A selective and fast engine for exploratory landscape analysis.

Documentation · Quick start · API reference · Changelog · Contributing

orivex is a successor to pflacco for exploratory landscape analysis (ELA). Compute individually selected features from sampled points and objective values, with shared intermediates, explicit costs, and versioned definitions. A NumPy/SciPy core is complemented by an optional differentiable PyTorch backend.

Installation

Python 3.10 or newer is required. Install from PyPI:

pip install orivex

Add the optional differentiable PyTorch backend with pip install "orivex[torch]".

To work from a checkout instead:

git clone https://github.com/helix-agh/orivex.git
cd orivex
uv sync

With pip, run python -m pip install -e . (add ".[torch]" for the PyTorch backend).

Quick start

import numpy as np

from orivex import LandscapeSample, compute

rng = np.random.default_rng(42)
x = rng.uniform(-5.0, 5.0, size=(200, 2))
y = np.sum(x**2, axis=1)
sample = LandscapeSample(x, y, lower=[-5.0, -5.0], upper=[5.0, 5.0])

result = compute(sample, ["ela_distr.skewness", "nbc.nn_nb.mean_ratio"])
for name, item in result.values.items():
    print(name, item.value, item.status.value)

Objective values are min-max normalized by default. Pass y_normalization=None to use raw canonical objectives. Undefined features return an invalid status with an explanation.

Documentation

Read the documentation for installation, feature selection, normalization, fitness-distance conventions, PyTorch support, and API reference. The feature overview, contributing guide, and benchmark guide are also available.

Preview the documentation locally:

uv sync --extra docs
uv run --no-sync mkdocs serve

See documentation development for strict builds and deployment.

License

orivex is licensed under the MIT License.