def test_bilby_to_lalinference(self):
mass_1 = [1, 20]
mass_2 = [1, 20]
chirp_mass = [1, 5]
mass_ratio = [0.125, 1]
bilby_prior = BBHPriorDict(dictionary=dict(
chirp_mass=Uniform(name='chirp_mass', minimum=chirp_mass[0], maximum=chirp_mass[1]),
mass_ratio=Uniform(name='mass_ratio', minimum=mass_ratio[0], maximum=mass_ratio[1]),
mass_2=Constraint(name='mass_2', minimum=mass_1[0], maximum=mass_1[1]),
mass_1=Constraint(name='mass_1', minimum=mass_2[0], maximum=mass_2[1])))
lalinf_prior = BBHPriorDict(dictionary=dict(
mass_ratio=Constraint(name='mass_ratio', minimum=mass_ratio[0], maximum=mass_ratio[1]),
chirp_mass=Constraint(name='chirp_mass', minimum=chirp_mass[0], maximum=chirp_mass[1]),
mass_2=Uniform(name='mass_2', minimum=mass_1[0], maximum=mass_1[1]),
mass_1=Uniform(name='mass_1', minimum=mass_2[0], maximum=mass_2[1])))
nsamples = 5000
bilby_samples = bilby_prior.sample(nsamples)
bilby_samples, _ = conversion.convert_to_lal_binary_black_hole_parameters(
bilby_samples)
# Quicker way to generate LA prior samples (rather than specifying Constraint)
lalinf_samples = []
while len(lalinf_samples) < nsamples:
s = lalinf_prior.sample()
if s["mass_1"] < s["mass_2"]:
s["mass_1"], s["mass_2"] = s["mass_2"], s["mass_1"]
if s["mass_2"] / s["mass_1"] > 0.125:
lalinf_samples.append(s)
lalinf_samples = pd.DataFrame(lalinf_samples)
lalinf_samples["mass_ratio"] = lalinf_samples["mass_2"] / lalinf_samples["mass_1"]
# Construct fake result object
result = bilby.core.result.Result()
result.search_parameter_keys = ["mass_ratio", "chirp_mass"]
result.meta_data = dict()
result.priors = bilby_prior
result.posterior = pd.DataFrame(bilby_samples)
result_converted = bilby.gw.prior.convert_to_flat_in_component_mass_prior(result)
if "plot" in sys.argv:
# Useful for debugging
plt.hist(bilby_samples["mass_ratio"], bins=50, density=True, alpha=0.5)
plt.hist(result_converted.posterior["mass_ratio"], bins=50, density=True, alpha=0.5)
plt.hist(lalinf_samples["mass_ratio"], bins=50, alpha=0.5, density=True)
plt.show()
# Check that the non-reweighted posteriors fail a KS test
ks = ks_2samp(bilby_samples["mass_ratio"], lalinf_samples["mass_ratio"])
print("Non-reweighted KS test = ", ks)
self.assertFalse(ks.pvalue > 0.05)
# Check that the non-reweighted posteriors pass a KS test
ks = ks_2samp(result_converted.posterior["mass_ratio"], lalinf_samples["mass_ratio"])
print("Reweighted KS test = ", ks)
self.assertTrue(ks.pvalue > 0.001)
def test_absolute_overlap(self):
priors = BBHPriorDict(aligned_spin=True)
del priors["mass_1"], priors["mass_2"]
priors["total_mass"] = Uniform(5, 50)
priors["mass_ratio"] = Uniform(0.5, 1)
priors["geocent_time"] = Uniform(-10, 10)
n_samples = 100
all_parameters = pd.DataFrame(priors.sample(n_samples))
overlaps = list()
for ii in range(n_samples):
parameters = dict(all_parameters.iloc[ii])
bilby_pols = self.bilby_wfg.frequency_domain_strain(parameters)
gpu_pols = self.gpu_wfg.frequency_domain_strain(parameters)
bilby_strain = self.ifo.get_detector_response(
waveform_polarizations=bilby_pols, parameters=parameters)
gpu_strain = self.ifo.get_detector_response(
waveform_polarizations=gpu_pols, parameters=parameters)
inner_product = noise_weighted_inner_product(
aa=bilby_strain,
bb=gpu_strain,
power_spectral_density=self.ifo.power_spectral_density_array,
duration=self.duration)
overlap = (inner_product /
self.ifo.optimal_snr_squared(signal=bilby_strain)**0.5 /
self.ifo.optimal_snr_squared(signal=gpu_strain)**0.5)
overlaps.append(overlap)
self.assertTrue(min(np.abs(overlaps)) > 0.995)
class Samples:
def __init__(self, prior_file, num_samples=1000):
self.prior_filename = prior_file
self.prior = BBHPriorDict(filename=prior_file)
self.num_samples = num_samples
self.samples = self.__generate_samples()
def __generate_samples(self):
samples = pd.DataFrame(self.prior.sample(size=self.num_samples))
samples = conversion.generate_all_bbh_parameters(samples)
samples = conversion.generate_component_spins(samples)
if 'chi_1' not in samples:
samples['chi_1'] = samples['spin_1z']
samples['chi_2'] = samples['spin_2z']
samples[REMNANT_KICK] = None
samples[REMNANT_MASS] = None
samples[REMNANT_SPIN] = None
return samples
def add_remnant_data_to_samples(self):
for idx, s in tqdm.tqdm(self.samples.iterrows(),
total=self.num_samples,
desc="Merging BH"):
remnant = merge_bbh_pair(
bh_1=BlackHole(mass=s.mass_1,
spin=[s.spin_1x, s.spin_1y, s.spin_1z]),
bh_2=BlackHole(mass=s.mass_2,
spin=[s.spin_2x, s.spin_2y, s.spin_2z]))
self.samples.at[idx, REMNANT_MASS] = remnant.mass
self.samples.at[idx, REMNANT_SPIN] = remnant.spin_mag
self.samples.at[idx, REMNANT_KICK] = remnant.kick_mag
def plot_corner(self, filename):
param_of_interest = [
"chi_1", "chi_2", "mass_ratio", "remnant_kick", "remnant_spin"
]
if "chi_1_in_plane" in self.samples:
param_of_interest += ["chi_1_in_plane", "chi_2_in_plane"]
corner.corner(self.samples[param_of_interest], **CORNER_KWARGS)
font = {'color': 'darkblue', 'weight': 'bold'}
plt.suptitle(filename, fontdict=font, fontsize=40)
plt.savefig(f"{filename}.png")
plt.close()
def save_samples(self):
self.samples.to_csv(self.prior_filename.replace(".prior", ".csv"),
index=False)