pysrat 0.2.2

Python package for software reliability analysis and modeling

pysrat

pysrat is a Python package for fitting and analysing NHPP (Non-Homogeneous Poisson Process) software reliability models. It provides a clean scikit-learn–style API (model.fit(data), params_, llf_, aic_) and high-performance EM core implementations exposed from C++ via pybind11.

Features

• Scikit-learn–style estimators (model.fit(data), params_, llf_, aic_)
• Fast EM updates implemented in C++ and exposed via pybind11
• A collection of classical NHPP models and CF1 (canonical phase-type)
• Plotting helpers for visualization (plot_mvf, plot_dmvf, plot_rate)
• Multi-factor models (MFLogitNHPP, MFProbitNHPP, MFCloglogNHPP)
• A Poisson regression–based framework (pr_nhpp_fit)

Installation

From source (recommended during development):

pip install pysrat

Requirements

• Python ≥ 3.9
• A C++20-capable compiler (required for some extension modules such as CF1)

Optional / developer tools

• ninja (recommended build backend for faster CMake builds)
• pytest for development and tests
• Python headers and development toolchain for building native extensions

Quick start

import numpy as np
from pysrat.data.nhpp import NHPPData
from pysrat.nhpp.models import ExponentialNHPP
from pysrat.nhpp.plot import plot_mvf

# Create NHPP data from grouped intervals
data = NHPPData.from_intervals(time=[1, 1, 1, 1], fault=[0, 1, 0, 5])

# Fit model
model = ExponentialNHPP().fit(data)

print("Parameters:", model.params_)
print("Log-likelihood:", model.llf_)
print("AIC:", model.aic_)

# Plot mean value function
plot_mvf(data, model)

Working with data

pysrat accepts several NHPP data formats via pysrat.data.nhpp.NHPPData:

• Interval/grouped data: NHPPData.from_intervals(time=..., fault=..., type=...)
• Counts per interval: NHPPData.from_counts(fault=...)
• Exact fault times: NHPPData.from_fault_times(times=..., te=... )

Available models

Common models are exported under pysrat.nhpp.models:

• ExponentialNHPP
• GammaNHPP
• Pareto2NHPP
• TruncatedNormalNHPP
• LogNormalNHPP
• TruncatedLogisticNHPP
• LogLogisticNHPP
• TruncatedExtremeValueMaxNHPP
• LogExtremeValueMaxNHPP
• TruncatedExtremeValueMinNHPP
• LogExtremeValueMinNHPP
• CanonicalPhaseTypeNHPP (CF1)

CF1 example

from pysrat.data.nhpp import NHPPData
from pysrat.nhpp.models import CanonicalPhaseTypeNHPP

data = NHPPData.from_intervals(time=[1, 2, 1.5], fault=[1, 0, 2], type=[0, 1, 0])
model = CanonicalPhaseTypeNHPP(3).fit(data)
print(model.params_)

Plotting

Use the plotting helpers from pysrat.nhpp.plot:

from pysrat.nhpp.plot import plot_mvf, plot_rate
plot_mvf(data, model)
plot_rate(data, model)

You can pass a dictionary of models to compare multiple MVFs:

models = {
    "Exp": ExponentialNHPP().fit(data),
    "Gamma": GammaNHPP().fit(data),
}
plot_mvf(data, models)

Model comparison

Compare models using aic_ or llf_:

from pysrat.nhpp.models import ExponentialNHPP, GammaNHPP
data = NHPPData.from_counts([0, 1, 0, 5])
m1 = ExponentialNHPP().fit(data)
m2 = GammaNHPP().fit(data)
best = min((m1, m2), key=lambda m: m.aic_)
print("Best model:", best.name)

Examples

See the examples/ notebooks included in the repository for end-to-end workflows (basic fitting, comparison, CF1 usage and plotting):

• examples/example1.ipynb — Exponential NHPP
• examples/example2.ipynb — CF1 fitting and comparison
• examples/example3.ipynb — Multi-factor models (MFLogitNHPP)
• examples/example4.ipynb — Poisson regression–based NHPP fitting (pr_nhpp_fit)

Contributing

Contributions are welcome. Please open an issue describing the change or a PR with tests and documentation updates.

License

This project is distributed under the terms of the MIT License. See LICENSE.