predictr

predict + reliability, in other words: A tool to predict the reliability.

predictr is a Python package for Weibull-based life data analysis (reliability engineering). It covers parameter estimation, bias-correction, confidence bounds, and publication-ready Weibull plots in a single, consistent API.

Installation

pip install predictr

Requires Python >= 3.6.

Quick start

from predictr import Analysis

failures = [0.4508831, 0.68564703, 0.76826143, 0.88231395, 1.48287253, 1.62876357]

weibull = Analysis(df=failures, bounds='fb', show=True)
weibull.mle()

print(weibull.beta, weibull.eta)  # shape and scale estimates

This fits a two-parameter Weibull distribution via Maximum Likelihood Estimation, adds Fisher confidence bounds, and renders the probability plot below.

See it in action

A few of predictr's capabilities, from bias-corrected estimates to comparing entire distributions.

Bias-corrected estimates	Confidence region, multiple levels

Ranked by AIC	Distributions compared

Regression: survival per covariate profile	Regression: goodness of fit with verdict

Main features

Parameter estimation

• Uncensored and type I / type II right-censored two-parameter Weibull distribution
• Maximum Likelihood Estimation (MLE) and Median Rank Regression (MRR)
• Bx-life calculator
• Normal, LogNormal and Exponential distributions, alongside Weibull
• Non-parametric Kaplan–Meier kaplan_meier() and Nelson–Aalen nelson_aalen() from the failure / suspension lists (no DataFrame), with pointwise bands and step plots

Lifetime regression (covariates)

• Weibull accelerated failure time (AFT) and Cox proportional hazards (Cox PH) models
• Uncensored and right-censored data, Efron/Breslow tie handling, Wald / profile-likelihood / bootstrap (parametric and non-parametric) bounds
• summary(), coefficient forest plot, survival-curve prediction per covariate profile (with pointwise and simultaneous confidence bands)
• Goodness of fit: goodness_of_fit() with a good / marginal / poor verdict (concordance, Cox–Snell slope, proportional-hazards test), Cox–Snell / martingale / deviance residuals, plot_gof(), check_ph()
• Non-parametric descriptors: the same kaplan_meier() / nelson_aalen() as Analysis, here also splittable by a covariate (by=), plus plot_km() / plot_na()
• Accelerated life testing: named aging laws (stress_model= — Arrhenius, inverse power, Eyring, Coffin–Manson), physical parameters (Ea, n) with CIs, acceleration_factor(), raw-unit predictions, plot_stress_life(), check_shape()
• Monte-Carlo power_analysis() and sample_size()

Bias-correction

• C4 method (reduced bias adjustment)
• Hirose and Ross method
• Parametric and non-parametric bootstrap correction (mean, median, trimmed mean)

Confidence bounds

• Fisher bounds
• Likelihood Ratio bounds (Weibull, Normal, LogNormal)
• Beta-Binomial bounds
• Monte Carlo Pivotal bounds
• Parametric and non-parametric bootstrap bounds
• Exact chi-square bounds (Exponential)

Plots

• Probability plots with all relevant statistics in the legend
• Multiple fits overlaid in one figure, for design comparisons
• Contour plots for the joint confidence region of shape and scale, with support for multiple confidence levels per dataset
• Distribution comparison: fit every supported distribution to one dataset, ranked by AIC or Anderson-Darling, plus a combined PDF plot

See the class documentation for the full method and parameter reference, including censored-data and bias-correction examples.

Documentation and links

• Documentation
• PyPI
• Changelog
• Citation / Zenodo
• Discussions

Citing predictr

If you use predictr in academic work, please cite it via its Zenodo DOI. See docs/citation.md for details.

License

MIT — see LICENSE.txt.

Contacte me

If you have any questions and / or suggestions, don't hesitate to contact me.