahypar/kahypar/pull/167)
• Performance and quality fix for Titan23 benchmark instances and hypergraphs with power-law edge-size distributions (https://github.com/kahypar/kahypar/pull/172)
• Allowing to write out partition in the python interface (https://github.com/kahypar/kahypar/pull/160)

• Fix block weights computation for fixed vertex partitioning (see https://github.com/kahypar/kahypar/pull/150)

• Fix segmentation fault for single-block partitioning (k=1).

This release contains some performance bug fixes that caused KaHyPar to become slow when partitioning instances using a large imbalance factor. Furthermore, this will be the last release that does not include the advanced balancing strategies for partitioning weighted instances (see https://github.com/kahypar/kahypar/pull/70).

As described in our upcoming SEA paper and in the corresponding technical report [1], KaHyPar now uses Weighted Hyperflow Cutter (WHFC) for flow-based refinement. The previous flow-based refinement algorithm [2,3] has been removed to reduce code complexity and compilation time.

[1] https://arxiv.org/abs/2003.12110 [2] https://dl.acm.org/doi/abs/10.1145/3329872 [3] https://drops.dagstuhl.de/opus/volltexte/2018/8936/

This KaHyPar release contains the source code that was used in the dissertation of Sebastian Schlag.