coincidence_HM.veto_and_optimize_single_detector#
Back to Module page
Summary#
Veto and optimize coincident candidates in a single detector
Signature#
def veto_and_optimize_single_detector(triggers, trig_obj, time_shift_tol, group_duration = 0.1, veto_triggers = True, min_veto_chi2 = None, apply_threshold = True, origin = 0, n_cores = 1, opt_format = 'new')
Name |
Type |
Default |
Description |
|---|---|---|---|
|
Processedclist with candidates in a detector |
||
|
Instance of trig.TriggerList with processed data in the detector |
||
|
Veto only one trigger every bucket of time_shift_tol seconds |
||
|
0.1 |
Divide triggers into groups every group_duration seconds and define subset for this group at once, saves on FFTs |
|
|
True |
Flag indicating whether to veto triggers |
|
|
None |
If given, veto only the candidates above this bar |
|
|
True |
Flag to apply threshold on single-detector chi2 when optimizing |
|
|
0 |
Origin to split the trigger times relative to |
|
|
1 |
Number of cores to use for splitting the computation |
|
|
‘new’ |
How we choose the finer grid, changed between O1 and O2 analyses Exposed here to replicate old runs if needed |
Output variables#
Return annotation |
Docstring type |
Description |
|---|---|---|
|
Dictionary indexed by bucket_ids with passed/failed veto, a processedclist of friends, a snr^2 correction factor, and metadata about data quality and glitch tests (see veto_and_optimize_group for an explanation) |
Docstring#
Veto and optimize coincident candidates in a single detector
:param triggers: Processedclist with candidates in a detector
:param trig_obj:
Instance of trig.TriggerList with processed data in the detector
:param time_shift_tol:
Veto only one trigger every bucket of time_shift_tol seconds
:param group_duration:
Divide triggers into groups every group_duration seconds and
define subset for this group at once, saves on FFTs
:param veto_triggers: Flag indicating whether to veto triggers
:param min_veto_chi2: If given, veto only the candidates above this bar
:param apply_threshold:
Flag to apply threshold on single-detector chi2 when optimizing
:param origin: Origin to split the trigger times relative to
:param n_cores: Number of cores to use for splitting the computation
:param opt_format:
How we choose the finer grid, changed between O1 and O2 analyses
Exposed here to replicate old runs if needed
:return:
Dictionary indexed by bucket_ids with passed/failed veto,
a processedclist of friends, a snr^2 correction factor, and metadata
about data quality and glitch tests
(see veto_and_optimize_group for an explanation)