beetroots 0.2.0

Beetroots (BayEsian infErence with spaTial Regularization of nOisy multi-line ObservaTion mapS)

beetroots

Beetroots (BayEsian infErence with spaTial Regularization of nOisy multi-line ObservaTion mapS) is a Python package that performs Bayesian inference of physical parameters from multispectral-structured cubes with a dedicated sampling algorithm. Thanks to this sampling algorithm, beetroots provides maps of credibility intervals along with estimated maps.

The sampling algorithm is introduced in

[1] P. Palud, P.-A. Thouvenin, P. Chainais, E. Bron, and F. Le Petit - Efficient sampling of non log-concave posterior distributions with mixture of noises, IEEE Transactions on Signal Processing, vol. 71, pp. 2491 -- 2501, 2023. doi:10.1109/TSP.2023.3289728

Such inversions rely on a forward model that is assumed to emulate accurately the physics of the observed environment. In parallel of the inversion, beetroots tests this hypothesis to evaluate the validity of the inference results. The testing method is described in (in French)

[2] P. Palud, P. Chainais, F. Le Petit, P.-A. Thouvenin and E. Bron - Problèmes inverses et test bayésien d'adéquation du modèle, GRETSI - Groupe de Recherche en Traitement du Signal et des Images in 29e Colloque sur le traitement du signal et des images, Grenoble, pp. 705 -- 708, 2023.

This package was applied e.g., to infer physical conditions in different regions of the interstellar medium in

[3] P. Palud, P.-A. Thouvenin, P. Chainais, E. Bron, F. Le Petit and ORION-B consortium - Bayesian inversion of large interstellar medium observation maps, in prep

It was also exploited to assert and compare the relevance of tracers and combination of tracers to constrain physical conditions in

[4] L. Einig, P. Palud, A. Roueff, P.-A. Thouvenin, P. Chainais, E. Bron, F. Le Petit, J. Pety, J. Chanussot and ORION-B consortium - Entropy-based selection of most informative observables for inference from interstellar medium observations, in prep

Complex forward models

In numerous real-life applications, the forward model is a complex numerical simulation. In the astrophysics applications presented in the documentation, the numerical simulation is replaced with a neural network-based approximation of the forward model for

• faster evaluation
• ability to evaluate derivatives

The package used to derive this approximation is nnbma (Neural Network-Based Model Approximation). The GitHub repository can be found here, the package here and the corresponding documentation here. The paper presenting this package is

[5] P. Palud, L. Einig, F. Le Petit, E. Bron, P. Chainais, J. Chanussot, J. Pety, P.-A. Thouvenin and ORION-B consortium - Neural network-based emulation of interstellar medium models, Astronomy & Astrophysics, 2023, 678, pp.A198. doi:10.1051/0004-6361/202347074

Installation and testing

To prepare and perform an inversion, we recommend installing the package. The package can be installed with pip:

pip install beetroots

or by cloning the repo. To clone, install and test the package, run:

git clone [e-mail]:pierrePalud/beetroots.git
cd beetroots
poetry install # or poetry install -E notebook -E docs for extra dependencies
poetry shell
pytest