PyGRB: Implement suitable ranking statistic from chisq tests

[a-r-williamson]

PyGRB (with pycbc_multi_inspiral) is in need of a power chi-sq implementation for the calculation of the detection statistic.

[spxiwh]

On this point we did have some discussion with @idorrington92 when he was writing pycbc_multi_inspiral. At the moment multi_inspiral does support the single detector chi-squareds:

https://github.com/gwastro/pycbc/blob/master/bin/pycbc_multi_inspiral#L390

but no form of coherent chi-squared (there's optimization questions here that will arise once short-slides/sky points are in ... and I think those sorts of questions are actually going to be the biggest hurdle in getting this code ready for production, but that's a different discussion).

The point is, that I think the consensus was (and I certainly believe) that these single-detector chi-squareds are more effective than the coherent chi-squared. Iain was looking at putting together a detection statistic using these (one basic idea is to somehow average the chi-squareds, proportionally to SNR^2 in each detector, and just weight the coherent SNR by that using the existing new SNR formula) ... I think that was in Iain's thesis.

[a-r-williamson]

Thanks for the information, @spxiwh. I think @tcarver1234 and co have been looking into this issue already, and I'm sure they've been doing as you say (I'm only just getting back in the loop on this). I'll leave it to them to comment further.

[idorrington92]

That was indeed in my thesis.

It's good to see work is still going on this. If there's anything a non-LIGO member might be able to help with in their spare time, then I'm happy to take a look. It would actually make me so happy to see this ready for production.

In fact, I might fix those doc strings on the functions to make them consistant with the rest of PyCBC this weekend. Then I think we can refactor the functions out to tidy things up a bit.

[JAMSADIQ]

Hi Everyone, I was using pycbc_multi_inspiral code with this version commit 6230eaa
When I printed chi-squared values of individual detectors from my runs these were negative values and that may be showing some issue in the code. Thanks

[tdent]

@a-r-williamson @pannarale - It seems this issue should be reassigned to someone who is able to work actively on it at the moment. As we have mentioned, Jam obtained some large negative values (more details can be provided if needed).