HIDEKI-SQ/intelligence-relativity: v2.3.0 - ARI-based SP_clu for all layout types
Abstract
What's Changed SP_clu Algorithm Update Adjusted Rand Index (ARI): SP_clu now uses ARI instead of simple label matching ARI is invariant to label permutation and corrects for chance agreement References: Hubert & Arabie (1985), Rand (1971) SP_clu Scope Change Now computed for all layout types (previously returned 1.0 for non-"cluster" layouts) SP_total reflects true three-component average across all experiments Updated Results (n=1,000 trials) | Experiment | Condition | SP (v2.2.0) | SP (v2.3.0) | |------------|-----------|-------------|-------------| | sp12 | Topology p=0.7 | 0.600 | 0.343 | | sp12 | Metric k=0.7 | 0.816 | 0.719 | | sp30 | λ=0.6 (Synth) | 0.574 | 0.301 | Key Findings Preserved ✅ Topology dominance (topology < metric) SP-SSC independence λ trade-off (SP↓ as SSC↑) Rotation invariance Migration Notes Breaking change: SP_total values differ from v2.2.0 SP_adj and SP_ord unchanged Re-run experiments to obtain v2.3.0 results Full Changelog: https://github.com/HIDEKI-SQ/intelligence-relativity/compare/v2.2.0...v2.3.0
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 25 January 2026