Efficient Recursive Estimation for Right-Censored Data with Iterate Averaging

Description

This dataset contains the empirical data, simulation results, and code accompanying the manuscript "Efficient Recursive Estimation for Right-Censored Data with Iterate Averaging". It includes: - Real dataset: kidney transplant survival data (n=863, censoring rate 83.8%) used in Section 5.6. - Monte Carlo results: MSE vs gain (stable regime), instability boundary (c>=7.5), transient regime, two-parameter Weibull experiment, methods comparison (MLE, EM, unaveraged SA, Kiefer-Wolfowitz), and sensitivity analyses (warm-up offset, projection radius). - All result files are provided as CSV or JSON, with dictionaries explaining each variable. - Python code to reproduce every table and figure (Tables 2–3, Figures 6–7, 9, and the real-data streaming application). All experiments use fixed random seed 2024 for full reproducibility. The real dataset is a publicly available benchmark from Klein & Moeschberger (2003), redistributed here for reproducibility. The code and derived results are released under CC BY 4.0. For detailed instructions, see README.md and the codebook files (data_dictionary.csv, results_dictionary.csv).

Steps to reproduce

1. Create Python environment (3.12) and install dependencies: pip install -r code/requirements.txt 2. Reproduce the real-data results (Section 5.6): python code/sim_real_data.py python code/make_fig_realdata.py Expected output: batch MLE ≈ 0.0429 /yr, recursive estimate ≈ 0.0430 /yr. 3. Reproduce the gain-sweep (Figure 7) and instability boundary: python code/sim_mse_vs_c.py python code/make_fig_mse_vs_c.py python code/sim_instability_boundary.py 4. Reproduce Tables 2 and 3: python code/sim_table2_mle_comparison.py python code/sim_table3_transient.py 5. Reproduce two-parameter Weibull experiment (Figure 6): python code/sim_weibull2d.py 6. Reproduce methods comparison and sensitivity analyses: python code/sim_methods_comparison.py python code/sim_sensitivity_n0R.py python code/sim_sensitivity_warmup_cmax.py All scripts use numpy.random.default_rng(2024); re-running yields identical numbers (up to floating-point). See README.md for full details.

Institutions

• National University of Uzbekistan

  Tashkent, Tashkent

Categories

Licence