#!/usr/bin/env python
# Copyright (C) 2014 Alex Nitz
#
# This program is free software; you can redistribute it and/or modify it
# under the terms of the GNU General Public License as published by the
# Free Software Foundation; either version 3 of the License, or (at your
# option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
# Public License for more details.
#
# You should have received a copy of the GNU General Public License along
# with this program; if not, write to the Free Software Foundation, Inc.,
# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
import logging, argparse, sys, numpy
from pycbc import vetoes, psd, waveform, events, strain, scheme, fft
from pycbc.vetoes.sgchisq import SingleDetSGChisq
from pycbc import DYN_RANGE_FAC
from pycbc.filter import MatchedFilterSkyMaxControl, make_frequency_series, MatchedFilterSkyMaxControlNoPhase
from pycbc.types import TimeSeries, FrequencySeries, zeros, float32, complex64
import pycbc.fft.fftw
import pycbc.opt
parser = argparse.ArgumentParser(usage='',
description="Find single detector precessing gravitational-wave triggers.")
parser.add_argument("-V", "--verbose", action="store_true",
help="print extra debugging information", default=False )
parser.add_argument("--output", type=str, help="FIXME: ADD")
parser.add_argument("--bank-file", type=str, help="FIXME: ADD")
parser.add_argument("--snr-threshold",
help="SNR threshold for trigger generation", type=float)
parser.add_argument("--newsnr-threshold", type=float, metavar='THRESHOLD',
help="Cut triggers with NewSNR less than THRESHOLD")
parser.add_argument("--low-frequency-cutoff", type=float,
help="The low frequency cutoff to use for filtering (Hz)")
parser.add_argument("--enable-bank-start-frequency", action='store_true',
help="Read the starting frequency of template waveforms"
" from the template bank.")
parser.add_argument("--max-template-length", type=float,
help="The maximum length of a template is seconds. The "
"starting frequency of the template is modified to "
"ensure the proper length")
parser.add_argument("--enable-q-transform", action='store_true',
help="compute the q-transform for each segment of a "
"given analysis run. (default = False)")
# add approximant arg
waveform.bank.add_approximant_arg(parser)
parser.add_argument("--order", type=int,
help="The integer half-PN order at which to generate"
" the approximant. Default is -1 which indicates to use"
" approximant defined default.", default=-1,
choices = numpy.arange(-1, 9, 1))
taper_choices = ["start","end","startend"]
parser.add_argument("--taper-template", choices=taper_choices,
help="For time-domain approximants, taper the start and/or "
"end of the waveform before FFTing.")
parser.add_argument("--cluster-method", choices=["template", "window"],
help="FIXME: ADD")
parser.add_argument("--cluster-function", choices=["findchirp", "symmetric"],
help="How to cluster together triggers within a window. "
"'findchirp' uses a forward sliding window; 'symmetric' "
"will compare each window to the one before and after, keeping "
"only a local maximum.", default="findchirp")
parser.add_argument("--cluster-window", type=float, default = -1,
help="Length of clustering window in seconds."
" Set to 0 to disable clustering.")
parser.add_argument("--maximization-interval", type=float, default=0,
help="Maximize triggers over the template bank (ms)")
parser.add_argument("--bank-veto-bank-file", type=str, help="FIXME: ADD")
parser.add_argument("--chisq-snr-threshold", type=float,
help="Minimum SNR to calculate the power chisq")
parser.add_argument("--chisq-bins", default=0, help=
"Number of frequency bins to use for power chisq. Specify"
" an integer for a constant number of bins, or a function "
"of template attributes. Math functions are "
"allowed, ex. "
"'10./math.sqrt((params.mass1+params.mass2)/100.)'. "
"Non-integer values will be rounded down.")
parser.add_argument("--chisq-threshold", type=float, default=0,
help="FIXME: ADD")
parser.add_argument("--chisq-delta", type=float, default=0, help="FIXME: ADD")
parser.add_argument("--autochi-number-points", type=int, default=0,
help="The number of points to use, in both directions if"
"doing a two-sided auto-chisq, to calculate the"
"auto-chisq statistic.")
parser.add_argument("--autochi-stride", type=int, default=0,
help="The gap, in sample points, between the points at"
"which to calculate auto-chisq.")
parser.add_argument("--autochi-two-phase", action="store_true",
default=False,
help="If given auto-chisq will be calculated by testing "
"against both phases of the SNR time-series. "
"If not given, only the phase matching the trigger "
"will be used.")
parser.add_argument("--autochi-onesided", action='store', default=None,
choices=['left','right'],
help="Decide whether to calculate auto-chisq using"
"points on both sides of the trigger or only on one"
"side. If not given points on both sides will be"
"used. If given, with either 'left' or 'right',"
"only points on that side (right = forward in time,"
"left = back in time) will be used.")
parser.add_argument("--autochi-reverse-template", action="store_true",
default=False,
help="If given, time-reverse the template before"
"calculating the auto-chisq statistic. This will"
"come at additional computational cost as the SNR"
"time-series will need recomputing for the time-"
"reversed template.")
parser.add_argument("--autochi-max-valued", action="store_true",
default=False,
help="If given, store only the maximum value of the auto-"
"chisq over all points tested. A disadvantage of this "
"is that the mean value will not be known "
"analytically.")
parser.add_argument("--autochi-max-valued-dof", action="store", metavar="INT",
type=int,
help="If using --autochi-max-valued this value denotes "
"the pre-calculated mean value that will be stored "
"as the auto-chisq degrees-of-freedom value.")
parser.add_argument("--downsample-factor", type=int,
help="Factor that determines the interval between the "
"initial SNR sampling. If not set (or 1) no sparse sample "
"is created, and the standard full SNR is calculated.", default=1)
parser.add_argument("--upsample-threshold", type=float,
help="The fraction of the SNR threshold to check the sparse SNR sample.")
parser.add_argument("--upsample-method", choices=["pruned_fft"],
help="The method to find the SNR points between the sparse SNR sample.", default='pruned_fft')
parser.add_argument("--user-tag", type=str, metavar="TAG", help="""
This is used to identify FULL_DATA jobs for
compatibility with pipedown post-processing.
Option will be removed when no longer needed.""")
parser.add_argument("--keep-loudest-interval", type=float,
help="Window in seconds to maximize triggers over bank")
parser.add_argument("--keep-loudest-num", type=int,
help="Number of triggers to keep from each maximization interval")
parser.add_argument("--keep-loudest-stat", default="newsnr",
choices=events.ranking.sngls_ranking_function_dict.keys(),
help="Statistic used to determine loudest to keep")
parser.add_argument("--keep-loudest-log-chirp-window", type=float,
help="Keep loudest triggers within ln chirp mass window")
parser.add_argument("--finalize-events-template-rate", default=None,
type=int, metavar="NUM TEMPLATES",
help="After NUM TEMPLATES perform the various clustering "
"and rejection tests that would be performed at the "
"end of this job. Default is to only do those things "
"at the end of the job. This can help control memory "
"usage if a lot of triggers that would be rejected "
"are being retained. A suggested value for this is "
"500, but a good number may depend on other settings "
"and your specific use-case.")
parser.add_argument("--gpu-callback-method", default='none')
parser.add_argument("--sky-maximization-method", required=True,
choices=["precessing", "hom"])
# Add options groups
psd.insert_psd_option_group(parser)
strain.insert_strain_option_group(parser)
strain.StrainSegments.insert_segment_option_group(parser)
scheme.insert_processing_option_group(parser)
fft.insert_fft_option_group(parser)
pycbc.opt.insert_optimization_option_group(parser)
SingleDetSGChisq.insert_option_group(parser)
opt = parser.parse_args()
# Check that the values returned for the options make sense
psd.verify_psd_options(opt, parser)
strain.verify_strain_options(opt, parser)
strain.StrainSegments.verify_segment_options(opt, parser)
scheme.verify_processing_options(opt, parser)
fft.verify_fft_options(opt,parser)
pycbc.opt.verify_optimization_options(opt, parser)
pycbc.init_logging(opt.verbose)
ctx = scheme.from_cli(opt)
gwstrain = strain.from_cli(opt, DYN_RANGE_FAC)
strain_segments = strain.StrainSegments.from_cli(opt, gwstrain)
with ctx:
fft.from_cli(opt)
flow = opt.low_frequency_cutoff
flen = strain_segments.freq_len
tlen = strain_segments.time_len
delta_f = strain_segments.delta_f
logging.info("Making frequency-domain data segments")
segments = strain_segments.fourier_segments()
psd.associate_psds_to_segments(opt, segments, gwstrain, flen, delta_f,
flow, dyn_range_factor=DYN_RANGE_FAC, precision='single')
# storage for values and types to be passed to event manager
out_types = {
'time_index' : int,
'snr' : float32,
'chisq' : float32,
'chisq_dof' : int,
'bank_chisq' : float32,
'bank_chisq_dof' : int,
'cont_chisq' : float32,
'psd_var_val' : float32,
'u_vals' : float32,
'coa_phase' : float32,
'hplus_cross_corr' : float32
}
out_types.update(SingleDetSGChisq.returns)
out_vals = {key: None for key in out_types}
names = sorted(out_vals.keys())
if len(strain_segments.segment_slices) == 0:
logging.info("--filter-inj-only specified and no injections in analysis time")
event_mgr = events.EventManager(
opt, names, [out_types[n] for n in names], psd=None,
gating_info=gwstrain.gating_info)
event_mgr.finalize_template_events()
event_mgr.write_events(opt.output)
logging.info("Finished")
sys.exit(0)
# FIXME: Maybe we should use the PSD corresponding to each trigger
if opt.psdvar_segment is not None:
logging.info("Calculating PSD variation")
psd_var = pycbc.psd.calc_filt_psd_variation(gwstrain, opt.psdvar_segment,
opt.psdvar_short_segment, opt.psdvar_long_segment,
opt.psdvar_psd_duration, opt.psdvar_psd_stride,
opt.psd_estimation, opt.psdvar_low_freq, opt.psdvar_high_freq)
if opt.sky_maximization_method == 'hom':
matched_filter = MatchedFilterSkyMaxControlNoPhase(opt.low_frequency_cutoff,
None, opt.snr_threshold, tlen, delta_f,
complex64)
else:
matched_filter = MatchedFilterSkyMaxControl(opt.low_frequency_cutoff,
None, opt.snr_threshold, tlen, delta_f,
complex64)
bank_chisq = vetoes.SingleDetSkyMaxBankVeto(opt.bank_veto_bank_file,
flen, delta_f, flow, complex64,
phase_order=opt.order,
approximant=opt.approximant)
power_chisq = vetoes.SingleDetSkyMaxPowerChisq(num_bins=opt.chisq_bins,
snr_threshold=opt.chisq_snr_threshold)
autochisq = vetoes.SingleDetSkyMaxAutoChisq(opt.autochi_stride,
opt.autochi_number_points,
onesided=opt.autochi_onesided,
twophase=opt.autochi_two_phase,
reverse_template=opt.autochi_reverse_template,
take_maximum_value=opt.autochi_max_valued,
maximal_value_dof=opt.autochi_max_valued_dof)
logging.info("Overwhitening frequency-domain data segments")
for seg in segments:
seg /= seg.psd
event_mgr = events.EventManager(opt, names,
[out_types[n] for n in names], psd=segments[0].psd,
gating_info=gwstrain.gating_info)
logging.info("Read in template bank")
lfc = None if opt.enable_bank_start_frequency else flow
bank = waveform.FilterBankSkyMax(opt.bank_file, flen,
delta_f, complex64, phase_order=opt.order,
taper=opt.taper_template, approximant=opt.approximant,
out_plus=zeros(tlen, dtype=complex64),
out_cross=zeros(tlen, dtype=complex64),
max_template_length=opt.max_template_length,
low_frequency_cutoff=lfc)
sg_chisq = SingleDetSGChisq.from_cli(opt, bank, opt.chisq_bins)
for t_num, (hplus, hcross) in enumerate(bank):
# FIXME: Here we need to store sigmasq for plus *and* cross. Derived
# quantities like effective distance may be more complicated!
# These may depend on the [complex] overlap between plus and
# cross.
event_mgr.new_template(tmplt=hplus.params,
sigmasq_plus=hplus.sigmasq(segments[0].psd),
sigmasq_cross=hcross.sigmasq(segments[0].psd))
if opt.cluster_method == "window":
cluster_window = int(opt.cluster_window * gwstrain.sample_rate)
elif opt.cluster_method == "template":
cluster_window = int(template.chirp_length * gwstrain.sample_rate)
for s_num, stilde in enumerate(segments):
logging.info("Filtering template %d/%d segment %d/%d" % \
(t_num + 1, len(bank), s_num + 1, len(segments)))
# FIXME: Why this takes more inputs than for non-spin?
snr, corr_plus, corr_cross, idx, snrv, u_vals, coa_phase,\
hplus_cross_corr, hpnorm, hcnorm =\
matched_filter.matched_filter_and_cluster(hplus, hcross,
hplus.sigmasq(stilde.psd),
hcross.sigmasq(stilde.psd),
stilde.psd, stilde, cluster_window)
if not len(idx):
continue
out_vals['u_vals'] = u_vals
out_vals['coa_phase'] = coa_phase
out_vals['hplus_cross_corr'] = numpy.repeat(hplus_cross_corr,
len(u_vals))
# This hasn't been implemented yet, but the stub is still here.
out_vals['bank_chisq'], out_vals['bank_chisq_dof'] = \
bank_chisq.values()
out_vals['chisq'], out_vals['chisq_dof'] = power_chisq.values(\
corr_plus, corr_cross, snrv, stilde.psd,
idx+stilde.analyze.start, hplus, hcross, u_vals,
hplus_cross_corr, hpnorm, hcnorm)
out_vals['sg_chisq'] = sg_chisq.values(stilde, hplus, stilde.psd,
snrv, 1.,
out_vals['chisq'],
out_vals['chisq_dof'],
idx+stilde.analyze.start)
# This hasn't been implemented yet, but the stub is still here.
out_vals['cont_chisq'] = \
autochisq.values()
idx += stilde.cumulative_index
out_vals['time_index'] = idx
out_vals['snr'] = snrv
if opt.psdvar_short_segment is not None:
out_vals['psd_var_val'] = \
pycbc.psd.find_trigger_value(psd_var,
out_vals['time_index'],
opt.gps_start_time, opt.sample_rate)
event_mgr.add_template_events(names, [out_vals[n] for n in names])
event_mgr.cluster_template_events("time_index", "snr", cluster_window)
event_mgr.finalize_template_events()
if opt.finalize_events_template_rate is not None and \
not (t_num+1) % opt.finalize_events_template_rate:
event_mgr.consolidate_events(opt, gwstrain=gwstrain)
event_mgr.consolidate_events(opt, gwstrain=gwstrain)
event_mgr.finalize_events()
logging.info("Found %s triggers" % str(len(event_mgr.events)))
logging.info("Writing out triggers")
event_mgr.write_events(opt.output)
if opt.fftw_output_float_wisdom_file:
fft.fftw.export_single_wisdom_to_filename(opt.fftw_output_float_wisdom_file)
if opt.fftw_output_double_wisdom_file:
fft.fftw.export_double_wisdom_to_filename(opt.fftw_output_double_wisdom_file)
logging.info("Finished")