Glitch model does not recover prior when --noWaveletPrior is on for srate != 2048
When running with --noWaveletPrior
we do not recover the flat prior in glitch dimension. This issue becomes more noticeable when runs have higher dimensions and when runs have a sampling rate that is not 2048.
Note: These runs were completed without BayesLine, without pointing towards an actual PSD. This means that the PSDs used for these runs look like:
For N = 6
This issue is not immediately obvious. The grey bars are the 1, 2, and 3stdev markers assuming Poisson error in each bin.
expected count = N_samples / (Dmax + 1) # +1 since we can have 0 wavelets
stdev = np.sqrt(expected_count)
My run We see that there is definitely more of a slope with srate = 4096, but it does not look too concerning.
Run settings
`Dmax6_srate2048` Command line: --ifo L1 --L1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_L1.cache --L1-channel L1:GWOSC-4KHZ_R1_STRAIN --psdlength 4 --noClean --glitchOnly --Dmax 6 --Nchain 5 --Niter 4000000 --trigtime 1165578635.450000048 --segment-start 1165578633.450000048 --psdstart 1165578638.450000048 --srate 1024.0 --seglen 4 --window 1 --prior --noWaveletPrior
Dmax6_srate4096
Command line: --H1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_H1.cache --H1-channel H1:GWOSC-4KHZ_R1_STRAIN --ifo L1 --L1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_L1.cache --L1-channel L1:GWOSC-4KHZ_R1_STRAIN --psdlength 4 --noClean --glitchOnly --Dmax 6 --Nchain 5 --Niter 4000000 --trigtime 1165578635.450000048 --segment-start 1165578633.450000048 --psdstart 1165578638.450000048 --srate 4096 --seglen 4 --window 1 --prior --noWaveletPrior
For N = 40
I then attempted to see if it was perhaps an issue with the --clusterWeight, so I tried with both 0 and 1: Again, 2048 is the only sampling rate that remains within a reasonable range for the entirety. (Although 40_srate1024_noCluster
is almost close)
Run settings N = 40
40_srate4096_clusterweight0
Command line: --H1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_H1.cache --H1-channel H1:GWOSC-4KHZ_R1_STRAIN --ifo L1 --L1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_L1.cache --L1-channel L1:GWOSC-4KHZ_R1_STRAIN --psdlength 4 --noClean --glitchOnly --Dmax 40 --Nchain 5 --Niter 4000000 --trigtime 1165578635.450000048 --segment-start 1165578633.450000048 --psdstart 1165578638.450000048 --srate 4096 --seglen 4 --window 1 --prior --noWaveletPrior --clusterWeight 0
40_srate4096_clusterweight1
Command line: --H1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_H1.cache --H1-channel H1:GWOSC-4KHZ_R1_STRAIN --ifo L1 --L1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_L1.cache --L1-channel L1:GWOSC-4KHZ_R1_STRAIN --psdlength 4 --noClean --glitchOnly --Dmax 40 --Nchain 5 --Niter 4000000 --trigtime 1165578635.450000048 --segment-start 1165578633.450000048 --psdstart 1165578638.450000048 --srate 4096 --seglen 4 --window 1 --prior --noWaveletPrior --clusterWeight 1
40_srate2048_clusterweight1
Command line: --H1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_H1.cache --H1-channel H1:GWOSC-4KHZ_R1_STRAIN --ifo L1 --L1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_L1.cache --L1-channel L1:GWOSC-4KHZ_R1_STRAIN --psdlength 4 --noClean --glitchOnly --Dmax 40 --Nchain 5 --Niter 4000000 --trigtime 1165578635.450000048 --segment-start 1165578633.450000048 --psdstart 1165578638.450000048 --srate 2048 --seglen 4 --window 1 --prior --noWaveletPrior --clusterWeight 1
40_srate1024_nocluster
Command line: --H1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_H1.cache --H1-channel H1:GWOSC-4KHZ_R1_STRAIN --ifo L1 --L1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_L1.cache --L1-channel L1:GWOSC-4KHZ_R1_STRAIN --psdlength 4 --noClean --glitchOnly --Dmax 40 --noClusterProposal --Nchain 5 --Niter 4000000 --trigtime 1165578635.450000048 --segment-start 1165578633.450000048 --psdstart 1165578638.450000048 --srate 1024 --seglen 4 --window 1 --prior --noWaveletPrior
N = 40 on Simulated data
I then tried to see if perhaps the fact that I am using very ugly PSDs (autogenerated by BW) was perhaps affecting the runs. So I tried running with simulated data, and the results were quite nice: (No fisher no cluster is referencing that I went into the code and manually turned off these samplers, and is so not reflected in the run settings below)
Run Settings N = 40 Simulated
N = 40 with fixed PSDs
Command line: --ifo L1 --L1-cache LALSimAdLIGO --L1-channel LALSimAdLIGO --dataseed 1234 --psdlength 4 --noClean --glitchOnly --Dmax 40 --Nchain 5 --Niter 4000000 --trigtime 1165578635.450000048 --segment-start 1165578633.450000048 --psdstart 1165578638.450000048 --srate 4096 --seglen 4 --window 1 --prior --noWaveletPrior
40_srate_4096_simulate
Command line: --ifo L1 --L1-cache LALSimAdLIGO --L1-channel LALSimAdLIGO --dataseed 1234 --psdlength 4 --noClean --glitchOnly --Dmax 40 --Nchain 5 --Niter 4000000 --trigtime 1165578635.450000048 --segment-start 1165578633.450000048 --psdstart 1165578638.450000048 --srate 4096 --seglen 4 --window 1 --prior --noWaveletPrior
N = 40, fixed PSD
I then came to the conclusion that this nasty PSD was the issue. However, running on the same data as before except this time using a fixed PSD, I got these results:
The 4096 result obviously being troubling. This is as far as I have been able to get on this issue. I would point to runs but I have been doing them all locally for speed / convenience.
Run Settings N = 40, fixed PSD
40_srate_2048_withPSD
Command line: --ifo L1 --L1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_L1.cache --L1-channel L1:GWOSC-4KHZ_R1_STRAIN --L1-psd /Users/sophie/LIGO/DEBUG/glitch_model/psds/psd_2048.dat --srate 2048 --dataseed 1234 --psdlength 4 --noClean --glitchOnly --Dmax 40 --Nchain 5 --Niter 4000000 --trigtime 1165578635.450000048 --segment-start 1165578633.450000048 --psdstart 1165578638.450000048 --seglen 4 --window 1 --noWaveletPrior --prior
40_srate_4096_withPSD
Command line: --ifo L1 --L1-cache /Users/sophie/LIGO/glitch_events/Frames/Blip2_L1.cache --L1-channel L1:GWOSC-4KHZ_R1_STRAIN --L1-psd /Users/sophie/LIGO/DEBUG/glitch_model/psds/psd_4096.dat --srate 4096 --dataseed 1234 --psdlength 4 --noClean --glitchOnly --Dmax 40 --Nchain 5 --Niter 4000000 --trigtime 1165578635.450000048 --segment-start 1165578633.450000048 --psdstart 1165578638.450000048 --seglen 4 --window 1 --noWaveletPrior --prior