Stable matrix parametrization — companion code

Code and data to reproduce the results in "A stable matrix parametrization for multivariate Ornstein-Uhlenbeck processes" (H-SSBP: Hurwitz Smooth Spectral Block Parametrization). Message-passing likelihood and gradient calculations are implemented in a branch of BEAST X; this repository holds everything else: benchmark code, simulation/application configs, compact outputs used for the paper, and the analysis/plotting scripts that turn them into every figure and table in the manuscript.

Layout

environment/    how to build BEAST, Python/R/Julia dependencies
benchmarks/     Julia kernel/stability benchmarks feeding the 2 benchmark figures below
figures/        turns benchmarks/ CSVs into the manuscript's benchmark figures
simulations/    K=5 MAP simulation study: generators, summarizers, and compact results
applications/   BitMEX trade-data example and the Anolis phylogenetic example
run_local_beast.sh   minimal single-machine runner: java -jar beast.jar ... on any XML here

This repository was trimmed to what the compiled manuscript and supplement (main.tex / Supplement.tex) actually cite — see "Excluded from this repository" below for what was cut and why, and each subdirectory's own README for details specific to it.

Every subdirectory has its own README with the run settings, BEAST commit, compact manuscript inputs, and regeneration commands.

Quickstart

# 1. Build BEAST (see environment/BEAST.md)
git clone https://github.com/beast-dev/beast-mcmc.git && cd beast-mcmc
git checkout 88a7b1e6a14f06778693581b527984d05509b814 && ant build && cd ..

# 2. Python dependencies for simulation generation/summarization helpers
#    (R/Quarto figure dependencies are listed in environment/README.md)
pip install -r environment/requirements.txt

# 3. Regenerate a figure from the results already shipped here, e.g. the
#    benchmark figures (fast — no BEAST/Julia run needed, just the R figures
#    pipeline over the checked-in CSVs):
cd figures && quarto render make_benchmark_figures.qmd --to html

Where raw BEAST outputs are shipped, the corresponding README records the XML, seed, and regeneration command. The final simulation study ships compact processed summaries plus scripts that regenerate the XMLs and raw outputs. To reproduce a run from scratch, use run_local_beast.sh with the BEAST jar you built and the XML you want to rerun.

Manuscript figure/table -> path in this repo

Checked directly against \includegraphics/\input in main.tex and Supplement.tex (not just file presence in manuscript/figures/ — several files there turned out to be unreferenced drafts; see below):

Manuscript output	Path	Used in
ou_exp_adjoint_cached_benchmarks	figures/make_benchmark_figures.qmd (+ benchmarks/branch/results/*.csv)	main.tex
ou_ssbp_vs_lyapunov_param_stability_summary	same	Supplement.tex
supp_simulations_combined_rmse_final_battery_lambda0p1_r25_best_likelihood	simulations/ - see simulations/README.md	main.tex
bitmex_jan2022_j20_trade_7h_cartesian_map_row	applications/bitmex/	Supplement.tex; summarized in main.tex
anole_phylogeny_drift_matrix_row_10m_fresh_2m	applications/anole/	main.tex

License

MIT — see LICENSE.