Background Generation

[wbenoit26]

• Open data is fetched from GWOSC using mldatafind's wrapper around the GWpy fetch_open_data function.
• Data is resampled to 2048 Hz
• Strain data used for training comes from between GPS times 1240579783 and 1241443783, a 10 day period in May 2019, as specified in the pyproject.toml
• Strain data used for testing comes from between GPS times 1241443783 and 1244035783, a 30 day period directly after the end of the training period.
• Only times at which both specified interferometers are science-ready. Note that coincidence is not actually required for the training data, as background instances are sampled independently. However, we use coincident segments for both training and testing.
• Segments are split such that they have a maximum length of 20,000 seconds. The minimum segment length is 8000 seconds for training segments and 1024 seconds for testing segments. We require a longer minimum length for the training segments because one of the training segments is used to calculate the SNR of our testing waveforms, see Testing Dataset #474.
• For an example of the resulting output, see /home/ethan.marx/aframe/methods-paper-data/train/background/ and /home/ethan.marx/aframe/methods-paper-data/test/background/ on LHO