qweibull-cr

Reproducible research package for:

q-Weibull Competing Risks under Weak Identifiability: Bayesian Inference via the No-U-Turn Sampler

Authors: Huirong Cao (Langfang Normal University, Langfang) and Fuchang Wang (University of Emergency Management, Sanhe). Corresponding author: Huirong Cao ([email protected]). Target journal: Quality and Reliability Engineering International (QREI, Wiley).

Repository layout

qweibull-cr/
  code/     Python (numpy/scipy/pandas/matplotlib/pymc) scripts
  data/    voltage.csv, simulation result JSONs, posterior quantities
  figures/ fig1..fig10 (PDF vector + PNG)
  paper/   main.tex, references.bib, compiled main.pdf, USG.cls, build support

Main scripts

• qweibull.py, bayes_model.py, cr_bayes_model.py — q-Weibull likelihood, cause-specific competing-risk log-posterior and analytic gradient.
• nuts.py — NUTS / HMC sampler with Hessian-based preconditioning and smooth-barrier wall.
• penalized_mle.py — unpenalised MLE vs penalised MLE / MAP decomposition.
• fisher_info.py — Fisher information matrix, eigenvalues and condition number across truncation designs.
• param_compare.py — quantitative comparison of the (q, log η, log β) vs (q, log λ, log β) parameterisations.
• pymc_baseline.py — mature-library (PyMC/ArviZ) reference fit.
• sim_eta_beta.py, prior_center_sensitivity.py, boundary_fraction.py, sample_size.py — recovery and sensitivity studies.
• ppc_voltage.py, dependent_cr.py — posterior predictive check and dependent-competing-risks misspecification experiment.
• run_voltage_bayes.py, run_voltage_bayes_weibull.py, verify_cif.py, compute_posterior_quantities.py — real-data analysis.

Reproduce

1. Install Python 3.10+ with numpy scipy pandas matplotlib (and pymc arviz for the external baseline).
2. Run the scripts in code/ to regenerate the JSON results in data/ and the figures in figures/.
3. Compile paper/main.tex with XeLaTeX (the Wiley NJD/USG class loads local STIX fonts in paper/Fonts/): xelatex → bibtex → xelatex → xelatex, or run paper/build_pdf.ps1.

The voltage data are derived from the weibulltools R package and are included as data/voltage.csv.

License

Code is released for reproducibility under the MIT License; the paper text and figures are © the authors.