Coverage for bilby/core/sampler/pymultinest.py: 53%
77 statements
« prev ^ index » next coverage.py v7.6.1, created at 2025-05-06 04:57 +0000
1import datetime
2import importlib
3import os
4import time
6import numpy as np
8from .base_sampler import NestedSampler, _TemporaryFileSamplerMixin, signal_wrapper
11class Pymultinest(_TemporaryFileSamplerMixin, NestedSampler):
12 """
13 bilby wrapper of pymultinest
14 (https://github.com/JohannesBuchner/PyMultiNest)
16 All positional and keyword arguments (i.e., the args and kwargs) passed to
17 `run_sampler` will be propagated to `pymultinest.run`, see documentation
18 for that class for further help. Under Other Parameters, we list commonly
19 used kwargs and the bilby defaults.
21 Parameters
22 ==========
23 npoints: int
24 The number of live points, note this can also equivalently be given as
25 one of [nlive, nlives, n_live_points]
26 importance_nested_sampling: bool, (False)
27 If true, use importance nested sampling
28 sampling_efficiency: float or {'parameter', 'model'}, ('parameter')
29 Defines the sampling efficiency
30 verbose: Bool
31 If true, print information information about the convergence during
32 resume: bool
33 If true, resume run from checkpoint (if available)
35 """
37 sampler_name = "pymultinest"
38 abbreviation = "pm"
39 default_kwargs = dict(
40 importance_nested_sampling=False,
41 resume=True,
42 verbose=True,
43 sampling_efficiency="parameter",
44 n_live_points=500,
45 n_params=None,
46 n_clustering_params=None,
47 wrapped_params=None,
48 multimodal=True,
49 const_efficiency_mode=False,
50 evidence_tolerance=0.5,
51 n_iter_before_update=100,
52 null_log_evidence=-1e90,
53 max_modes=100,
54 mode_tolerance=-1e90,
55 outputfiles_basename=None,
56 seed=-1,
57 context=0,
58 write_output=True,
59 log_zero=-1e100,
60 max_iter=0,
61 init_MPI=False,
62 dump_callback=None,
63 )
64 short_name = "pm"
65 hard_exit = True
66 sampling_seed_key = "seed"
68 def __init__(
69 self,
70 likelihood,
71 priors,
72 outdir="outdir",
73 label="label",
74 use_ratio=False,
75 plot=False,
76 exit_code=77,
77 skip_import_verification=False,
78 temporary_directory=True,
79 **kwargs
80 ):
81 super(Pymultinest, self).__init__(
82 likelihood=likelihood,
83 priors=priors,
84 outdir=outdir,
85 label=label,
86 use_ratio=use_ratio,
87 plot=plot,
88 skip_import_verification=skip_import_verification,
89 exit_code=exit_code,
90 temporary_directory=temporary_directory,
91 **kwargs
92 )
93 self._apply_multinest_boundaries()
95 def _translate_kwargs(self, kwargs):
96 kwargs = super()._translate_kwargs(kwargs)
97 if "n_live_points" not in kwargs:
98 for equiv in self.npoints_equiv_kwargs:
99 if equiv in kwargs:
100 kwargs["n_live_points"] = kwargs.pop(equiv)
102 def _verify_kwargs_against_default_kwargs(self):
103 """Check the kwargs"""
105 self.outputfiles_basename = self.kwargs.pop("outputfiles_basename", None)
107 # for PyMultiNest >=2.9 the n_params kwarg cannot be None
108 if self.kwargs["n_params"] is None:
109 self.kwargs["n_params"] = self.ndim
110 if self.kwargs["dump_callback"] is None:
111 self.kwargs["dump_callback"] = self._dump_callback
112 NestedSampler._verify_kwargs_against_default_kwargs(self)
114 def _dump_callback(self, *args, **kwargs):
115 if self.use_temporary_directory:
116 self._copy_temporary_directory_contents_to_proper_path()
117 self._calculate_and_save_sampling_time()
119 def _apply_multinest_boundaries(self):
120 if self.kwargs["wrapped_params"] is None:
121 self.kwargs["wrapped_params"] = list()
122 for param in self.search_parameter_keys:
123 if self.priors[param].boundary == "periodic":
124 self.kwargs["wrapped_params"].append(1)
125 else:
126 self.kwargs["wrapped_params"].append(0)
128 @signal_wrapper
129 def run_sampler(self):
130 import pymultinest
132 self._verify_kwargs_against_default_kwargs()
134 self._setup_run_directory()
135 self._check_and_load_sampling_time_file()
136 if not self.kwargs["resume"]:
137 self.total_sampling_time = 0.0
139 # Overwrite pymultinest's signal handling function
140 pm_run = importlib.import_module("pymultinest.run")
141 pm_run.interrupt_handler = self.write_current_state_and_exit
143 self.start_time = time.time()
144 out = pymultinest.solve(
145 LogLikelihood=self.log_likelihood,
146 Prior=self.prior_transform,
147 n_dims=self.ndim,
148 **self.kwargs
149 )
150 self._calculate_and_save_sampling_time()
152 self._clean_up_run_directory()
154 post_equal_weights = os.path.join(
155 self.outputfiles_basename, "post_equal_weights.dat"
156 )
157 post_equal_weights_data = np.loadtxt(post_equal_weights)
158 self.result.log_likelihood_evaluations = post_equal_weights_data[:, -1]
159 self.result.sampler_output = out
160 self.result.samples = post_equal_weights_data[:, :-1]
161 self.result.log_evidence = out["logZ"]
162 self.result.log_evidence_err = out["logZerr"]
163 self.calc_likelihood_count()
164 self.result.outputfiles_basename = self.outputfiles_basename
165 self.result.sampling_time = datetime.timedelta(seconds=self.total_sampling_time)
166 self.result.nested_samples = self._nested_samples
167 return self.result
169 @property
170 def _nested_samples(self):
171 """
172 Extract nested samples from the pymultinest files.
173 This requires combining the "dead" points from `ev.dat` and the "live"
174 points from `phys_live.points`.
175 The prior volume associated with the current live points is the simple
176 estimate of `remaining_prior_volume / N`.
177 """
178 import pandas as pd
180 dir_ = self.kwargs["outputfiles_basename"]
181 dead_points = np.genfromtxt(dir_ + "/ev.dat")
182 live_points = np.genfromtxt(dir_ + "/phys_live.points")
184 nlive = self.kwargs["n_live_points"]
185 final_log_prior_volume = -len(dead_points) / nlive - np.log(nlive)
186 live_points = np.insert(live_points, -1, final_log_prior_volume, axis=-1)
188 nested_samples = pd.DataFrame(
189 np.vstack([dead_points, live_points]).copy(),
190 columns=self.search_parameter_keys
191 + ["log_likelihood", "log_prior_volume", "mode"],
192 )
193 return nested_samples