risdr
Regularised and Information-Theoretic Sufficient Dimension Reduction
Implements covariance-stabilised sufficient dimension reduction for continuous responses with information-theoretic structural dimension selection. Supported methods include sliced inverse regression, sliced average variance estimation, directional regression, and principal Hessian directions. Sample, ridge, Oracle Approximating Shrinkage, Ledoit-Wolf, and Maximum Entropy Covariance (MEC) estimators are provided alongside prediction, resampling, simulation, and diagnostic utilities. The sufficient dimension reduction methods build on Li (1991) <doi:10.1080/01621459.1991.10475035>, Li (1992) <doi:10.1080/01621459.1992.10476258>, and Li and Wang (2007) <doi:10.1198/016214507000000536>. Covariance shrinkage follows Olorede and Yahya (2019) <doi:10.48550/arXiv.1909.13017>, Ledoit and Wolf (2004) <doi:10.1016/S0047-259X(03)00096-4> and Chen et al. (2010) <doi:10.1109/TSP.2010.2053029>.
README
# Supplied research-result fixtures The summary CSV files were selected from the supplied `Thesis_Results_Adebiyi.zip` archive. The four EPI input matrices were supplied separately for v0.3.0. They support vignette examples, reproducible empirical analysis, and provenance checks. The `epi` directory contains `X_old.csv`, `X_new.csv`, `y_old.csv`, and `y_new.csv`, plus selected empirical comparison and repeated cross-validation summaries. Each predictor matrix has 180 rows and the same 70 columns. The `simulation` directory contains selected Simulation A, B1, and B2 summaries and rankings. Input-file SHA-256 digests are: - `X_old.csv`: `b18bf20cacb1f51245451edfc2384f4431d9ee2969f20ad1aa4696d7a6558b90`; - `X_new.csv`: `d03dcf03f0be248ed39405abf829e6fb6c1ce4b57770b44bda72732a79217be5`; - `y_old.csv`: `0200104325dd20ee7893b766c5df5a4a9138159db3f60978d717337548424f2b`; - `y_new.csv`: `d1c093a3095e9679e9bfc05cf5a5ea9168cec62f018793e0b4f33c957c5b4144`. Simulation A and B1 external-prediction columns have legacy status because the historical script independently generated the random-sparse basis in the training and test samples. Simulation B2 does not use an external test sample, but remains a supplied historical record pending the unified corrected rerun. Use `analysis/reproduce_simulations.R` from the repository archive to generate new corrected outputs. The script writes distinct filenames and does not overwrite these fixtures.
Versions across snapshots
| Version | Repository | File | Size |
|---|---|---|---|
0.3.1 |
rolling linux/jammy R-4.5 | risdr_0.3.1.tar.gz |
418.4 KiB |
0.3.1 |
rolling linux/noble R-4.5 | risdr_0.3.1.tar.gz |
418.3 KiB |
0.3.1 |
rolling source/ R- | risdr_0.3.1.tar.gz |
177.0 KiB |
0.3.1 |
latest linux/jammy R-4.5 | risdr_0.3.1.tar.gz |
418.4 KiB |
0.3.1 |
latest linux/noble R-4.5 | risdr_0.3.1.tar.gz |
418.3 KiB |
0.3.1 |
latest source/ R- | risdr_0.3.1.tar.gz |
177.0 KiB |
0.3.1 |
2026-04-23 source/ R- | risdr_0.3.1.tar.gz |
0 B |