ml4gw-buoy 0.3.0

Deploy Aframe and AMPLFI models over open data

Installation

This library is pip installable with

pip install ml4gw-buoy

It is recommended that you install buoy in a virtual environment such as conda.

Usage

The function of this library is to run trained Aframe and AMPLFI models over a gravitaional wave event reported by the LIGO-Virgo-KAGRA collaboration during their third observing run, O3.

Note: the trained models will be downloaded from HuggingFace and require about 320 MB of space in total.

To produce model outputs, first identify an event of interest. This can either be a catalog event, e.g., from GWTC-3, formatted like GW190521, or it can be a G event or superevent from GraceDB, formatted like G363842 or S200213t. Note that LIGO credentials are required to use the latter option, and also that this library uses public data, meaning that events from O4 cannot be analyzed.

Once an event has been identified, run:

buoy --events <EVENT_NAME> --outdir <OUTPUT_DIRECTORY>

The output directory is structured as follows will contain a directory matching the name of the event. Inside, there will be a data directory containing data created during the analysis, and a plots directory containing Aframe's response to the event as well as a skymap and corner plot from AMPLFI.

Multiple events can be specified at once, e.g.:

buoy --events '["GW190828_063405", "GW190521", "S200213t"]' --outdir <OUTPUT_DIRECTORY>

About 10 MB of space is required for each event.