geoews 0.1.0

Information-geometric early warning signals (KL rate and geodesic acceleration).

geoews

geoews is a pip-installable Python package for information-geometric early warning signals.

This package extracts the canonical implementations from the source research repository:

• sliding-window Gaussian fitting
• KL divergence rate between consecutive Gaussian windows
• Fisher-Rao step distance (univariate exact, multivariate midpoint approximation)
• geodesic acceleration indicator

Install

pip install .

Quick start

import numpy as np
from geoews.windows import estimate_gaussian_params
from geoews.indicators import kl_divergence_rate, geodesic_acceleration

x = np.sin(np.linspace(0, 12, 2000))
times, mus, sigmas = estimate_gaussian_params(x, window_size=50, step=1)

kl = kl_divergence_rate(mus, sigmas)
acc = geodesic_acceleration(mus, sigmas, cumul_window=30)

Canonical constants

• COVARIANCE_REGULARIZATION = 1e-6

Development

pip install -e ".[dev]"
pytest