What does this task do?

This task generates OmegaScans of h(t) and processes them with Gravity-Spy’s Convolutional Neural Network (CNN) model to generate glitch categorization confidences. Omegascan PNGs and tabular results are included in the report. We note that standard CNN models are included within the DQR source code (under ~dqr/utils/) and are referenced automatically by some derived tasks.

What are its return states?

  • human_input_needed

  • error

How was it reviewed?

This has not been reviewed!

How should results be interpreted?

Evaluate the output of gravityspy by assessing the omegascans in conjunction with label and confidence level assigned. Any confidence > 0.1 assigned to the event will be highlighted. We expect strong signals without overlapping glitches will likely be assigned gravityspy labels no_glitch or chirp. Thee labels will be be highlighted in green. If the event has a confidence > 0.1 assigned to another label (usually associated with a glitch) the label will be highlighted in red.

There is no automatic pass or fail for this task. Sometimes an event can be assigned multiple event-like and glitch-like labels. In this scenario use your judgment to assess why gravityspy might assign those labels. Compare the h(t) omegascan with known classes of glitches, particularly those known to affect CBC templates [1, 2, 3]. Make note of any mis-classifications in your report. Jointly with search and PE experts, evaluate the impact of any identified glitches on the reported trigger and recommend appropriate mitigation (i.e. gating, linear subtraction with a witness, glitch model subtraction, veto/retraction, etc.).

Gravityspy’s classification is very time sensitive. In the plot below you will see how gravityspy’s classification jumps between different labels around an event time on the 1ms timescale. Red indicates the greater liklihood of a particular label. You should bear this in mind when you are interpreting gravityspy results.

alternate text

What INI options, config files are required?

  • path_to_cnn_model (string)

    • the path to the CNN model used by Gravity-Spy. This is specified automatically for derived tasks and they will ignore this option if it is present.

    • reference Gravity-Spy documentation for information on the file format required for the CNN models.

Are there any derived tasks that are based on this one?

The following reference CNN models stored within the DQR source code and therefore will ignore path_to_cnn_model if it is supplied.

  • H1 gravityspy

  • L1 gravityspy

  • V1 gravityspy