Advanced computational analysis of AIME problem technique classification. Accompanies: "A Multi-Scale Computational Analysis of Mathematical Structure, Embedding Geometry, and Topological Invariants in the AIME Corpus"
Paper: Nischay Kommisetty - Probing Technique-Structured Geometry in AIME Embedding Space_ A Computational and Statistical Analysis of Mathematical Technique Classification Across Four Decades of Competition Mathematics.pdf
embedding_geometry.py— Riemannian geometry of TF-IDF manifold; sectional curvature estimationmanifold_analysis.py— Intrinsic dimensionality, persistent homology (Betti numbers), geodesic distanceneural_probe.py— Multi-layer nonlinear probe; representational similarity analysis (CKA)information_theory.py— PMI/PPMI matrices, information gain, mutual information estimation (KSG)statistical_tests.py— Permutation testing, bootstrap CIs, FDR correction, power analysisfeature_geometry.py— Gradient-based feature attribution; SHAP values for LR coefficientstemporal_drift.py— Vocabulary drift analysis; year-stratified embedding trajectoriesrun_full_pipeline.py— End-to-end reproducible pipeline with seeding and logging
python run_full_pipeline.py --corpus AIME_labeled_full.csv --seed 42 --output results/