Reimplement PhaseTDStatistic

[titodalcanton]

Reimplement PhaseTDStatistic based on the code that was in place before #3535, and modify the PyCBC Live test to use it. Some extra changes are needed to fix some errors related to a from_cli() method modifying some of the arguments and breaking further calls to the same function.

Closes #3580. Closes #3584.

I have not tested that the resulting values are actually correct or equivalent to the old code, though, so marking as WIP.

[spxiwh]

This looks good to me. As long as you/others are happy that the new code in stat.py reproduces what was done with the old 2-det coinc code then please merge.

[titodalcanton]

Hmm, interesting failure in PyCBC Live, apparently reproducible.

>> [Thu Jan 28 10:00:54 UTC 2021] Checking results
2021-01-28 10:00:55,585 No triggers found in V1
2021-01-28 10:00:55,585 Single Trigger Test Failed
2021-01-28 10:00:55,585 2 coincident trigger(s) detected
2021-01-28 10:00:55,586 Checking trigger output/coinc-1272790100.1669922-zimpiz.xml.gz
2021-01-28 10:00:55,668 IFO SNRs: [9.41339   7.7795067]
2021-01-28 10:00:55,669 Network SNR: 12.211988
2021-01-28 10:00:55,669 IFO End Time: 1272790100.000000
2021-01-28 10:00:55,669 Mass 1: 290.929321
2021-01-28 10:00:55,669 Mass 2: 3.675545
2021-01-28 10:00:55,669 Spin1z: 0.993485
2021-01-28 10:00:55,669 Spin2z: 0.927135
2021-01-28 10:00:55,669 Checking trigger output/coinc-1272790260.0966797-nrxeea.xml.gz
2021-01-28 10:00:55,749 IFO SNRs: [8.146549 6.880121]
2021-01-28 10:00:55,750 Network SNR: 10.663130
2021-01-28 10:00:55,750 IFO End Time: 1272790260.000000
2021-01-28 10:00:55,750 Mass 1: 1.133169
2021-01-28 10:00:55,750 Mass 2: 1.010624
2021-01-28 10:00:55,750 Spin1z: 0.029544
2021-01-28 10:00:55,750 Spin2z: 0.020994
2021-01-28 10:00:55,750 Test Failed
    FAILED!

???

[titodalcanton]

For reference, the failing test was due to #3603.

[titodalcanton]

Here is a very simple test using the current example. I compared the rankings of the recovered injections between this code version and 503873c (just before the stat refactor).

2020_05_06/H1L1V1-Live-1272790096.069336-8.hdf
  Difference in coa_phase_H1: -1.1920928955078125e-07
  Difference in coa_phase_V1: -2.384185791015625e-07
  Difference in coinc_rank: 0.40607452392578125
  Difference in coinc_rank_squared: 10.332295827902271
  Difference in end_time_H1: 0.0
  Difference in end_time_V1: 0.0
  Difference in ifos: set()
  Difference in snr_H1: 4.76837158203125e-06
  Difference in snr_V1: -7.152557373046875e-06
2020_05_06/H1L1V1-Live-1272790256.069336-8.hdf
  Difference in coa_phase_H1: -1.1920928955078125e-07
  Difference in coa_phase_V1: -3.5762786865234375e-07
  Difference in coinc_rank: 0.42845726013183594
  Difference in coinc_rank_squared: 10.54476247247294
  Difference in end_time_H1: 0.0
  Difference in end_time_V1: 0.0
  Difference in ifos: set()
  Difference in snr_H1: -3.814697265625e-06
  Difference in snr_V1: 9.5367431640625e-07

The difference in coinc ranks seems a bit high (both injections have coinc rank ~12). Part of this difference may be due to the necessary change in the format of the phase-time-amplitude files, however.

[titodalcanton]

I have now poked at the old and new phase-time-amplitude files for some time (see plots here) and I convinced myself that the ~100x difference in expected signal rate comes from the phase-time-amplitude files themselves and not from this code change. I might look into this further, but I would like to merge this so we can have a working O3-like configuration of PyCBC Live.

[tdent]

@titodalcanton have you had any feedback from @ahnitz on the change in PTA files and resulting change in statistic?

[titodalcanton]

@ahnitz and I discussed this and agreed that the rate is expected to have differences due to the PTA file format change.