Pair-based Estimators of Infection and Removal Rates

This Python package is an AI-translated version of the R package github.com/sdtemple/peirrs.

Install

From PyPI (recommended)

pip install peirrs

From source (development)

Install from the local source directory using pip:

pip install -e .

Requires NumPy and SciPy for numerical computations.

Requirements

This package requires Python 3.6+ with the following dependencies:

• NumPy 1.19+
• SciPy 1.5+

I developed the package with versions:

• Python 3.9+
• NumPy 2.0+
• SciPy 1.7+

Usage

Estimate infection and removal rates with partially observed removal and infection times. The following functions are the ones you would likely use, in order of relevance:

Real data analysis

• peirrs.estimators.peirr_tau() - EM-based estimation
• peirrs.estimators.peirr_bayes() - Bayesian estimation with MCMC
• peirrs.estimators.peirr_bootstrap() - Bootstrap resampling
• peirrs.estimators.peirr_imputed() - Imputation-based estimation

Simulation experiments

• peirrs.simulate.simulator() - Core simulation wrapper

All functions have docstrings. As a result, you can get help for instance with:

from peirrs.estimators import peirr_tau
help(peirr_tau)
print(peirr_tau.__doc__)

There are also functions with the suffixes _multitype() and _spatial() for estimators with multiple classes and spatial kernels, respectively:

Multitype estimators (in peirrs.multitype.estimators_multitype)

• peirr_tau_multitype() - Class-specific EM estimation
• peirr_bayes_multitype() - Class-specific Bayesian MCMC
• peirr_bootstrap_multitype() - Class-specific bootstrap

Spatial tools (in peirrs.spatial)

• peirr_tau_spatial() - Spatial EM estimation
• peirr_bayes_spatial() - Spatial Bayesian MCMC
• simulate_distance_matrix() - Generate spatial distance matrices
• simulator_spatial() - Spatial simulation wrapper

The peirr_bootstrap() function does not provide confidence intervals but rather bootstrap samples. You can perform bias correction or interval estimation according to Wikipedia.

Warning !!!

• I used AI chatbots to translate this from an R package.
  • Mostly as a personal experiment ...
• PBLA functions are not available.
• Some light units tests are available.
  • Except for complete data, the _bayes functions are barely tested.
• Some spot reading resulted in:
  • Manually fixes to the utils.tau_moment function
  • Manually fixed to the _bayes functions
• I checked that some simulation study results were similar.
  • The results are qualitatively the same.
  • But, the estimator computes adds many small float numbers.
  • The snowball effect leads to some slight differences between Python and R estimates.
• I checked for sensible results in a small IPython notebook.