Patrick Godwin authored 6 years ago and Kipp Cannon committed 6 years ago

fxtools/utils.py: change doctest for TimeseriesFeatureQueue since doctest does not play nice with defaultdict

Patrick Godwin authored 6 years ago and Kipp Cannon committed 6 years ago

Makefile.offline_tutorial_test: changed the DATASTART, DATASTOP times for fake frame production so frames cover the entire span of the analysis

Patrick Godwin authored 6 years ago and Kipp Cannon committed 6 years ago

fxtools/multirate_datasource.py: add audioconvert element after interpolator for negotiation to work properly, switch back from tdwhiten to firbank with whitening kernel for performance while that is sorted out

Patrick Godwin authored 6 years ago and Kipp Cannon committed 6 years ago

feature_extractor.py + utils.py: remove unused columns in generating triggers and made some performance tweaks

Patrick Godwin authored 6 years ago and Kipp Cannon committed 6 years ago

gstlal_sngltrigger: factor out code that adds single peaks to buffers to avoid looping through channels to select max snr channel when using this option

Patrick Godwin authored 6 years ago and Kipp Cannon committed 6 years ago

gstlal_peakfinder.c: save result of abs value of SNR in gstlal_peak_max_over_channels() to avoid repeated evaluation

Chad Hanna authored 6 years ago and Kipp Cannon committed 6 years ago

gstlal_interpolator.c: fix caps negotiation bug, add debugging statements and some general cleanup of warnings

- redefine the ranking statistic to be the largest ln L for any subset of the triggers in a candidate to prevent a glitch in one detector from making a good candidate look bad

- instead of a dictionary of SNRs and keyword arguments of (SNR,
  \chi^2/SNR^2) pairs, switch to a dictionary of SNRs and a dictionary of
  \chi^2/SNR^2
- this is the first step toward re-defining the ranking stat as the ln L
  maximized over subsets of triggers.  this re-organization will simplify
  the iterations require to implement that feature

- in .ln_lr_from_triggers(), select reference event as first in alphabetical order by instrument.  avoids cost of GPS time reconstruction to choose reference.

- rename a keyword argument of RankingStatPDF() to more accurately reflect its purpose

Patrick Godwin authored 6 years ago and Kipp Cannon committed 6 years ago

fxtools/multirate_datasource.py: replace resample with interpolator elem for downsampling, update firbank elem for whitening with tdwhiten elem, update comments, add check for GstAudio 1.0

- left-over bug from rewrite

- move Handler to its own module
- rename .gen_psd_xmldoc() to .__get_psd_xmldoc() following conventions used in inspiral.Data
- inspiral.py: excise gracedb code from Data object, encapsulate in GracedBWrapper class
- remove now unused write_psd_fileobj() from reference_psd.py
- move Data class to lloidhandler.py
- rename ._gatehandler() .__gatehandler() for consistency
- remove .flush(), unused locking wrapper of .__flush()
- remove .do_gracedb_alerts(), unused locking wrapper of .__do_gracedb_alerts()
- excise eye candy from Data class, move to stand-alone EyeCandy class
- merge what remains of Data and Handler into a single class
- excise segment tracking from result, move to stand-alone SegmentsTracker class
- use Handler's PSDs for GracedB uploads instead of getting new ones
  (addresses a FIXME)

gstlal-calibration:  new element lal_dqtukey that reads in a bit vector and outputs a stream that makes smooth transitions between zero and one based on the bit vector.

snglbursttable.py: fix import to _snglbursttable to not be relative, which causes ValueError: Attempted relative import in non-package

feature_extraction: add fake live data source (white_live), add special case for default gps start time since fake live sources do not seem to be seekable or able to change gps time offsets, add seeking for fake sources. fxtools/utils.py: fix issue with feature saving directory structure when gps times are less than 10 digits

- If EOS is reached before the required number of FFTs are computed to make transfer functions, a transfer function will be written with the available data
- Added option to take a median of transfer functions instead of averaging them with a mean
- Added option to store the first transfer function computed after a gap, so that it can be used after the next gap.