Merge branch 'fix-pos0-setting-and-add-example' into 'master'

Adds an example of custom initialising emcee chains and fixes a bug

See merge request !269

bilby/core/sampler/emcee.py

@@ -121,7 +121,7 @@ class Emcee(MCMCSampler):
         return self.result
 
     def _set_pos0(self):
-        if self.pos0:
+        if self.pos0 is not None:
             logger.debug("Using given initial positions for walkers")
             if isinstance(self.pos0, DataFrame):
                 self.pos0 = self.pos0[self.search_parameter_keys].values

examples/other_examples/starting_mcmc_chains_near_to_the_peak.py

+#!/usr/bin/env python
+"""
+An example of using emcee, but starting the walkers off close to the peak (or
+any other arbitrary point). This is based off the
+linear_regression_with_unknown_noise.py example.
+"""
+from __future__ import division
+import bilby
+import numpy as np
+import pandas as pd
+
+# A few simple setup steps
+label = 'starting_mcmc_chains_near_to_the_peak'
+outdir = 'outdir'
+
+
+# First, we define our "signal model", in this case a simple linear function
+def model(time, m, c):
+    return time * m + c
+
+
+# Now we define the injection parameters which we make simulated data with
+injection_parameters = dict(m=0.5, c=0.2)
+
+# For this example, we'll inject standard Gaussian noise
+sigma = 1
+
+# These lines of code generate the fake data
+sampling_frequency = 10
+time_duration = 10
+time = np.arange(0, time_duration, 1 / sampling_frequency)
+N = len(time)
+data = model(time, **injection_parameters) + np.random.normal(0, sigma, N)
+
+# Now lets instantiate the built-in GaussianLikelihood, giving it
+# the time, data and signal model. Note that, because we do not give it the
+# parameter, sigma is unknown and marginalised over during the sampling
+likelihood = bilby.core.likelihood.GaussianLikelihood(time, data, model)
+
+# Here we define the prior distribution used while sampling
+priors = bilby.core.prior.PriorDict()
+priors['m'] = bilby.core.prior.Uniform(0, 5, 'm')
+priors['c'] = bilby.core.prior.Uniform(-2, 2, 'c')
+priors['sigma'] = bilby.core.prior.Uniform(0, 10, 'sigma')
+
+# Set values to determine how to sample with emcee
+nwalkers = 100
+nsteps = 1000
+sampler = 'emcee'
+
+# Run the sampler from the default pos0 (which is samples drawn from the prior)
+result = bilby.run_sampler(
+    likelihood=likelihood, priors=priors, sampler=sampler, nsteps=nsteps,
+    nwalkers=nwalkers, outdir=outdir, label=label + 'default_pos0')
+result.plot_walkers()
+
+
+# Here we define a distribution from which to start the walkers off from.
+start_pos = bilby.core.prior.PriorDict()
+start_pos['m'] = bilby.core.prior.Normal(injection_parameters['m'], 0.1)
+start_pos['c'] = bilby.core.prior.Normal(injection_parameters['c'], 0.1)
+start_pos['sigma'] = bilby.core.prior.Normal(sigma, 0.1)
+
+# This line generated the initial starting position data frame by sampling
+# nwalkers-times from the start_pos distribution. Note, you can
+# generate this is anyway you like, but it must be a DataFrame with a length
+# equal to the number of walkers
+pos0 = pd.DataFrame(start_pos.sample(nwalkers))
+
+# Run the sampler with our created pos0
+result = bilby.run_sampler(
+    likelihood=likelihood, priors=priors, sampler=sampler, nsteps=nsteps,
+    nwalkers=nwalkers, outdir=outdir, label=label + 'user_pos0', pos0=pos0)
+result.plot_walkers()
+
+
+# After running this script, in the outdir, you'll find to png images showing
+# the result of the runs with and without the initialisation.