Multiple detectors, software injection (circular polarisation)#

Here we compare lalpulsar_parameter_estimation_nested with cwinpy in the case of simulated Gaussian noise from multiple detectors (H1, L1 and V1) containing a software injected signal with close-to circular polarisation. The parameters being estimated are \(h_0\), \(\phi_0\), \(\psi\) and \(\cos{\iota}\), all with uniform priors.

The script for this comparison, using the dynesty nested sampling algorithm, is shown at the bottom of the page. It produces the following comparison data:

../_images/multi_detector_software_injection_circular_corner.png

Evidence table#
Method	\(\ln{(Z)}\)	\(\ln{(Z)}\) noise	\(\ln{}\) Odds
`lalpulsar_parameter_estimation_nested`	488839.112	488320.478	518.634±0.109
`cwinpy_pe`	488302.601	487788.855	513.746±0.190
`cwinpy_pe` (grid)	488302.351		513.497

Parameter table#
Method	\(h_0\)	\(\phi_0\) (rad)	\(\psi\) (rad)	\(\cos{\iota}\)
`lalpulsar_parameter_estimation_nested`	1.27±0.16×10^-25	0.96±0.44	0.75±0.44	0.82±0.11
90% credible intervals	[1.06, 1.56]×10^-25	[0.23, 1.63]	[0.08, 1.48]	[0.64, 0.98]
`cwinpy_pe`	1.27±0.16×10^-25	0.97±0.44	0.74±0.44	0.81±0.11
90% credible intervals	[1.05, 1.55]×10^-25	[0.24, 1.62]	[0.09, 1.47]	[0.64, 0.98]

Maximum a-posteriori#
Method	\(h_0\)	\(\phi_0\) (rad)	\(\psi\) (rad)	\(\cos{\iota}\)	\(\ln{(L)}\) max
`lalpulsar_parameter_estimation_nested`	1.25×10^-25	1.11	0.60	0.83	488852.25
`cwinpy_pe`	1.22×10^-25	1.33	0.38	0.84	488315.67

Combined K-S test p-value: 0.5952

Maximum Jensen-Shannon divergence: 0.0009

CWInPy version: 1.0.0

bilby version: 2.1.1

#!/usr/bin/env python

"""
Compare cwinpy with lalpulsar_parameter_estimation_nested for noise-only
data for a multiple detectors (H1, L1 and V1) containing a software injection
with close-to-circular polarisation.
"""

import os
import subprocess as sp

import h5py
import matplotlib
import numpy as np
from astropy.utils.data import download_file
from bilby.core.prior import Uniform
from comparitors import comparisons
from lalinference import LALInferenceHDF5PosteriorSamplesDatasetName
from lalinference.io import read_samples
from matplotlib import pyplot as plt

from cwinpy import HeterodynedData
from cwinpy.pe import pe
from cwinpy.plot import Plot

matplotlib.use("Agg")

# URL for ephemeris files
DOWNLOAD_URL = "https://git.ligo.org/lscsoft/lalsuite/raw/master/lalpulsar/lib/{}"

# create a fake pulsar parameter file
parcontent = """\
PSRJ     J0123+3456
RAJ      01:23:45.6789
DECJ     34:56:54.321
F0       567.89
F1       -1.2e-12
PEPOCH   56789
H0       1.1e-25
COSIOTA  0.95
PSI      0.5
PHI0     1.2
"""

injection_parameters = {}
injection_parameters["h0"] = 1.1e-25
injection_parameters["phi0"] = 1.2
injection_parameters["psi"] = 0.5
injection_parameters["cosiota"] = 0.95

label = "multi_detector_software_injection_circular"
outdir = "outputs"

if not os.path.isdir(outdir):
    os.makedirs(outdir)

# add content to the par file
parfile = os.path.join(outdir, "{}.par".format(label))
with open(parfile, "w") as fp:
    fp.write(parcontent)

# create some fake heterodyned data
detectors = ["H1", "L1", "V1"]  # the detector to use
asd = 1e-24  # noise amplitude spectral density
times = np.linspace(1000000000.0, 1000086340.0, 1440)  # times
het = {}
hetfiles = []
for detector in detectors:
    het[detector] = HeterodynedData(
        times=times,
        par=parfile,
        injpar=parfile,
        inject=True,
        fakeasd=asd,
        detector=detector,
    )

    # output the data
    hetfile = os.path.join(outdir, "{}_{}_data.txt".format(label, detector))
    het[detector].write(hetfile)
    hetfiles.append(hetfile)

# create priors
phi0range = [0.0, np.pi]
psirange = [0.0, np.pi / 2.0]
cosiotarange = [-1.0, 1.0]
h0range = [0.0, 1e-23]

# set prior for lalpulsar_parameter_estimation_nested
priorfile = os.path.join(outdir, "{}_prior.txt".format(label))
priorcontent = """H0 uniform {} {}
PHI0 uniform {} {}
PSI uniform {} {}
COSIOTA uniform {} {}
"""
with open(priorfile, "w") as fp:
    fp.write(priorcontent.format(*(h0range + phi0range + psirange + cosiotarange)))

# set prior for bilby
priors = {}
priors["h0"] = Uniform(h0range[0], h0range[1], "h0", latex_label=r"$h_0$")
priors["phi0"] = Uniform(
    phi0range[0], phi0range[1], "phi0", latex_label=r"$\phi_0$", unit="rad"
)
priors["psi"] = Uniform(
    psirange[0], psirange[1], "psi", latex_label=r"$\psi$", unit="rad"
)
priors["cosiota"] = Uniform(
    cosiotarange[0], cosiotarange[1], "cosiota", latex_label=r"$\cos{\iota}$"
)

# run lalpulsar_parameter_estimation_nested
try:
    execpath = os.environ["CONDA_PREFIX"]
except KeyError:
    raise KeyError(
        "Please work in a conda environment with lalsuite and cwinpy installed"
    )

execpath = os.path.join(execpath, "bin")

lppen = os.path.join(execpath, "lalpulsar_parameter_estimation_nested")
n2p = os.path.join(execpath, "lalinference_nest2pos")

Nlive = 1000  # number of nested sampling live points
Nmcmcinitial = 0  # set to 0 so that prior samples are not resampled

outfile = os.path.join(outdir, "{}_nest.hdf".format(label))

# set ephemeris files
efile = download_file(DOWNLOAD_URL.format("earth00-40-DE405.dat.gz"), cache=True)
sfile = download_file(DOWNLOAD_URL.format("sun00-40-DE405.dat.gz"), cache=True)
tfile = download_file(DOWNLOAD_URL.format("te405_2000-2040.dat.gz"), cache=True)

# set the command line arguments
runcmd = " ".join(
    [
        lppen,
        "--verbose",
        "--input-files",
        ",".join(hetfiles),
        "--detectors",
        ",".join(detectors),
        "--par-file",
        parfile,
        "--prior-file",
        priorfile,
        "--Nlive",
        "{}".format(Nlive),
        "--Nmcmcinitial",
        "{}".format(Nmcmcinitial),
        "--outfile",
        outfile,
        "--ephem-earth",
        efile,
        "--ephem-sun",
        sfile,
        "--ephem-timecorr",
        tfile,
    ]
)

with sp.Popen(
    runcmd,
    stdout=sp.PIPE,
    stderr=sp.PIPE,
    shell=True,
    bufsize=1,
    universal_newlines=True,
) as p:
    for line in p.stderr:
        print(line, end="")

# convert nested samples to posterior samples
outpost = os.path.join(outdir, "{}_post.hdf".format(label))
runcmd = " ".join([n2p, "-p", outpost, outfile])
with sp.Popen(
    runcmd,
    stdout=sp.PIPE,
    stderr=sp.PIPE,
    shell=True,
    bufsize=1,
    universal_newlines=True,
) as p:
    for line in p.stdout:
        print(line, end="")

# get posterior samples
post = read_samples(outpost, tablename=LALInferenceHDF5PosteriorSamplesDatasetName)
lp = len(post["H0"])
postsamples = np.zeros((lp, len(priors)))
for i, p in enumerate(priors.keys()):
    postsamples[:, i] = post[p.upper()]

# get evidence
hdf = h5py.File(outpost, "r")
a = hdf["lalinference"]["lalinference_nest"]
evsig = a.attrs["log_evidence"]
evnoise = a.attrs["log_noise_evidence"]
hdf.close()

# run bilby via the pe interface
runner = pe(
    data_file=hetfiles,
    par_file=parfile,
    prior=priors,
    detector=detectors,
    outdir=outdir,
    label=label,
)

result = runner.result

# evaluate the likelihood on a grid
gridpoints = 50
grid_size = dict()
for p in priors.keys():
    grid_size[p] = np.linspace(
        np.min(result.posterior[p]), np.max(result.posterior[p]), gridpoints
    )

grunner = pe(
    data_file=hetfiles,
    par_file=parfile,
    prior=priors,
    detector=detectors,
    outdir=outdir,
    label=label,
    grid=True,
    grid_kwargs={"grid_size": grid_size},
)

grid = grunner.grid

# output comparisons
comparisons(label, outdir, grid, priors, cred=0.9)

# create results plot
allresults = {
    "lalpulsar_parameter_estimation_nested": outpost,
    "cwinpy_pe": result,
    "cwinpy_pe (grid)": grid,
}

colors = {
    key: plt.rcParams["axes.prop_cycle"].by_key()["color"][i]
    for i, key in enumerate(allresults.keys())
}

plot = Plot(
    results=allresults,
    parameters=list(priors.keys()),
    plottype="corner",
    pulsar=parfile,
)

plot.plot(
    bins=50,
    smooth=0.9,
    quantiles=[0.16, 0.84],
    levels=(1 - np.exp(-0.5), 1 - np.exp(-2), 1 - np.exp(-9 / 2.0)),
    fill_contours=True,
    colors=colors,
)

plot.savefig(os.path.join(outdir, "{}_corner.png".format(label)), dpi=150)