# Fitting a model to data with both x and y errors with Bilby¶

Usually when we fit a model to data with a Gaussian Likelihood we assume that we know x values exactly. This is almost never the case. Here we show how to fit a model with errors in both x and y.

[1]:

import bilby
import matplotlib.pyplot as plt
import numpy as np

%matplotlib inline


## Simulate data¶

First we create the data and plot it

[2]:

# define our model, a line
def model(x, m, c, **kwargs):
y = m * x + c
return y

# make a function to create and plot our data
def make_data(points, m, c, xerr, yerr, seed):
np.random.seed(int(seed))
xtrue = np.linspace(0, 100, points)
ytrue = model(x=xtrue, m=m, c=c)

xerr_vals = xerr * np.random.randn(points)
yerr_vals = yerr * np.random.randn(points)
xobs = xtrue + xerr_vals
yobs = ytrue + yerr_vals

plt.errorbar(xobs, yobs, xerr=xerr, yerr=yerr, fmt="x")
plt.errorbar(xtrue, ytrue, yerr=yerr, color="black", alpha=0.5)
plt.xlim(0, 100)
plt.show()
plt.close()

data = {
"xtrue": xtrue,
"ytrue": ytrue,
"xobs": xobs,
"yobs": yobs,
"xerr": xerr,
"yerr": yerr,
}

return data

data = make_data(points=30, m=5, c=10, xerr=5, yerr=5, seed=123)


## Define our prior and sampler settings¶

Now lets set up the prior and bilby output directory/sampler settings

[3]:

# setting up bilby priors
priors = dict(
m=bilby.core.prior.Uniform(0, 30, "m"), c=bilby.core.prior.Uniform(0, 30, "c")
)

sampler_kwargs = dict(priors=priors, sampler="bilby_mcmc", nsamples=1000, printdt=5, outdir="outdir", verbose=False, clean=True)


## Fit with exactly known x-values¶

Our first step is to recover the straight line using a simple Gaussian Likelihood that only takes into account the y errors. Under the assumption we know x exactly. In this case, we pass in xtrue for x

[4]:

known_x = bilby.core.likelihood.GaussianLikelihood(
x=data["xtrue"], y=data["yobs"], func=model, sigma=data["yerr"]
)
result_known_x = bilby.run_sampler(
likelihood=known_x,
label="known_x",
**sampler_kwargs,
)

17:28 bilby INFO    : Running for label 'known_x', output will be saved to 'outdir'
17:28 bilby INFO    : Analysis priors:
17:28 bilby INFO    : m=Uniform(minimum=0, maximum=30, name='m', latex_label='m', unit=None, boundary=None)
17:28 bilby INFO    : c=Uniform(minimum=0, maximum=30, name='c', latex_label='c', unit=None, boundary=None)
17:28 bilby INFO    : Analysis likelihood class: <class 'bilby.core.likelihood.GaussianLikelihood'>
17:28 bilby INFO    : Analysis likelihood noise evidence: nan
17:28 bilby INFO    : Single likelihood evaluation took 8.364e-05 s
17:28 bilby INFO    : Using sampler Bilby_MCMC with kwargs {'nsamples': 1000, 'nensemble': 1, 'pt_ensemble': False, 'ntemps': 1, 'Tmax': None, 'Tmax_from_SNR': 20, 'initial_betas': None, 'adapt': True, 'adapt_t0': 100, 'adapt_nu': 10, 'pt_rejection_sample': False, 'burn_in_nact': 10, 'thin_by_nact': 1, 'fixed_discard': 0, 'autocorr_c': 5, 'L1steps': 100, 'L2steps': 3, 'printdt': 5, 'min_tau': 1, 'proposal_cycle': 'default', 'stop_after_convergence': False, 'fixed_tau': None, 'tau_window': None, 'evidence_method': 'stepping_stone', 'initial_sample_method': 'prior', 'initial_sample_dict': None}
17:28 bilby INFO    : Initializing BilbyPTMCMCSampler with:
Convergence settings: ConvergenceInputs(autocorr_c=5, burn_in_nact=10, thin_by_nact=1, fixed_discard=0, target_nsamples=1000, stop_after_convergence=False, L1steps=100, L2steps=3, min_tau=1, fixed_tau=None, tau_window=None)
proposal_cycle: default
pt_rejection_sample: False
17:28 bilby INFO    : Initializing BilbyPTMCMCSampler with:ntemps=1, nensemble=1, pt_ensemble=False, initial_betas=[1], initial_sample_method=prior, initial_sample_dict=None

17:28 bilby INFO    : Using initial sample {'m': 9.01830406509614, 'c': 19.033268036983173}
17:28 bilby INFO    : Using ProposalCycle:
DifferentialEvolutionProposal(acceptance_ratio:nan,n:0,)
UniformProposal(acceptance_ratio:nan,n:0,)
KDEProposal(acceptance_ratio:nan,n:0,trained:0,)
GMMProposal(acceptance_ratio:nan,n:0,trained:0,)

17:28 bilby INFO    : Setting convergence_inputs=ConvergenceInputs(autocorr_c=5, burn_in_nact=10, thin_by_nact=1, fixed_discard=0, target_nsamples=1000, stop_after_convergence=False, L1steps=100, L2steps=3, min_tau=1, fixed_tau=None, tau_window=None)
17:28 bilby INFO    : Drawing 1000 samples
17:28 bilby INFO    : Checkpoint every 1800s
17:28 bilby INFO    : Print update every 5s
17:28 bilby WARNING : Non-negligible print progress time (ppt_frac=0.02)
17:28 bilby INFO    : Reached convergence: exiting sampling
17:28 bilby INFO    : Checkpoint start
17:28 bilby INFO    : Written checkpoint file outdir/known_x_resume.pickle
17:28 bilby INFO    : Zero-temperature proposals:
17:28 bilby INFO    : DifferentialEvolutionProposal(acceptance_ratio:0.46,n:3.6e+04,)
17:28 bilby INFO    : UniformProposal(acceptance_ratio:1,n:5.9e+02,)
17:28 bilby INFO    : KDEProposal(acceptance_ratio:0.00015,n:3.4e+04,trained:0,)
17:28 bilby INFO    : GMMProposal(acceptance_ratio:5.9e-05,n:3.4e+04,trained:0,)
17:28 bilby INFO    : Current taus={'m': 1.2, 'c': 1.1}
17:28 bilby INFO    : Creating diagnostic plots
17:28 bilby INFO    : Checkpoint finished
17:28 bilby INFO    : Sampling time: 0:00:20.036122
17:28 bilby INFO    : Summary of results:
nsamples: 1279
ln_noise_evidence:    nan
ln_evidence:    nan +/-    nan
ln_bayes_factor:    nan +/-    nan


[5]:

_ = result_known_x.plot_corner(truth=dict(m=5, c=10), titles=True, save=False)
plt.show()
plt.close()


## Fit with unmodeled uncertainty in the x-values¶

As expected this is easy to recover and the sampler does a good job. However this was made too easy - by passing in the ‘true’ values of x. Lets see what happens when we pass in the observed values of x

[6]:

incorrect_x = bilby.core.likelihood.GaussianLikelihood(
x=data["xobs"], y=data["yobs"], func=model, sigma=data["yerr"]
)
result_incorrect_x = bilby.run_sampler(
likelihood=incorrect_x,
label="incorrect_x",
**sampler_kwargs,
)

17:28 bilby INFO    : Running for label 'incorrect_x', output will be saved to 'outdir'
17:29 bilby INFO    : Analysis priors:
17:29 bilby INFO    : m=Uniform(minimum=0, maximum=30, name='m', latex_label='m', unit=None, boundary=None)
17:29 bilby INFO    : c=Uniform(minimum=0, maximum=30, name='c', latex_label='c', unit=None, boundary=None)
17:29 bilby INFO    : Analysis likelihood class: <class 'bilby.core.likelihood.GaussianLikelihood'>
17:29 bilby INFO    : Analysis likelihood noise evidence: nan
17:29 bilby INFO    : Single likelihood evaluation took 9.753e-05 s
17:29 bilby INFO    : Using sampler Bilby_MCMC with kwargs {'nsamples': 1000, 'nensemble': 1, 'pt_ensemble': False, 'ntemps': 1, 'Tmax': None, 'Tmax_from_SNR': 20, 'initial_betas': None, 'adapt': True, 'adapt_t0': 100, 'adapt_nu': 10, 'pt_rejection_sample': False, 'burn_in_nact': 10, 'thin_by_nact': 1, 'fixed_discard': 0, 'autocorr_c': 5, 'L1steps': 100, 'L2steps': 3, 'printdt': 5, 'min_tau': 1, 'proposal_cycle': 'default', 'stop_after_convergence': False, 'fixed_tau': None, 'tau_window': None, 'evidence_method': 'stepping_stone', 'initial_sample_method': 'prior', 'initial_sample_dict': None}
17:29 bilby INFO    : Initializing BilbyPTMCMCSampler with:
Convergence settings: ConvergenceInputs(autocorr_c=5, burn_in_nact=10, thin_by_nact=1, fixed_discard=0, target_nsamples=1000, stop_after_convergence=False, L1steps=100, L2steps=3, min_tau=1, fixed_tau=None, tau_window=None)
proposal_cycle: default
pt_rejection_sample: False
17:29 bilby INFO    : Initializing BilbyPTMCMCSampler with:ntemps=1, nensemble=1, pt_ensemble=False, initial_betas=[1], initial_sample_method=prior, initial_sample_dict=None

17:29 bilby INFO    : Using initial sample {'m': 17.118647293988026, 'c': 13.52706901382501}
17:29 bilby INFO    : Using ProposalCycle:
DifferentialEvolutionProposal(acceptance_ratio:nan,n:0,)
UniformProposal(acceptance_ratio:nan,n:0,)
KDEProposal(acceptance_ratio:nan,n:0,trained:0,)
GMMProposal(acceptance_ratio:nan,n:0,trained:0,)

17:29 bilby INFO    : Setting convergence_inputs=ConvergenceInputs(autocorr_c=5, burn_in_nact=10, thin_by_nact=1, fixed_discard=0, target_nsamples=1000, stop_after_convergence=False, L1steps=100, L2steps=3, min_tau=1, fixed_tau=None, tau_window=None)
17:29 bilby INFO    : Drawing 1000 samples
17:29 bilby INFO    : Checkpoint every 1800s
17:29 bilby INFO    : Print update every 5s
17:29 bilby INFO    : Reached convergence: exiting sampling
17:29 bilby INFO    : Checkpoint start
17:29 bilby INFO    : Written checkpoint file outdir/incorrect_x_resume.pickle
17:29 bilby INFO    : Zero-temperature proposals:
17:29 bilby INFO    : DifferentialEvolutionProposal(acceptance_ratio:0.46,n:3.8e+04,)
17:29 bilby INFO    : UniformProposal(acceptance_ratio:1,n:4.3e+02,)
17:29 bilby INFO    : KDEProposal(acceptance_ratio:0.00014,n:3.5e+04,trained:0,)
17:29 bilby INFO    : GMMProposal(acceptance_ratio:8.8e-05,n:3.4e+04,trained:0,)
17:29 bilby INFO    : Current taus={'m': 1, 'c': 1.0}
17:29 bilby INFO    : Creating diagnostic plots
17:29 bilby INFO    : Checkpoint finished
17:29 bilby INFO    : Sampling time: 0:00:20.018757
17:29 bilby INFO    : Summary of results:
nsamples: 1318
ln_noise_evidence:    nan
ln_evidence:    nan +/-    nan
ln_bayes_factor:    nan +/-    nan


[7]:

_ = result_incorrect_x.plot_corner(truth=dict(m=5, c=10), titles=True, save=False)
plt.show()
plt.close()


## Fit with modeled uncertainty in x-values¶

This is not good as there is unmodelled uncertainty in our x values. Getting around this requires marginalisation of the true x values or sampling over them. See discussion in section 7 of https://arxiv.org/pdf/1008.4686.pdf.

For this, we will have to define a new likelihood class. By subclassing the base bilby.core.likelihood.Likelihood class we can do this fairly simply.

[8]:

class GaussianLikelihoodUncertainX(bilby.core.likelihood.Likelihood):
def __init__(self, xobs, yobs, xerr, yerr, function):
"""

Parameters
----------
xobs, yobs: array_like
The data to analyse
xerr, yerr: array_like
The standard deviation of the noise
function:
The python function to fit to the data
"""
super(GaussianLikelihoodUncertainX, self).__init__(dict())
self.xobs = xobs
self.yobs = yobs
self.yerr = yerr
self.xerr = xerr
self.function = function

def log_likelihood(self):
variance = (self.xerr * self.parameters["m"]) ** 2 + self.yerr**2
model_y = self.function(self.xobs, **self.parameters)
residual = self.yobs - model_y

ll = -0.5 * np.sum(residual**2 / variance + np.log(variance))

return -0.5 * np.sum(residual**2 / variance + np.log(variance))

[9]:

gaussian_unknown_x = GaussianLikelihoodUncertainX(
xobs=data["xobs"],
yobs=data["yobs"],
xerr=data["xerr"],
yerr=data["yerr"],
function=model,
)
result_unknown_x = bilby.run_sampler(
likelihood=gaussian_unknown_x,
label="unknown_x",
**sampler_kwargs,
)

17:29 bilby INFO    : Running for label 'unknown_x', output will be saved to 'outdir'
17:29 bilby INFO    : Analysis priors:
17:29 bilby INFO    : m=Uniform(minimum=0, maximum=30, name='m', latex_label='m', unit=None, boundary=None)
17:29 bilby INFO    : c=Uniform(minimum=0, maximum=30, name='c', latex_label='c', unit=None, boundary=None)
17:29 bilby INFO    : Analysis likelihood class: <class '__main__.GaussianLikelihoodUncertainX'>
17:29 bilby INFO    : Analysis likelihood noise evidence: nan
17:29 bilby INFO    : Single likelihood evaluation took 1.050e-04 s
17:29 bilby INFO    : Using sampler Bilby_MCMC with kwargs {'nsamples': 1000, 'nensemble': 1, 'pt_ensemble': False, 'ntemps': 1, 'Tmax': None, 'Tmax_from_SNR': 20, 'initial_betas': None, 'adapt': True, 'adapt_t0': 100, 'adapt_nu': 10, 'pt_rejection_sample': False, 'burn_in_nact': 10, 'thin_by_nact': 1, 'fixed_discard': 0, 'autocorr_c': 5, 'L1steps': 100, 'L2steps': 3, 'printdt': 5, 'min_tau': 1, 'proposal_cycle': 'default', 'stop_after_convergence': False, 'fixed_tau': None, 'tau_window': None, 'evidence_method': 'stepping_stone', 'initial_sample_method': 'prior', 'initial_sample_dict': None}
17:29 bilby INFO    : Initializing BilbyPTMCMCSampler with:
Convergence settings: ConvergenceInputs(autocorr_c=5, burn_in_nact=10, thin_by_nact=1, fixed_discard=0, target_nsamples=1000, stop_after_convergence=False, L1steps=100, L2steps=3, min_tau=1, fixed_tau=None, tau_window=None)
proposal_cycle: default
pt_rejection_sample: False
17:29 bilby INFO    : Initializing BilbyPTMCMCSampler with:ntemps=1, nensemble=1, pt_ensemble=False, initial_betas=[1], initial_sample_method=prior, initial_sample_dict=None

17:29 bilby INFO    : Using initial sample {'m': 19.69974063033767, 'c': 17.996037868874076}
17:29 bilby INFO    : Using ProposalCycle:
DifferentialEvolutionProposal(acceptance_ratio:nan,n:0,)
UniformProposal(acceptance_ratio:nan,n:0,)
KDEProposal(acceptance_ratio:nan,n:0,trained:0,)
GMMProposal(acceptance_ratio:nan,n:0,trained:0,)

17:29 bilby INFO    : Setting convergence_inputs=ConvergenceInputs(autocorr_c=5, burn_in_nact=10, thin_by_nact=1, fixed_discard=0, target_nsamples=1000, stop_after_convergence=False, L1steps=100, L2steps=3, min_tau=1, fixed_tau=None, tau_window=None)
17:29 bilby INFO    : Drawing 1000 samples
17:29 bilby INFO    : Checkpoint every 1800s
17:29 bilby INFO    : Print update every 5s
17:29 bilby INFO    : Reached convergence: exiting sampling
17:29 bilby INFO    : Checkpoint start
17:29 bilby INFO    : Written checkpoint file outdir/unknown_x_resume.pickle
17:29 bilby INFO    : Zero-temperature proposals:
17:29 bilby INFO    : DifferentialEvolutionProposal(acceptance_ratio:0.46,n:3.4e+04,)
17:29 bilby INFO    : UniformProposal(acceptance_ratio:1,n:5e+02,)
17:29 bilby INFO    : KDEProposal(acceptance_ratio:0.0014,n:3.6e+04,trained:0,)
17:29 bilby INFO    : GMMProposal(acceptance_ratio:0.00085,n:3.5e+04,trained:0,)
17:29 bilby INFO    : Current taus={'m': 1.0, 'c': 1}
17:29 bilby INFO    : Creating diagnostic plots
17:29 bilby INFO    : Checkpoint finished
17:29 bilby INFO    : Sampling time: 0:00:20.042728
17:29 bilby INFO    : Summary of results:
nsamples: 1332
ln_noise_evidence:    nan
ln_evidence:    nan +/-    nan
ln_bayes_factor:    nan +/-    nan


[10]:

_ = result_unknown_x.plot_corner(truth=dict(m=5, c=10), titles=True, save=False)
plt.show()
plt.close()


Success! The inferred posterior is consistent with the true values.