energy_operators.py 12.3 KB
Newer Older
Martin Reinecke's avatar
Martin Reinecke committed
1
2
3
4
5
6
7
8
9
10
11
12
13
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.
#
14
# Copyright(C) 2013-2019 Max-Planck-Society
Martin Reinecke's avatar
Martin Reinecke committed
15
#
16
# NIFTy is being developed at the Max-Planck-Institut fuer Astrophysik.
Martin Reinecke's avatar
Martin Reinecke committed
17

Philipp Arras's avatar
Philipp Arras committed
18
19
import numpy as np

Philipp Arras's avatar
Philipp Arras committed
20
from .. import utilities
Martin Reinecke's avatar
Martin Reinecke committed
21
from ..domain_tuple import DomainTuple
Philipp Arras's avatar
Philipp Arras committed
22
from ..field import Field
23
from ..multi_field import MultiField
Philipp Arras's avatar
Philipp Arras committed
24
from ..linearization import Linearization
Philipp Arras's avatar
Philipp Arras committed
25
26
from ..sugar import makeDomain, makeOp
from .linear_operator import LinearOperator
Martin Reinecke's avatar
Martin Reinecke committed
27
from .operator import Operator
Martin Reinecke's avatar
fix    
Martin Reinecke committed
28
from .sampling_enabler import SamplingEnabler
Philipp Arras's avatar
Philipp Arras committed
29
from .sandwich_operator import SandwichOperator
Martin Reinecke's avatar
Martin Reinecke committed
30
from .simple_linear_operators import VdotOperator
Martin Reinecke's avatar
Martin Reinecke committed
31
32
33


class EnergyOperator(Operator):
Philipp Arras's avatar
Philipp Arras committed
34
    """Operator which has a scalar domain as target domain.
35

Martin Reinecke's avatar
Martin Reinecke committed
36
    It is intended as an objective function for field inference.
37

Philipp Arras's avatar
Philipp Arras committed
38
39
40
    Examples
    --------
     - Information Hamiltonian, i.e. negative-log-probabilities.
Martin Reinecke's avatar
Martin Reinecke committed
41
     - Gibbs free energy, i.e. an averaged Hamiltonian, aka Kullback-Leibler
Philipp Arras's avatar
Philipp Arras committed
42
       divergence.
43
    """
Martin Reinecke's avatar
Martin Reinecke committed
44
45
46
47
    _target = DomainTuple.scalar_domain()


class SquaredNormOperator(EnergyOperator):
Philipp Arras's avatar
Philipp Arras committed
48
    """Computes the L2-norm of the output of an operator.
49

Philipp Arras's avatar
Philipp Arras committed
50
51
52
    Parameters
    ----------
    domain : Domain, DomainTuple or tuple of Domain
53
        Domain of the operator in which the L2-norm shall be computed.
Martin Reinecke's avatar
Martin Reinecke committed
54
    """
Philipp Arras's avatar
Philipp Arras committed
55

Martin Reinecke's avatar
Martin Reinecke committed
56
57
58
59
    def __init__(self, domain):
        self._domain = domain

    def apply(self, x):
60
        self._check_input(x)
Martin Reinecke's avatar
Martin Reinecke committed
61
        if isinstance(x, Linearization):
Martin Reinecke's avatar
Martin Reinecke committed
62
            val = Field.scalar(x.val.vdot(x.val))
Martin Reinecke's avatar
Martin Reinecke committed
63
            jac = VdotOperator(2*x.val)(x.jac)
64
            return x.new(val, jac)
Martin Reinecke's avatar
Martin Reinecke committed
65
        return Field.scalar(x.vdot(x))
Martin Reinecke's avatar
Martin Reinecke committed
66
67
68


class QuadraticFormOperator(EnergyOperator):
Philipp Arras's avatar
Philipp Arras committed
69
    """Computes the L2-norm of a Field or MultiField with respect to a
70
    specific kernel given by `endo`.
Philipp Arras's avatar
Philipp Arras committed
71
72
73

    .. math ::
        E(f) = \\frac12 f^\\dagger \\text{endo}(f)
74
75
76

    Parameters
    ----------
Philipp Arras's avatar
Philipp Arras committed
77
    endo : EndomorphicOperator
78
         Kernel of the quadratic form
Martin Reinecke's avatar
Martin Reinecke committed
79
    """
Philipp Arras's avatar
Philipp Arras committed
80
81

    def __init__(self, endo):
Martin Reinecke's avatar
Martin Reinecke committed
82
        from .endomorphic_operator import EndomorphicOperator
Philipp Arras's avatar
Philipp Arras committed
83
        if not isinstance(endo, EndomorphicOperator):
Martin Reinecke's avatar
Martin Reinecke committed
84
            raise TypeError("op must be an EndomorphicOperator")
Philipp Arras's avatar
Philipp Arras committed
85
86
        self._op = endo
        self._domain = endo.domain
Martin Reinecke's avatar
Martin Reinecke committed
87
88

    def apply(self, x):
89
        self._check_input(x)
Martin Reinecke's avatar
Martin Reinecke committed
90
        if isinstance(x, Linearization):
Martin Reinecke's avatar
Martin Reinecke committed
91
92
            t1 = self._op(x.val)
            jac = VdotOperator(t1)(x.jac)
Martin Reinecke's avatar
Martin Reinecke committed
93
            val = Field.scalar(0.5*x.val.vdot(t1))
94
            return x.new(val, jac)
Martin Reinecke's avatar
Martin Reinecke committed
95
        return Field.scalar(0.5*x.vdot(self._op(x)))
Martin Reinecke's avatar
Martin Reinecke committed
96
97
98


class GaussianEnergy(EnergyOperator):
Philipp Arras's avatar
Docs    
Philipp Arras committed
99
    """Computes a negative-log Gaussian.
100

Philipp Arras's avatar
Philipp Arras committed
101
    Represents up to constants in :math:`m`:
Martin Reinecke's avatar
Martin Reinecke committed
102

Philipp Arras's avatar
Philipp Arras committed
103
104
    .. math ::
        E(f) = - \\log G(f-m, D) = 0.5 (f-m)^\\dagger D^{-1} (f-m),
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
105

Philipp Arras's avatar
Philipp Arras committed
106
107
    an information energy for a Gaussian distribution with mean m and
    covariance D.
108

Philipp Arras's avatar
Philipp Arras committed
109
110
111
112
113
114
    Parameters
    ----------
    mean : Field
        Mean of the Gaussian. Default is 0.
    covariance : LinearOperator
        Covariance of the Gaussian. Default is the identity operator.
Philipp Arras's avatar
Fixup    
Philipp Arras committed
115
    domain : Domain, DomainTuple, tuple of Domain or MultiDomain
Philipp Arras's avatar
Philipp Arras committed
116
117
118
119
120
121
        Operator domain. By default it is inferred from `mean` or
        `covariance` if specified

    Note
    ----
    At least one of the arguments has to be provided.
Martin Reinecke's avatar
Martin Reinecke committed
122
    """
Martin Reinecke's avatar
Martin Reinecke committed
123

Martin Reinecke's avatar
Martin Reinecke committed
124
    def __init__(self, mean=None, covariance=None, domain=None):
Martin Reinecke's avatar
Martin Reinecke committed
125
126
        if mean is not None and not isinstance(mean, (Field, MultiField)):
            raise TypeError
Philipp Arras's avatar
Philipp Arras committed
127
128
129
130
        if covariance is not None and not isinstance(covariance,
                                                     LinearOperator):
            raise TypeError

Martin Reinecke's avatar
Martin Reinecke committed
131
132
133
134
135
136
137
138
139
140
        self._domain = None
        if mean is not None:
            self._checkEquivalence(mean.domain)
        if covariance is not None:
            self._checkEquivalence(covariance.domain)
        if domain is not None:
            self._checkEquivalence(domain)
        if self._domain is None:
            raise ValueError("no domain given")
        self._mean = mean
Martin Reinecke's avatar
Martin Reinecke committed
141
142
143
144
        if covariance is None:
            self._op = SquaredNormOperator(self._domain).scale(0.5)
        else:
            self._op = QuadraticFormOperator(covariance.inverse)
Martin Reinecke's avatar
Martin Reinecke committed
145
146
147
        self._icov = None if covariance is None else covariance.inverse

    def _checkEquivalence(self, newdom):
Martin Reinecke's avatar
fix    
Martin Reinecke committed
148
        newdom = makeDomain(newdom)
Martin Reinecke's avatar
Martin Reinecke committed
149
        if self._domain is None:
Philipp Arras's avatar
Philipp Arras committed
150
            self._domain = newdom
Martin Reinecke's avatar
Martin Reinecke committed
151
        else:
Philipp Arras's avatar
Philipp Arras committed
152
            if self._domain != newdom:
Martin Reinecke's avatar
Martin Reinecke committed
153
154
155
                raise ValueError("domain mismatch")

    def apply(self, x):
156
        self._check_input(x)
Philipp Arras's avatar
Philipp Arras committed
157
        residual = x if self._mean is None else x - self._mean
Philipp Arras's avatar
Changes    
Philipp Arras committed
158
        res = self._op(residual).real
159
        if not isinstance(x, Linearization) or not x.want_metric:
Martin Reinecke's avatar
Martin Reinecke committed
160
161
162
163
164
165
            return res
        metric = SandwichOperator.make(x.jac, self._icov)
        return res.add_metric(metric)


class PoissonianEnergy(EnergyOperator):
Philipp Arras's avatar
Docs    
Philipp Arras committed
166
167
    """Computes likelihood Hamiltonians of expected count field constrained by
    Poissonian count data.
168

Philipp Arras's avatar
Philipp Arras committed
169
    Represents up to an f-independent term :math:`log(d!)`:
170

Philipp Arras's avatar
Philipp Arras committed
171
172
    .. math ::
        E(f) = -\\log \\text{Poisson}(d|f) = \\sum f - d^\\dagger \\log(f),
173

Philipp Arras's avatar
Philipp Arras committed
174
    where f is a :class:`Field` in data space with the expectation values for
Martin Reinecke's avatar
Martin Reinecke committed
175
    the counts.
Philipp Arras's avatar
Philipp Arras committed
176
177
178
179
180
181

    Parameters
    ----------
    d : Field
        Data field with counts. Needs to have integer dtype and all field
        values need to be non-negative.
Martin Reinecke's avatar
Martin Reinecke committed
182
    """
Philipp Arras's avatar
Philipp Arras committed
183

184
    def __init__(self, d):
Philipp Arras's avatar
Philipp Arras committed
185
186
187
188
        if not isinstance(d, Field) or not np.issubdtype(d.dtype, np.integer):
            raise TypeError
        if np.any(d.local_data < 0):
            raise ValueError
189
190
        self._d = d
        self._domain = DomainTuple.make(d.domain)
Martin Reinecke's avatar
Martin Reinecke committed
191
192

    def apply(self, x):
193
        self._check_input(x)
Martin Reinecke's avatar
Martin Reinecke committed
194
195
196
197
198
199
200
        res = x.sum()
        tmp = (res.val.local_data if isinstance(res, Linearization)
            else res.local_data)
        # if we have no infinity here, we can continue with the calculation;
        # otherwise we know that the result must also be infinity
        if not np.any(np.isinf(tmp)):
            res = res - x.log().vdot(self._d)
Martin Reinecke's avatar
Martin Reinecke committed
201
        if not isinstance(x, Linearization):
Martin Reinecke's avatar
Martin Reinecke committed
202
            return Field.scalar(res)
203
204
        if not x.want_metric:
            return res
Martin Reinecke's avatar
Martin Reinecke committed
205
206
207
        metric = SandwichOperator.make(x.jac, makeOp(1./x.val))
        return res.add_metric(metric)

208

209
class InverseGammaLikelihood(EnergyOperator):
Philipp Arras's avatar
Docs    
Philipp Arras committed
210
    """Computes the negative log-likelihood of the inverse gamma distribution.
211
212
213

    It negative log-pdf(x) is given by

Martin Reinecke's avatar
Martin Reinecke committed
214
215
216
217
218
219
220
    .. math ::

        \\sum_i (\\alpha_i+1)*\\ln(x_i) + \\beta_i/x_i

    This is the likelihood for the variance :math:`x=S_k` given data
    :math:`\\beta = 0.5 |s_k|^2` where the Field :math:`s` is known to have
    the covariance :math:`S_k`.
221
222
223
224
225
226
227

    Parameters
    ----------
    beta : Field
        beta parameter of the inverse gamma distribution
    alpha : Scalar, Field, optional
        alpha parameter of the inverse gamma distribution
228
    """
Philipp Arras's avatar
Philipp Arras committed
229

230
231
    def __init__(self, beta, alpha=-0.5):
        if not isinstance(beta, Field):
Philipp Arras's avatar
Philipp Arras committed
232
            raise TypeError
233
234
        self._beta = beta
        if np.isscalar(alpha):
Martin Reinecke's avatar
Martin Reinecke committed
235
236
            alpha = Field.from_local_data(
                beta.domain, np.full(beta.local_data.shape, alpha))
237
238
239
240
        elif not isinstance(alpha, Field):
            raise TypeError
        self._alphap1 = alpha+1
        self._domain = DomainTuple.make(beta.domain)
241
242

    def apply(self, x):
243
        self._check_input(x)
244
        res = x.log().vdot(self._alphap1) + (1./x).vdot(self._beta)
245
246
        if not isinstance(x, Linearization):
            return Field.scalar(res)
247
248
        if not x.want_metric:
            return res
249
        metric = SandwichOperator.make(x.jac, makeOp(self._alphap1/(x.val**2)))
250
251
252
        return res.add_metric(metric)


Martin Reinecke's avatar
Martin Reinecke committed
253
class BernoulliEnergy(EnergyOperator):
Philipp Arras's avatar
Philipp Arras committed
254
    """Computes likelihood energy of expected event frequency constrained by
255
256
    event data.

Philipp Arras's avatar
Philipp Arras committed
257
258
259
260
261
262
263
    .. math ::
        E(f) = -\\log \\text{Bernoulli}(d|f)
             = -d^\\dagger \\log f  - (1-d)^\\dagger \\log(1-f),

    where f is a field defined on `d.domain` with the expected
    frequencies of events.

264
265
    Parameters
    ----------
Martin Reinecke's avatar
Martin Reinecke committed
266
    d : Field
Philipp Arras's avatar
Philipp Arras committed
267
        Data field with events (1) or non-events (0).
Martin Reinecke's avatar
Martin Reinecke committed
268
    """
Philipp Arras's avatar
Philipp Arras committed
269

270
    def __init__(self, d):
Philipp Arras's avatar
Philipp Arras committed
271
272
273
274
        if not isinstance(d, Field) or not np.issubdtype(d.dtype, np.integer):
            raise TypeError
        if not np.all(np.logical_or(d.local_data == 0, d.local_data == 1)):
            raise ValueError
Martin Reinecke's avatar
Martin Reinecke committed
275
        self._d = d
276
        self._domain = DomainTuple.make(d.domain)
Martin Reinecke's avatar
Martin Reinecke committed
277
278

    def apply(self, x):
279
        self._check_input(x)
Philipp Arras's avatar
Philipp Arras committed
280
        v = -(x.log().vdot(self._d) + (1. - x).log().vdot(1. - self._d))
Martin Reinecke's avatar
Martin Reinecke committed
281
        if not isinstance(x, Linearization):
Martin Reinecke's avatar
Martin Reinecke committed
282
            return Field.scalar(v)
283
284
        if not x.want_metric:
            return v
Philipp Arras's avatar
Philipp Arras committed
285
        met = makeOp(1./(x.val*(1. - x.val)))
Martin Reinecke's avatar
Martin Reinecke committed
286
287
288
289
        met = SandwichOperator.make(x.jac, met)
        return v.add_metric(met)


290
class StandardHamiltonian(EnergyOperator):
Philipp Arras's avatar
Philipp Arras committed
291
292
    """Computes an information Hamiltonian in its standard form, i.e. with the
    prior being a Gaussian with unit covariance.
293

Philipp Arras's avatar
Philipp Arras committed
294
    Let the likelihood energy be :math:`E_{lh}`. Then this operator computes:
295

Philipp Arras's avatar
Philipp Arras committed
296
297
    .. math ::
         H(f) = 0.5 f^\\dagger f + E_{lh}(f):
298

Martin Reinecke's avatar
Martin Reinecke committed
299
    Other field priors can be represented via transformations of a white
300
301
    Gaussian field into a field with the desired prior probability structure.

Martin Reinecke's avatar
Martin Reinecke committed
302
    By implementing prior information this way, the field prior is represented
303
304
305
    by a generative model, from which NIFTy can draw samples and infer a field
    using the Maximum a Posteriori (MAP) or the Variational Bayes (VB) method.

Philipp Arras's avatar
Philipp Arras committed
306
307
308
309
310
311
312
313
    The metric of this operator can be used as covariance for drawing Gaussian
    samples.

    Parameters
    ----------
    lh : EnergyOperator
        The likelihood energy.
    ic_samp : IterationController
314
        Tells an internal :class:`SamplingEnabler` which convergence criterion
Philipp Arras's avatar
Philipp Arras committed
315
316
317
318
319
320
        to use to draw Gaussian samples.

    See also
    --------
    `Encoding prior knowledge in the structure of the likelihood`,
    Jakob Knollmüller, Torsten A. Ensslin,
Martin Reinecke's avatar
Martin Reinecke committed
321
    `<https://arxiv.org/abs/1812.04403>`_
Martin Reinecke's avatar
Martin Reinecke committed
322
    """
Philipp Arras's avatar
Philipp Arras committed
323

Martin Reinecke's avatar
Martin Reinecke committed
324
325
326
327
    def __init__(self, lh, ic_samp=None):
        self._lh = lh
        self._prior = GaussianEnergy(domain=lh.domain)
        self._ic_samp = ic_samp
Martin Reinecke's avatar
Martin Reinecke committed
328
        self._domain = lh.domain
Martin Reinecke's avatar
Martin Reinecke committed
329
330

    def apply(self, x):
331
        self._check_input(x)
Philipp Arras's avatar
Philipp Arras committed
332
333
334
        if (self._ic_samp is None or not isinstance(x, Linearization)
                or not x.want_metric):
            return self._lh(x) + self._prior(x)
Martin Reinecke's avatar
Martin Reinecke committed
335
        else:
336
            lhx, prx = self._lh(x), self._prior(x)
337
338
            mtr = SamplingEnabler(lhx.metric, prx.metric,
                                  self._ic_samp)
Philipp Arras's avatar
Philipp Arras committed
339
            return (lhx + prx).add_metric(mtr)
Martin Reinecke's avatar
Martin Reinecke committed
340

Philipp Arras's avatar
Philipp Arras committed
341
342
343
    def __repr__(self):
        subs = 'Likelihood:\n{}'.format(utilities.indent(self._lh.__repr__()))
        subs += '\nPrior: Quadratic{}'.format(self._lh.domain.keys())
Martin Reinecke's avatar
Martin Reinecke committed
344
        return 'StandardHamiltonian:\n' + utilities.indent(subs)
Philipp Arras's avatar
Philipp Arras committed
345

Martin Reinecke's avatar
Martin Reinecke committed
346

Martin Reinecke's avatar
Martin Reinecke committed
347
class AveragedEnergy(EnergyOperator):
Philipp Arras's avatar
Docs    
Philipp Arras committed
348
    """Averages an energy over samples.
Martin Reinecke's avatar
Martin Reinecke committed
349

350
351
352
    Parameters
    ----------
    h: Hamiltonian
Philipp Arras's avatar
Philipp Arras committed
353
       The energy to be averaged.
Martin Reinecke's avatar
Martin Reinecke committed
354
    res_samples : iterable of Fields
Torsten Ensslin's avatar
Torsten Ensslin committed
355
356
       Set of residual sample points to be added to mean field for
       approximate estimation of the KL.
357

Philipp Arras's avatar
Docs    
Philipp Arras committed
358
359
360
361
362
    Notes
    -----
    - Having symmetrized residual samples, with both :math:`v_i` and
      :math:`-v_i` being present, ensures that the distribution mean is
      exactly represented.
Torsten Ensslin's avatar
Fix te    
Torsten Ensslin committed
363

Philipp Arras's avatar
Docs    
Philipp Arras committed
364
365
366
    - :class:`AveragedEnergy(h)` approximates
      :math:`\\left< H(f) \\right>_{G(f-m,D)}` if the residuals :math:`f-m`
      are drawn from a Gaussian distribution with covariance :math:`D`.
Martin Reinecke's avatar
Martin Reinecke committed
367
    """
Martin Reinecke's avatar
Martin Reinecke committed
368
369
370

    def __init__(self, h, res_samples):
        self._h = h
Martin Reinecke's avatar
Martin Reinecke committed
371
        self._domain = h.domain
Martin Reinecke's avatar
Martin Reinecke committed
372
373
374
        self._res_samples = tuple(res_samples)

    def apply(self, x):
375
        self._check_input(x)
Philipp Arras's avatar
Philipp Arras committed
376
377
        mymap = map(lambda v: self._h(x + v), self._res_samples)
        return utilities.my_sum(mymap)*(1./len(self._res_samples))