forked from joshspeagle/dynesty
-
Notifications
You must be signed in to change notification settings - Fork 0
/
priors.py
325 lines (251 loc) · 8.79 KB
/
priors.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
# Module containg various objects to be used as priors.
# These return the ln-prior-probability
import numpy as np
import scipy.stats
__all__ = ["prior_transform",
"Prior", "TopHat", "Normal", "ClippedNormal",
"LogNormal", "LogUniform", "Beta"]
def prior_transform(self, unit_coords, priors, prior_args=[]):
"""An example of one way to use the `Prior` objects below to go from unit
cube to parameter space, for nested sampling. This takes and returns a
list instead of an array, to accomodate possible vector parameters. Thus
one will need something like ``theta_array=np.concatenate(*theta)``
:param unit_coords:
Coordinates on the unit prior hyper-cube. Iterable.
:param priors:
A list of `Prior` objects, iterable of same length as `unit_coords`.
:param prior_args: (optional)
A list of dictionaries of prior function keyword arguments.
:returns theta:
A list of parameter values corresponding to the given coordinates on
the prior unit hypercube.
"""
theta = []
for i, (u, p) in enumerate(zip(unit_coords, priors)):
func = p.unit_transform
try:
kwargs = prior_args[i]
except(IndexError):
kwargs = {}
theta.append(func(u, **kwargs))
return theta
class Prior(object):
"""Encapsulate the priors in an object. Each prior should have a
distribution name and optional parameters specifying scale and location
(e.g. min/max or mean/sigma). These can be aliased. When called, the
argument should be a variable and it should return the ln-prior-probability
of that value.
Should be able to sample from the prior, and to get the gradient of the
prior at any variable value. Methods should also be avilable to give a
useful plotting range and, if there are bounds, to return them.
"""
def __init__(self, parnames=[], name='', **kwargs):
"""Constructor.
:param parnames:
A list of names of the parameters, used to alias the intrinsic
parameter names. This way different instances of the same Prior
can have different parameter names in case they are being fit for.
"""
if len(parnames) == 0:
parnames = self.prior_params
assert len(parnames) == len(self.prior_params)
self.alias = dict(zip(self.prior_params, parnames))
self.params = {}
self.name = name
self.update(**kwargs)
def update(self, **kwargs):
"""Update `params` values using alias.
"""
for k in self.prior_params:
try:
self.params[k] = kwargs[self.alias[k]]
except(KeyError):
pass
def __len__(self):
"""The length is set by the maximum size of any of the prior_params.
Note that the prior params must therefore be scalar of same length as
the maximum size of any of the parameters. This is not checked.
"""
return max([np.size(self.params.get(k, 1)) for k in self.prior_params])
def __call__(self, x, **kwargs):
"""Compute the value of the probability desnity function at x and
return the ln of that.
:param x:
Value of the parameter, scalar or iterable of same length as the
Prior object.
:param kwargs: optional
All extra keyword arguments are sued to update the `prior_params`.
:returns lnp:
The natural log of the prior probability at x, scalar or ndarray of
same length as the prior object.
"""
if len(kwargs) > 0:
self.update(**kwargs)
p = self.distribution.pdf(x, *self.args,
loc=self.loc, scale=self.scale)
return np.log(p)
def sample(self, nsample=None, **kwargs):
"""Draw a sample from the prior distribution.
:param nsample: (optional)
Unused
"""
if len(kwargs) > 0:
self.update(**kwargs)
return self.distribution.rvs(*self.args, size=len(self),
loc=self.loc, scale=self.scale)
def unit_transform(self, x, **kwargs):
"""Go from a value of the CDF (between 0 and 1) to the corresponding
parameter value.
:param x:
A scalar or vector of same length as the Prior with values between
zero and one corresponding to the value of the CDF.
:returns theta:
The parameter value corresponding to the value of the CDF given by
`x`.
"""
if len(kwargs) > 0:
self.update(**kwargs)
return self.distribution.ppf(x, *self.args,
loc=self.loc, scale=self.scale)
def inverse_unit_transform(self, x, **kwargs):
"""Go from the parameter value to the unit coordinate using the cdf.
"""
if len(kwargs) > 0:
self.update(**kwargs)
return self.distribution.cdf(x, *self.args,
loc=self.loc, scale=self.scale)
def gradient(self, theta):
raise(NotImplementedError)
@property
def loc(self):
"""This should be overridden.
"""
return 0
@property
def scale(self):
"""This should be overridden.
"""
return 1
@property
def args(self):
return []
@property
def range(self):
raise(NotImplementedError)
@property
def bounds(self):
raise(NotImplementedError)
def serialize(self):
raise(NotImplementedError)
class TopHat(Prior):
prior_params = ['mini', 'maxi']
distribution = scipy.stats.uniform
@property
def scale(self):
return self.params['maxi'] - self.params['mini']
@property
def loc(self):
return self.params['mini']
@property
def range(self):
return (self.params['mini'], self.params['maxi'])
def bounds(self, **kwargs):
if len(kwargs) > 0:
self.update(**kwargs)
return self.range
class Normal(Prior):
"""A simple gaussian prior.
"""
prior_params = ['mean', 'sigma']
distribution = scipy.stats.norm
@property
def scale(self):
return self.params['sigma']
@property
def loc(self):
return self.params['mean']
@property
def range(self):
nsig = 4
return (self.params['mean'] - nsig * self.params['sigma'],
self.params['mean'] + self.params['sigma'])
def bounds(self, **kwargs):
# if len(kwargs) > 0:
# self.update(**kwargs)
return (-np.inf, np.inf)
class ClippedNormal(Prior):
"""A Gaussian prior clipped to some range.
"""
prior_params = ['mean', 'sigma', 'mini', 'maxi']
distribution = scipy.stats.truncnorm
@property
def scale(self):
return self.params['sigma']
@property
def loc(self):
return self.params['mean']
@property
def range(self):
return (self.params['mini'], self.params['maxi'])
@property
def args(self):
a = (self.params['mini'] - self.params['mean']) / self.params['sigma']
b = (self.params['maxi'] - self.params['mean']) / self.params['sigma']
return [a, b]
def bounds(self, **kwargs):
if len(kwargs) > 0:
self.update(**kwargs)
return self.range
class LogUniform(Prior):
"""Like log-normal, but the distribution of ln of the variable is
distributed uniformly instead of normally.
"""
prior_params = ['mini', 'maxi']
distribution = scipy.stats.reciprocal
@property
def args(self):
a = self.params['mini']
b = self.params['maxi']
return [a, b]
@property
def range(self):
return (self.params['mini'], self.params['maxi'])
def bounds(self, **kwargs):
if len(kwargs) > 0:
self.update(**kwargs)
return self.range
class Beta(Prior):
"""A Beta distribution.
"""
prior_params = ['mini', 'maxi', 'alpha', 'beta']
distribution = scipy.stats.beta
@property
def scale(self):
return self.params.get('maxi', 1) - self.params.get('mini', 0)
@property
def loc(self):
return self.params.get('mini', 0)
@property
def args(self):
a = self.params['alpha']
b = self.params['beta']
return [a, b]
@property
def range(self):
return (self.params.get('mini', 0), self.params.get('maxi', 1))
def bounds(self, **kwargs):
if len(kwargs) > 0:
self.update(**kwargs)
return self.range
class LogNormal(Prior):
prior_params = ['mode', 'sigma']
distribution = scipy.stats.lognorm
@property
def args(self):
pass
@property
def scale(self):
pass
@property
def loc(self):
pass