search

dingo-gw / dingo

Dingo: Deep inference for gravitational-wave observations
MIT License
55 stars 18 forks source link

Implementation of Injections into Dingo Pipe #198

Open nihargupte-ph opened 1 year ago

nihargupte-ph commented 1 year ago

It would be nice to have dingo pipe take injections. The behaviour would be to specify an injection-dict column in dingo_pipe which would then call the dingo Injection module to generate an event dataset. We can also think about implementing a zero-noise feature which would run multiple iterations of the pipeline and average the results at the end, though this could maybe be a separate PR.

mpuerrer commented 1 year ago

That would be very useful and I'd be happy to help implement this!

stephengreen commented 1 year ago

This could be implemented in dingo_pipe_generation, replacing the Bilby injection code with one that uses the Dingo waveform generator exactly consistent with the network. It's slightly unclear how / if to allow for changes in data conditioning settings during importance sampling, which we currently allow for using provided data. If we simply regenerate the data with the new settings then we will get a different noise realization; we could try to downsample the EventDataset, although this is not currently implemented and would change the structure of the code. Maybe it's best to not allow changes in data conditioning for injections.