CLAWDIA is an open-source Python framework for applying sparse dictionary learning (SDL) to gravitational-wave (GW) data analysis.
The framework systematises previously isolated SDL workflows into a unified, modular environment with a consistent, NumPy-style API. The current release focuses on time-domain denoising and classification under realistic detector noise. Denoising is implemented through several LASSO-regularised sparse reconstruction strategies (including simple sliding-window, margin-constrained, iterative-residual, and reference-guided methods) built on top of the SPAMS LASSO solver. Classification is provided by a dedicated dictionary model based on Low-Rank Shared Dictionary Learning (LRSDL), tailored to fixed-length GW signals. All dictionaries are exposed as Python classes that handle training, reconstruction, and prediction, and can be used independently or combined in custom workflows.
A lightweight classification pipeline is included as a reference implementation. It chains together preprocessing (via GWADAMA), sparse denoising, and an LRSDL classifier to perform supervised classification in low signal-to-noise ratio conditions. The pipeline is not the architectural centre of the framework but a convenient example of how individual components can be assembled into a reproducible SDL-based workflow.
Beyond these specific methods, CLAWDIA is intended as a general-purpose, community-driven library for sparse modelling in GW data analysis. Its design targets scarce-data regimes, class imbalance, and interpretability, aiming to provide robust, physically meaningful representations of GW morphology. The framework is designed to remain extensible: future releases are planned to include additional SDL-based classifiers, patch-based models for variable-length signals, frequency- or band-targeted dictionaries, adaptive and multi-detector setups, and support for further tasks such as detection, parameter estimation, regression, and controlled data generation. Optimisation tools, curriculum-like training schemes, and more efficient back-ends (including compiled extensions) are also foreseen.
Users of clawdia are kindly requested to cite the corresponding framework paper when using the software in academic work:
M. Llorens-Monteagudo, A. Torres-Forné, and J. A. Font, 2025,
"CLAWDIA: A dictionary learning framework for gravitational-wave data analysis",
arXiv:2511.16750 [astro-ph.IM], https://arxiv.org/abs/2511.16750.
This paper should be taken as the primary reference for clawdia, and provides further details together with illustrative applications to real and simulated GW data.
CLAWDIA was developed as part of the PhD thesis:
M. Llorens-Monteagudo, 2025,
"Gravitational-wave signal denoising, reconstruction and classification via Sparse Dictionary Learning",
PhD thesis, Universitat de València, Spain.
Publicly available at https://hdl.handle.net/10550/110046