Resolve "Add histogram distribution as option in the hierarchical analysis"
Resolves #45 (closed).
This adds a non-parametric histogram-style distribution to the hierarchical analysis options.
Currently the tests pass, but trying to use this for an analysis fails (e.g., using test result in '/home/matt/cwinpy_test/pe/outdir') with:
from bilby.core.result import ResultList
from cwinpy.hierarchical import MassQuadrupoleDistribution
data = ResultList(["gaussian_example_result.json"])
data[0].posterior["Q22"] = data[0].posterior["x"]
data[0].priors["Q22"] = data[0].priors["x"]
distkwargs = {"low": data[0].posterior["Q22"].min(), "high": data[0].posterior["Q22"].max(), "nbins": 10}
intmethod = "numerical"
sampler_kwargs = {
"sample": "unif",
"nlive": 500,
"gzip": True,
"outdir": "exponential_distribution",
"label": intmethod,
"check_point_plot": False,
}
mqd = MassQuadrupoleDistribution(
data=data,
distribution="histogram",
distkwargs=distkwargs,
sampler_kwargs=sampler_kwargs,
integration_method=intmethod,
nsamples=500, # number of Q22 samples to store/use
)
res = mqd.sample()gives:
19:06 bilby INFO : Running for label 'numerical', output will be saved to 'exponential_distribution'
19:06 bilby WARNING : Parameter weight has no default prior and is set to None, this will not be sampled and may cause an error.
19:06 bilby INFO : Search parameters:
19:06 bilby INFO : weight0 = Dirichlet(order=0, n_dimensions=10, label='weight')
19:06 bilby INFO : weight1 = Dirichlet(order=1, n_dimensions=10, label='weight')
19:06 bilby INFO : weight2 = Dirichlet(order=2, n_dimensions=10, label='weight')
19:06 bilby INFO : weight3 = Dirichlet(order=3, n_dimensions=10, label='weight')
19:06 bilby INFO : weight4 = Dirichlet(order=4, n_dimensions=10, label='weight')
19:06 bilby INFO : weight5 = Dirichlet(order=5, n_dimensions=10, label='weight')
19:06 bilby INFO : weight6 = Dirichlet(order=6, n_dimensions=10, label='weight')
19:06 bilby INFO : weight7 = Dirichlet(order=7, n_dimensions=10, label='weight')
19:06 bilby INFO : weight8 = Dirichlet(order=8, n_dimensions=10, label='weight')
19:06 bilby INFO : Single likelihood evaluation took 3.556e-04 s
0it [00:00, ?it/s]19:06 bilby INFO : Using sampler Dynesty with kwargs {'bound': 'multi', 'sample': 'unif', 'verbose': True, 'periodic': None, 'reflective': None, 'check_point_delta_t': 600, 'nlive': 500, 'first_update': None, 'walks': 100, 'npdim': None, 'rstate': None, 'queue_size': 1, 'pool': None, 'use_pool': None, 'live_points': None, 'logl_args': None, 'logl_kwargs': None, 'ptform_args': None, 'ptform_kwargs': None, 'enlarge': 1.5, 'bootstrap': None, 'vol_dec': 0.5, 'vol_check': 8.0, 'facc': 0.2, 'slices': 5, 'update_interval': 300, 'print_func': <bound method Dynesty._print_func of <bilby.core.sampler.dynesty.Dynesty object at 0x7fad2e9e8c40>>, 'dlogz': 0.1, 'maxiter': None, 'maxcall': None, 'logl_max': inf, 'add_live': True, 'print_progress': True, 'save_bounds': False, 'n_effective': None, 'maxmcmc': 5000, 'nact': 5}
19:06 bilby INFO : Checkpoint every check_point_delta_t = 600s
19:06 bilby INFO : Using dynesty version 1.0.1
19:06 bilby INFO : Resume file exponential_distribution/numerical_resume.pickle does not exist.
19:06 bilby INFO : Generating initial points from the prior
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
<ipython-input-99-62c69578641e> in <module>
----> 1 res = mqd.sample()
~/miniconda3/envs/cwinpy/lib/python3.8/site-packages/cwinpy/hierarchical.py in sample(self, **run_kwargs)
1673 )
1674 else:
-> 1675 self._result = bilby.run_sampler(
1676 likelihood=self._likelihood,
1677 priors=self._prior,
~/miniconda3/envs/cwinpy/lib/python3.8/site-packages/bilby/core/sampler/__init__.py in run_sampler(likelihood, priors, label, outdir, sampler, use_ratio, injection_parameters, conversion_function, plot, default_priors_file, clean, meta_data, save, gzip, result_class, npool, **kwargs)
182 result = sampler._run_test()
183 else:
--> 184 result = sampler.run_sampler()
185 end_time = datetime.datetime.now()
186
~/miniconda3/envs/cwinpy/lib/python3.8/site-packages/bilby/core/sampler/dynesty.py in run_sampler(self)
354 if self.kwargs['live_points'] is None:
355 self.kwargs['live_points'] = (
--> 356 self.get_initial_points_from_prior(self.kwargs['nlive'])
357 )
358 self.sampler = dynesty.NestedSampler(
~/miniconda3/envs/cwinpy/lib/python3.8/site-packages/bilby/core/sampler/base_sampler.py in get_initial_points_from_prior(self, npoints)
440 while len(unit_cube) < npoints:
441 unit = np.random.rand(self.ndim)
--> 442 theta = self.prior_transform(unit)
443 if self.check_draw(theta, warning=False):
444 unit_cube.append(unit)
~/miniconda3/envs/cwinpy/lib/python3.8/site-packages/bilby/core/sampler/dynesty.py in prior_transform(self, theta)
689
690 """
--> 691 return self.priors.rescale(self._search_parameter_keys, theta)
692
693
~/miniconda3/envs/cwinpy/lib/python3.8/site-packages/bilby/core/prior/dict.py in rescale(self, keys, theta)
688 required_variables = {k: result[k] for k in getattr(self[key], 'required_variables', [])}
689 result[key] = self[key].rescale(theta[index], **required_variables)
--> 690 return [result[key] for key in keys]
691
692 def _update_rescale_keys(self, keys):
~/miniconda3/envs/cwinpy/lib/python3.8/site-packages/bilby/core/prior/dict.py in <listcomp>(.0)
688 required_variables = {k: result[k] for k in getattr(self[key], 'required_variables', [])}
689 result[key] = self[key].rescale(theta[index], **required_variables)
--> 690 return [result[key] for key in keys]
691
692 def _update_rescale_keys(self, keys):
I'm not sure if this is due to the implementation in CWInPy, or just an issue with bilby.
Edited by Matthew Pitkin