Closed wbenoit26 closed 2 weeks ago
Great stuff looks good. Couple Qs
How big is the file? Might make sense to add a function that downloads the file, or even just point people to the zenodo link
Would it be possible down the line to massage the data into a format that can be ingested by our own SV code (as opposed to Tris)? More of a sanity checking thing.
The file seems to be bigger on github for some reason: du -sh
says it's 36 MB, but github says it's 78 MB. I like the idea of downloading it, I'll add that in.
In theory, that should be possible. I tried it a while back and got turned around somewhere, but it's worth trying that again.
Right, I remember now: the dataset doesn't have the parameters from the rejected draws. They normalize by the total number of draw, rather than by the sum of the weights of the accepted and rejected draws, which I'm fairly certain we decided is incorrect, though I'd need to go back and look at our work. To be fair, for our data, there's a less than 0.2% difference between the values.
So we can't quite get the data into the format we'd need, but we can get close enough to do it in a notebook.
The plotting code will now calculate the SV vs FAR for cWB, gstLAL, MBTA, PyCBC-BBH, and PyCBC-Broad and plots those values alongside the values for Aframe. This PR also switches the default units of FAR on the plots to 1/years and reduces the default maximum FAR to 1/day.
The calculation is able to calculate SV vs p_astro just as well, so we can integrate those plots if we want to once our own calculation is ready.
I don't see a way around including the large data file we need for making these calculations. I've stripped out any columns that we don't use, and it's still pretty big.