Bayesian Software Reliability Growth Models (Static vs. Time-Varying Defect Detection)

Aperta
#899 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 2 giorni

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
48/100
Tipo di issue
Documentazione
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
jupyter-notebook, python

Direzione di ricerca

Read the linked working notebook gist and compare its proposed models with the related reliability and Poisson regression notebooks named in the issue. Build a case study around the Musa System 1 data, covering both NHPP models, model comparison, release metrics, and posterior predictive checks; done means the notebook explains these analyses and validates the arrival trajectory.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

proposal

Notebook proposal

Title: Case Study: Bayesian Software Reliability Growth Models (Static vs. Time-Varying Defect Detection)

Why should this notebook be added to pymc-examples?

Currently, pymc-examples includes hardware survival analysis (e.g. case_studies/reliability_and_calibrated_prediction.ipynb using Weibull/LogNormal for component lifespans), but lacks any case studies covering software reliability engineering (SRE) and defect arrival processes.

This notebook demonstrates a complete Bayesian case study modeling software defect discovery as a Non-Homogeneous Poisson Process (NHPP):

  1. Model 1 (Goel-Okumoto): Assumes constant detection efficiency per fault ($b$), yielding an exponential decay in defect arrivals: $m(t) = a(1 - e^{-bt})$.
  2. Model 2 (Yamada Delayed S-Shaped): Relaxes the constant-efficiency assumption by modeling a time-varying detection efficiency $b(t) = \frac{b^2 t}{1 + bt}$ to capture tester learning curves and ramp-up, yielding an S-shaped discovery trajectory: $m(t) = a(1 - (1+bt)e^{-bt})$.
  3. Out-of-Sample Model Comparison (az.compare): Compares predictive accuracy using PSIS-LOO, demonstrating why Goel-Okumoto outperforms Yamada on projects with immediate high discovery.
Image
  1. Actionable Release Metrics: Direct posterior estimation of undetected defects remaining in the codebase ($N_{\text{remaining}}(T) = a \cdot e^{-bT}$, strictly positive) and future operational mission reliability $R(T, \Delta t) = \exp[-(m(T+\Delta t) - m(T))]$ to guide SLA release gating.
  2. Validation via PPC: Posterior predictive checks confirming that the model faithfully captures the arrival trajectory.
Image

The case study uses the canonical Musa System 1 benchmark (136 observed software failures over 24.6 CPU hours from Bell Laboratories).

Suggested categories:

  • Level: Intermediate
  • Diataxis type: Case study / How-to guide

Related notebooks

  • case_studies/reliability_and_calibrated_prediction.ipynb: Hardware survival analysis with right-censored time-to-failure. This proposal focuses on software defect counting processes (NHPP).
  • generalized_linear_models/GLM-poisson-regression.ipynb: Demonstrates standard linear Poisson regression, whereas this case study covers physical non-linear mean-value functions, operational reliability decay, and model comparison.

Working Draft

Lingua principale
Python
Stelle
402
Fork
327
Merge medio
5g 10m
PR unite (30g)
12

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di pymc-devs/pymc-examples

Tutte le issue di pymc-devs/pymc-examples

Issue simili

Altre issue su Python

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.