Skip to content

Commit

Permalink
Dev comments for the pycbc_multi_inspiral executable (gwastro#4513)
Browse files Browse the repository at this point in the history 
* Dev comments for the pycbc_multi_inspiral executable

* minor changes to dev comments pycbc_multi_inspiral

* minor change in dev comments pycbc_multi_inspiral

* Update dev comments pycbc_multi_inspiral

* Update pycbc_multi_inspiral

* Update pycbc_multi_inspiral

* Update dev comments for pycbc_multi_inspiral

* Update on dev comments for pycbc_multi_inspiral

Simplified the comments in the argparse section of the code. And moved some of the important details to gwastro#4513 (comment) .

This has been done in order to keep track of them and migrate them to  issues and PRs in the near future

* Update to dev comments pycbc_multi_inspiral

* Update dev comments pycbc_multi_inspiral

* Update dev comments pycbc_multi_inspiral

* Update dev comments pycbc_multi_inspiral

* Update dev comments pycbc_multi_inspiral
  • Loading branch information
@sebastiangomezlopez @acorreia61201
sebastiangomezlopez authored and acorreia61201 committed Apr 4, 2024
1 parent 9645f34 commit d248865
Showing 1 changed file with 84 additions and 31 deletions.
115 changes: 84 additions & 31 deletions bin/pycbc_multi_inspiral
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,10 @@
"""
Find multi-detector gravitational wave triggers and calculate the
coherent SNRs and related statistics.
To see an example on how to analyze the GW170817 event using this executable,
take a look at:
https://github.com/gwastro/pycbc/blob/master/examples/multi_inspiral/run.sh
"""

import logging
Expand All @@ -36,10 +40,14 @@ from pycbc.filter import MatchedFilterControl
from pycbc.types import TimeSeries, zeros, float32, complex64
from pycbc.types import MultiDetOptionAction
from pycbc.vetoes import sgchisq


# The following block of lines sets up the command-line interface (CLI) for the
# pycbc_multi_inspiral executable.
time_init = time.time()
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("-V", "--verbose", action="store_true",
help="print extra debugging information",
help="prints extra debugging information during runtime",
default=False)
parser.add_argument("--output", type=str)
parser.add_argument("--instruments", nargs="+", type=str, required=True,
Expand Down Expand Up @@ -74,7 +82,7 @@ parser.add_argument("--bank-veto-bank-file", type=str, help="Path to the "
"bank file used to compute the the bank chi-square veto.")
parser.add_argument("--chisq-bins", default=0)
# Commenting out options which are not yet implemented
# parser.add_argument("--chisq-threshold", type=float, default=0)
# parser.add_argument("--chisq-threshold", type=float, default=0)
# parser.add_argument("--chisq-delta", type=float, default=0)
parser.add_argument("--autochi-number-points", type=int, default=0)
parser.add_argument("--autochi-stride", type=int, default=0)
Expand Down Expand Up @@ -120,7 +128,7 @@ parser.add_argument("--trigger-time", type=int,
help="Time of the GRB, used to set the antenna patterns.")
parser.add_argument("--projection", default="standard",
choices=["standard", "left", "right", "left+right"],
help="Choice of projection matrix. 'left' and 'right' "
help="Choice of projection matrix. 'left' and 'right' "
"correspond to face-away and face-on")
parser.add_argument("--num-slides", type=int, default=0,
help="Number of time slides to perform.")
Expand All @@ -139,19 +147,38 @@ args = parser.parse_args()
init_logging(args.verbose)
# Set GRB time variable for convenience
t_gps = args.trigger_time
# Put the ifos in alphabetical order so they are always called in
# the same order.
# The following line aranges the ifos in alphabetical order so they are
# always called in the same order.
args.instruments.sort()
strain.verify_strain_options_multi_ifo(args, parser, args.instruments)
nifo = len(args.instruments[:])
# The following class verification methods are used to check whether input CLI
# options provided by parser to pycbc.strain, pycbc.strain.StrainSegments,
# pycbc.psd, pycbc.scheme & pycbc.fft modules are sane.
# pycbc.strain CLI verifier:
strain.verify_strain_options_multi_ifo(
args, parser, args.instruments)
strain.StrainSegments.verify_segment_options_multi_ifo(
args, parser, args.instruments)
psd.verify_psd_options_multi_ifo(args, parser, args.instruments)
args, parser, args.instruments)
# The pycbc.psd CLI verifier:
psd.verify_psd_options_multi_ifo(
args, parser, args.instruments)
# The pycbc.scheme CLI verifier:
scheme.verify_processing_options(args, parser)
# The pycbc.fft CLI verifier:
fft.verify_fft_options(args, parser)
# The following stores a Dictionary of "InjFilterRejector" objects
# which is used to avoid using computing power on injections corresponding
# to templates with significantly different chirp mass.
inj_filter_rejector = inject.InjFilterRejector.from_cli_multi_ifos(
args, args.instruments)
# The following stores a dictionary of "timeseries" objects. Each of them
# corresponds to the strain each detectors.
# strain data is taken from args.gps_start_time to args.gps_end_time
# using the sampling rate args.sample_rate
strain_dict = strain.from_cli_multi_ifos(
args, args.instruments, inj_filter_rejector, dyn_range_fac=DYN_RANGE_FAC)
# The following stores a dictionary of python slice objects, that indicate
# where the segments begin and end,for each detector's timeseries.
strain_segments_dict = strain.StrainSegments.from_cli_multi_ifos(
args, strain_dict, args.instruments)
ctx = scheme.from_cli(args)
Expand All @@ -160,6 +187,9 @@ with ctx:
# Set some often used variables for easy access
flow = args.low_frequency_cutoff
flow_dict = defaultdict(lambda : flow)
# The following for loop is used to check whether
# the sampling rate, flen and tlen agree for all detectors
# taking the zeroth detector in the list as a reference.
for count, ifo in enumerate(args.instruments):
if count == 0:
sample_rate = strain_dict[ifo].sample_rate
Expand All @@ -180,12 +210,17 @@ with ctx:
err_msg = "Sample rate, frequency length and time length "
err_msg += "must all be consistent across ifos."
raise ValueError(err_msg)
# segments is a dictionary of "frequency domain" objects, each
# of which is the Fourier transform of the segments in strain_segments_dict
logging.info("Making frequency-domain data segments")
segments = {
ifo: strain_segments_dict[ifo].fourier_segments()
for ifo in args.instruments
}
# The timedomain segments dictionary is deleted to save memory.
del strain_segments_dict
# The follwowing pycbc.psd method
# associates PSDs to segments for all ifos when using the multi-detector CLI
psd.associate_psds_to_multi_ifo_segments(
args, segments, strain_dict, flen, delta_f, flow, args.instruments,
dyn_range_factor=DYN_RANGE_FAC, precision='single')
Expand All @@ -194,28 +229,32 @@ with ctx:
# change if needed. Segments is only used to get tlen etc. which is
# same for all ifos, so just send the first ifo
template_mem = zeros(tlen, dtype=complex64)

#Read the sky grid or the single sky position
if args.sky_grid is not None and args.ra is not None and args.dec is not None:
parser.error('Give either a sky grid or a sky position, not both')

if args.sky_grid is not None:
sky_grid = h5py.File(args.sky_grid, 'r')
ra = np.array(sky_grid['ra'])
dec = np.array(sky_grid['dec'])
if args.ra is not None and args.dec is not None:
ra = np.array([args.ra])
dec = np.array([args.dec])

sky_positions = np.array([ra, dec])
num_sky_positions = sky_positions.shape[1]
positions_array = np.arange(num_sky_positions)

# Calculate time delays to each detector for each sky position and apply time slide shifts
# Calculate time delays to each detector and store time slide shifts
# in a dedicated dictionary
slide_ids = np.arange(1 + args.num_slides)
time_slides = {
ifo: args.slide_shift * slide_ids * n_ifo
for n_ifo, ifo in enumerate(args.instruments)}
# Given the time delays in time_slides, the following dictionary keeps the
# time delay indices evaluated wrt the geocenter, in units of samples, i.e.
# (time delay from geocenter + timeslide)*sampling_rate
time_delay_idx = {
slide: {
position_index: {
Expand All @@ -229,10 +268,10 @@ with ctx:
} for position_index in positions_array
} for slide in slide_ids
}

# Matched filter each ifo. Don't cluster here for a coherent search.
# Clustering happens at the end of the template loop.
# FIXME: The single detector SNR threshold should not necessarily be
# NOTE: The single detector SNR threshold should not necessarily be
# applied to every IFO (usually only 2 most sensitive in
# network)
matched_filter = {
Expand All @@ -243,6 +282,9 @@ with ctx:
upsample_threshold=args.upsample_threshold,
upsample_method=args.upsample_method, cluster_function='symmetric')
for ifo in args.instruments}
# Vetoes
# The existing SingleDetPowerChisq can calculate the single detector
# chisq for multiple ifos, so just use that directly.
logging.info("Initializing signal-based vetoes.")
# The existing SingleDetPowerChisq can calculate the single detector
# chisq for multiple ifos, so just use that directly.
Expand All @@ -256,10 +298,13 @@ with ctx:
autochisq = vetoes.SingleDetAutoChisq(
args.autochi_stride, args.autochi_number_points,
onesided=args.autochi_onesided)
# Here we take all frequency-domain segments and over-whiten them
# dividing by the PSD estimate
logging.info("Overwhitening frequency-domain data segments")
for ifo in args.instruments:
for seg in segments[ifo]:
seg /= seg.psd
# The following dicts are used to initialize and feed the event manager
ifo_out_types = {
'time_index': int,
'ifo': int, # IFO is stored as an int internally!
Expand Down Expand Up @@ -301,17 +346,18 @@ with ctx:
'coherent_snr': None,
'null_snr': None,
'nifo': None,
'my_network_chisq': None,
'my_network_chisq': None,
'reweighted_snr': None,
'slide_id': None
}
network_names = sorted(network_out_vals.keys())
event_mgr = EventManagerCoherent(
args, args.instruments, ifo_names,
[ifo_out_types[n] for n in ifo_names], network_names,
[network_out_types[n] for n in network_names],
[network_out_types[n] for n in network_names],
segments=segments, time_slides=time_slides)

# Template bank: filtering and thinning
logging.info("Read in template bank")
bank = waveform.FilterBank(
args.bank_file, flen, delta_f, complex64, low_frequency_cutoff=flow,
Expand All @@ -327,7 +373,7 @@ with ctx:
if not len(bank) == n_bank:
n_bank = len(bank)
logging.info("Template bank size after thinning: %d", n_bank)

# Antenna patterns
antenna_patterns = [[[0 for i in range(2)] for position_index in positions_array] for i in range(len(args.instruments))]
for i, ifo in enumerate(args.instruments):
Expand All @@ -344,7 +390,7 @@ with ctx:
for s_num,stilde in enumerate(segments[args.instruments[0]]):
stilde = {ifo : segments[ifo][s_num] for ifo in args.instruments}
# Filter check checks the 'inj_filter_rejector' options to
# determine whether to filter this template/segment
# determine whether to filter this template/segment
# if injections are present.
analyse_segment = True
for ifo in args.instruments:
Expand All @@ -361,45 +407,52 @@ with ctx:
ifo : template.sigmasq(segments[ifo][s_num].psd)
for ifo in args.instruments}
sigma = {ifo : np.sqrt(sigmasq[ifo]) for ifo in args.instruments}
# Every time s_num is zero or we skip the segment, we run new
# Every time s_num is zero or we skip the segment, we run new
# template to increment the template index
if s_num==0:
event_mgr.new_template(tmplt=template.params, sigmasq=sigmasq)
if not analyse_segment: continue
logging.info(
"Analyzing segment %d/%d", s_num + 1, len(segments[ifo]))
# The following dicts are created to store copies of the
# matched filtering results computed below
snr_dict = dict.fromkeys(args.instruments)
norm_dict = dict.fromkeys(args.instruments)
corr_dict = dict.fromkeys(args.instruments)
idx = dict.fromkeys(args.instruments)
snrv_dict = dict.fromkeys(args.instruments)
snr = dict.fromkeys(args.instruments)
# FIXME: 2 lines that can be moved outside the loops
# We do not really use ifo_list, and we reassign
# nifo identically a few lines below
ifo_list = args.instruments[:]
nifo = len(ifo_list)
for ifo in args.instruments:
logging.info(
"Filtering template %d/%d, ifo %s", t_num + 1, n_bank, ifo)
# No clustering in the coherent search until the end.
# The correlation vector is the FFT of the snr (so inverse
# FFT it to get the snr).
# FFT it to get the snr).
# The following lines Unpack & store copies of the matched
# filtering results for the current segment and template.
snr_ts, norm, corr, ind, snrv = \
matched_filter[ifo].matched_filter_and_cluster(
s_num, template.sigmasq(stilde[ifo].psd), window=0)
# snr time series for each ifo
snr_dict[ifo] = (
snr_ts[matched_filter[ifo].segments[s_num].analyze]
* norm)
norm_dict[ifo] = norm
# The correlation vector for each ifo. it is the FFT of the snr
# (so inverse FFT it to get the snr).
corr_dict[ifo] = corr.copy()
# Trigger indices list for each ifo
idx[ifo] = ind.copy()
# Trigger snr list for each ifo
snrv_dict[ifo] = snrv.copy()
# Normalized trigger snr list for each ifo
snr[ifo] = snrv * norm

# FIXME: wrong comment?

# Move onto next segment if there are no triggers.
if len(ifo_list)==0: continue
n_trigs = [len(snrv_dict[ifo]) for ifo in args.instruments]
if not any(n_trigs):
continue
# Short timeslides start here !
# Loop through slides, staring with the zero-lag
for slide in range(args.num_slides + 1):
logging.info(
Expand All @@ -411,8 +464,8 @@ with ctx:
# Even if we have none, need to keep an empty dictionary.
# Only do this if idx doesn't get time shifted out of the
# time we are looking at, i.e., require
# idx[ifo] - time_delay_idx[slide][position_index][ifo] to be in
# (0, len(snr_dict[ifo]))
# idx[ifo] - time_delay_idx[slide][position_index][ifo]
# to be in (0, len(snr_dict[ifo]))
idx_dict = {
ifo: idx[ifo][
np.logical_and(
Expand All @@ -422,7 +475,7 @@ with ctx:
]
for ifo in args.instruments
}

# Find triggers that are coincident (in geocent time) in
# multiple ifos. If a single ifo analysis then just use the
# indexes from that ifo.
Expand Down

0 comments on commit d248865

Please sign in to comment.