powerlawrs 0.0.3

A fast package for analyzing Power-Law distributions in empirical data written in Rust.

powerlawrs

powerlawrs is a Python package for analyzing power-law distributions in empirical data. It is built on a high-performance Rust crate powerlaw, providing both speed and ease of use for Python users. The methodology is heavily based on the techniques and statistical framework described in the paper 'Power-Law Distributions in Empirical Data' by Aaron Clauset, Cosma Rohilla Shalizi, and M. E. J. Newman.

Features

• Parameter Estimation: Estimates the parameters (x_min, alpha) of a power-law distribution from data.
• Goodness-of-Fit: Uses the Kolmogorov-Smirnov (KS) statistic to find the best-fitting parameters.
• High Performance: Computationally intensive tasks are parallelized in the Rust core for significant speedups.
• Flexible API: Offers both a simple functional API for quick analyses and a class-based API for more detailed work.

Installation

Prerequisites

• Python 3.8+
• Rust (the package is built from Rust source)
• uv (this project uses uv for environment and package management)

Setup and Installation

1. Clone the repository:

   git clone https://github.com/aulichny3/powerlawrs_dev.git
   cd powerlawrs_dev
   

2. Create and activate a virtual environment: This project is configured to use the power virtual environment with uv.

   # Create the environment
   uv venv -p python3.8 power
   
   # Activate the environment
   source power/bin/activate
   

3. Install the package: To install the package in editable mode and include all development dependencies, run:

   # Install the package using maturin
   maturin develop
   
   # Install development dependencies
   uv pip install -r requirements.txt
   

   This installs powerlawrs in editable mode, so any changes you make to the source code will be immediately available.

Dependencies

• The core powerlawrs library has no Python dependencies.
• Development dependencies (for running the example notebook) are listed in requirements.txt and include:
  • jupyterlab: For running the example notebook.
  • numpy: Used in the example notebook.
  • polars: Used for data loading in the example notebook.

Usage

powerlawrs offers two primary ways to analyze your data: a simple functional API and a more detailed class-based API.

Functional API (Recommended)

The powerlawrs.fit() function is the most straightforward way to fit a power-law distribution to your data.

import powerlawrs
import polars as pl

# 1. Load your data into a list or Polars Series
# The data should be a 1-dimensional array of numbers.
file = "reference_data/blackouts.txt"
data = pl.read_csv(file, has_header=False).to_series().to_list()

# 2. Fit the data
fit_results = powerlawrs.fit(data)

# 3. Print the results
print(f"Alpha: {fit_results.alpha}")
print(f"X_min: {fit_results.x_min}")
print(f"KS Statistic: {fit_results.D}")
print(f"Tail Length: {fit_results.len_tail}")

Class-based API

For more fine-grained control, you can see the API examples in Notebooks/Example.ipynb.

Jupyter Notebook Example

The Notebooks/Example.ipynb file provides a detailed walkthrough of the package's functionalities. After installing the development dependencies, you can run it with:

# Make sure your virtual environment is active
source power/bin/activate

# Start Jupyter Lab
uv run jupyter lab

Limitations

1. Only the continuous case of the Pareto Type I Distribution is considered for parameter estimation, goodness of fit, and hypothesis testing at this time. The example data in the documentation is discrete, thus the results are only an approximation.
2. Domain knowledge of the data generating process is critical given the methodology used by this package is based on that proposed by the referenced material. Specifically the 1-sample Kolmogorov-Smirnov test is used for goodness of fit testing which assumes i.i.d data. Many natural processes data are serially correlated, thus KS testing is not appropriate.
3. This is highly alpha code; backwards compatibility is not guaranteed and should not be expected.
4. Many more known and unknown.

License

This project is licensed under either of:

• Apache License, Version 2.0, (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.