Welcome to the first machine learning gravitational-wave search mock data challenge hosted by the Albert-Einstein-Institut Hannover and the Friedrich-Schiller Universität Jena. In this challenge participants are tasked with finding gravitational-wave signals of varying complexity in a noisy background. Entries are evaluated on metrics which are used to classify the performance of real-world, state-of-the-art search algorithms.

The goal of this challenge is to create a collaborative publication that collects state-of-the-art machine learning based gravitational-wave search algorithms and enables a comparison to classical approaches such as matched filtering or coherent burst searches. Through this, we strive to highlight the advantages of different entries for specific tasks and want to pinpoint areas where further research seems fruitful.

Because this is a collaborative work, all teams that submit an algorithm and choose not to retract it before final publication will gain co-authorship. We nonetheless encourage publications on the individual algorithms to describe details of pre-processing, post-processing, training, etc. We, furthermore, encourage the publication of the source code used for training and evaluation to foster reproducability. However, open source code is not required for submission.

Although this challenge is focused on machine learning approaches, we do accept submissions which do not make use of this relatively new area of research.

If you want to partipate in this mock data challenge, please get in contact with us by sending a mail to [email protected]. We accept registrations up to a maximum number of 30 participating groups until December 31st, 2021 (We have remaining capacity. Please get in touch if you would still like to participate). The deadline for the final submission of the algorithm is March 31st, 2022.

On submission, we will evaluate your algorithm on a validation set. The performance on this validation set will then be reported back to you to check that the algorithm behaves as expected. Once we have confirmation by the group that the algorithm performs within the expected margins of error, we will evaluate the submission on a secret test set that is the same for all entries. The performance on this set will only be reported back to the groups on the first circulation of the publication draft. Submissions may be retracted at any point prior to final publication of the manuscript. For more information please refer to this page.

This repository contains source code to generate data of the kind that will be used for final evaluation as well as the source code that will be used to carry out this final evaluation. It also contains a few configuration files that are required for data generation.

Submissions must be able to process a file of HDF5 format that contains the raw strain data for 2 detectors. Any required pre-processing is expected to be performed by the submitted code. The output is expected to be another file of HDF5 format which contains times, ranking-statistic like values, and timing accuracies for candidate events. The ranking-statistic like values are numbers where a larger value is supposed to correspond to a larger probability of an astrophysical event to be present. For details on the input- and output-format please refer to the Wiki of this repository.

To run the code you need to have a working installation of Python 3.7 or higher. You will then need to install dependencies using

pip install -r requirements.txt

This installs a version of the PyCBC github that was tested and confirmed to be working. Older versions may be missing required functions.

For more detailed installation instructions please refer to this page.

If you make use of the code in this repository please cite it accordingly. Please cite as

@misc{https://doi.org/10.48550/arxiv.2209.11146,
    doi = {10.48550/ARXIV.2209.11146},
    url = {https://arxiv.org/abs/2209.11146},
    author = {Schäfer, Marlin B. and Zelenka, Ondřej and Nitz, Alexander H. and Wang, He and Wu, Shichao and Guo, Zong-Kuan and Cao, Zhoujian and Ren, Zhixiang and Nousi, Paraskevi and Stergioulas, Nikolaos and Iosif, Panagiotis and Koloniari, Alexandra E. and Tefas, Anastasios and Passalis, Nikolaos and Salemi, Francesco and Vedovato, Gabriele and Klimenko, Sergey and Mishra, Tanmaya and Brügmann, Bernd and Cuoco, Elena and Huerta, E. A. and Messenger, Chris and Ohme, Frank},
    keywords = {Instrumentation and Methods for Astrophysics (astro-ph.IM), High Energy Astrophysical Phenomena (astro-ph.HE), Machine Learning (cs.LG), General Relativity and Quantum Cosmology (gr-qc), FOS: Physical sciences, FOS: Physical sciences, FOS: Computer and information sciences, FOS: Computer and information sciences},
    title = {MLGWSC-1: The first Machine Learning Gravitational-Wave Search Mock Data Challenge},
    publisher = {arXiv},
    year = {2022},
    copyright = {arXiv.org perpetual, non-exclusive license}
}