Sung or Danced? Rhythm, Meter, and Melody as Computational Discriminators in Aragonese Folk Music—Replication Bundle

Derived data, trained classifiers, and the full source code for the experiments reported in the paper. It is derivative-only: the underlying MusicXML score transcriptions of the Cancionero del Alto Aragón (Garcés Til et al., 1999) are not included, because the edition's rights holder has not granted redistribution. Everything needed to re-evaluate the released classifiers and inspect the derived features is here; re-running the pattern-mining stage from scratch requires obtaining the score files separately. root/├── README.md ← this file├── LICENSE ← CC BY 4.0├── CITATION.cff ← citation metadata├── REPRODUCE.md ← step-by-step reproduction guide├── data/│ ├── metadata.csv piece-level metadata (filename, genre, label)│ ├── labels.csv sung/danced target labels│ ├── features_pattern.csv pattern-level feature matrix (496 × 20)│ ├── features_global.csv global feature matrix (496 × 18)│ ├── pattern_vocabulary.csv abstract pattern vocabulary (2829 patterns)│ └── pattern_occurrences.csv piece × pattern occurrence counts├── models/│ ├── svc_pattern_final.joblib SVC on 10 α-selected pattern features (α=0.2, paper Table 1 headline)│ ├── svc_global_baseline.joblib SVC on 17 RFECV-selected global features│ ├── xgboost_pattern.joblib XGBoost on the same 10 pattern features│ └── feature_selection.json RFECV + α-sweep trace, selected feature names└── code/ ├── pyproject.toml ├── .python-version ├── configs/baseline.yaml ├── src/ (preprocessing, mining, classification) └── experiments/ (run_baseline, run_viewpoint_ablation, run_compression_analysis) The paper and this release agree on the following pipeline: the published edition contains 528 pieces; excluding the 12 unlabelled "Varios" entries leaves 516 analytical pieces; twenty of those produce no patterns under the canonical mining configuration and are dropped from the classification analysis, leaving **496 pieces** (265 sung,231 danced). The feature matrices and models in this release are built on those 496 pieces.