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
    Parameters
    ----------
    mean : Field
        Mean of the Gaussian. Default is 0.
113
114
    inverse_covariance : LinearOperator
        Inverse 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

124
    def __init__(self, mean=None, inverse_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
127
        if inverse_covariance is not None and not isinstance(inverse_covariance, LinearOperator):
Philipp Arras's avatar
Philipp Arras committed
128
129
            raise TypeError

Martin Reinecke's avatar
Martin Reinecke committed
130
131
132
        self._domain = None
        if mean is not None:
            self._checkEquivalence(mean.domain)
133
134
        if inverse_covariance is not None:
            self._checkEquivalence(inverse_covariance.domain)
Martin Reinecke's avatar
Martin Reinecke committed
135
136
137
138
139
        if domain is not None:
            self._checkEquivalence(domain)
        if self._domain is None:
            raise ValueError("no domain given")
        self._mean = mean
140
        if inverse_covariance is None:
Martin Reinecke's avatar
Martin Reinecke committed
141
142
            self._op = SquaredNormOperator(self._domain).scale(0.5)
        else:
143
144
            self._op = QuadraticFormOperator(inverse_covariance)
        self._icov = None if inverse_covariance is None else inverse_covariance
Martin Reinecke's avatar
Martin Reinecke committed
145
146

    def _checkEquivalence(self, newdom):
Martin Reinecke's avatar
fix    
Martin Reinecke committed
147
        newdom = makeDomain(newdom)
Martin Reinecke's avatar
Martin Reinecke committed
148
        if self._domain is None:
Philipp Arras's avatar
Philipp Arras committed
149
            self._domain = newdom
Martin Reinecke's avatar
Martin Reinecke committed
150
        else:
Philipp Arras's avatar
Philipp Arras committed
151
            if self._domain != newdom:
Martin Reinecke's avatar
Martin Reinecke committed
152
153
154
                raise ValueError("domain mismatch")

    def apply(self, x):
155
        self._check_input(x)
Philipp Arras's avatar
Philipp Arras committed
156
        residual = x if self._mean is None else x - self._mean
Philipp Arras's avatar
Changes    
Philipp Arras committed
157
        res = self._op(residual).real
158
        if not isinstance(x, Linearization) or not x.want_metric:
Martin Reinecke's avatar
Martin Reinecke committed
159
160
161
162
163
164
            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
165
166
    """Computes likelihood Hamiltonians of expected count field constrained by
    Poissonian count data.
167

Philipp Arras's avatar
Philipp Arras committed
168
    Represents up to an f-independent term :math:`log(d!)`:
169

Philipp Arras's avatar
Philipp Arras committed
170
171
    .. math ::
        E(f) = -\\log \\text{Poisson}(d|f) = \\sum f - d^\\dagger \\log(f),
172

Philipp Arras's avatar
Philipp Arras committed
173
    where f is a :class:`Field` in data space with the expectation values for
Martin Reinecke's avatar
Martin Reinecke committed
174
    the counts.
Philipp Arras's avatar
Philipp Arras committed
175
176
177
178
179
180

    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
181
    """
Philipp Arras's avatar
Philipp Arras committed
182

183
    def __init__(self, d):
Philipp Arras's avatar
Philipp Arras committed
184
185
186
187
        if not isinstance(d, Field) or not np.issubdtype(d.dtype, np.integer):
            raise TypeError
        if np.any(d.local_data < 0):
            raise ValueError
188
189
        self._d = d
        self._domain = DomainTuple.make(d.domain)
Martin Reinecke's avatar
Martin Reinecke committed
190
191

    def apply(self, x):
192
        self._check_input(x)
Martin Reinecke's avatar
Martin Reinecke committed
193
        res = x.sum()
Martin Reinecke's avatar
Martin Reinecke committed
194
        tmp = res.val.local_data if isinstance(res, Linearization) else res
Martin Reinecke's avatar
Martin Reinecke committed
195
196
        # if we have no infinity here, we can continue with the calculation;
        # otherwise we know that the result must also be infinity
Martin Reinecke's avatar
Martin Reinecke committed
197
        if not np.isinf(tmp):
Martin Reinecke's avatar
Martin Reinecke committed
198
            res = res - x.log().vdot(self._d)
Martin Reinecke's avatar
Martin Reinecke committed
199
        if not isinstance(x, Linearization):
Martin Reinecke's avatar
Martin Reinecke committed
200
            return Field.scalar(res)
201
202
        if not x.want_metric:
            return res
Martin Reinecke's avatar
Martin Reinecke committed
203
204
205
        metric = SandwichOperator.make(x.jac, makeOp(1./x.val))
        return res.add_metric(metric)

206

207
class InverseGammaLikelihood(EnergyOperator):
Philipp Arras's avatar
Docs    
Philipp Arras committed
208
    """Computes the negative log-likelihood of the inverse gamma distribution.
209
210
211

    It negative log-pdf(x) is given by

Martin Reinecke's avatar
Martin Reinecke committed
212
213
214
215
216
217
218
    .. 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`.
219
220
221
222
223
224
225

    Parameters
    ----------
    beta : Field
        beta parameter of the inverse gamma distribution
    alpha : Scalar, Field, optional
        alpha parameter of the inverse gamma distribution
226
    """
Philipp Arras's avatar
Philipp Arras committed
227

228
229
    def __init__(self, beta, alpha=-0.5):
        if not isinstance(beta, Field):
Philipp Arras's avatar
Philipp Arras committed
230
            raise TypeError
231
232
        self._beta = beta
        if np.isscalar(alpha):
Martin Reinecke's avatar
Martin Reinecke committed
233
234
            alpha = Field.from_local_data(
                beta.domain, np.full(beta.local_data.shape, alpha))
235
236
237
238
        elif not isinstance(alpha, Field):
            raise TypeError
        self._alphap1 = alpha+1
        self._domain = DomainTuple.make(beta.domain)
239
240

    def apply(self, x):
241
        self._check_input(x)
242
        res = x.log().vdot(self._alphap1) + (1./x).vdot(self._beta)
243
244
        if not isinstance(x, Linearization):
            return Field.scalar(res)
245
246
        if not x.want_metric:
            return res
247
        metric = SandwichOperator.make(x.jac, makeOp(self._alphap1/(x.val**2)))
248
249
250
        return res.add_metric(metric)


Martin Reinecke's avatar
Martin Reinecke committed
251
class BernoulliEnergy(EnergyOperator):
Philipp Arras's avatar
Philipp Arras committed
252
    """Computes likelihood energy of expected event frequency constrained by
253
254
    event data.

Philipp Arras's avatar
Philipp Arras committed
255
256
257
258
259
260
261
    .. 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.

262
263
    Parameters
    ----------
Martin Reinecke's avatar
Martin Reinecke committed
264
    d : Field
Philipp Arras's avatar
Philipp Arras committed
265
        Data field with events (1) or non-events (0).
Martin Reinecke's avatar
Martin Reinecke committed
266
    """
Philipp Arras's avatar
Philipp Arras committed
267

268
    def __init__(self, d):
Philipp Arras's avatar
Philipp Arras committed
269
270
271
272
        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
273
        self._d = d
274
        self._domain = DomainTuple.make(d.domain)
Martin Reinecke's avatar
Martin Reinecke committed
275
276

    def apply(self, x):
277
        self._check_input(x)
Philipp Arras's avatar
Philipp Arras committed
278
        v = -(x.log().vdot(self._d) + (1. - x).log().vdot(1. - self._d))
Martin Reinecke's avatar
Martin Reinecke committed
279
        if not isinstance(x, Linearization):
Martin Reinecke's avatar
Martin Reinecke committed
280
            return Field.scalar(v)
281
282
        if not x.want_metric:
            return v
Philipp Arras's avatar
Philipp Arras committed
283
        met = makeOp(1./(x.val*(1. - x.val)))
Martin Reinecke's avatar
Martin Reinecke committed
284
285
286
287
        met = SandwichOperator.make(x.jac, met)
        return v.add_metric(met)


288
class StandardHamiltonian(EnergyOperator):
Philipp Arras's avatar
Philipp Arras committed
289
290
    """Computes an information Hamiltonian in its standard form, i.e. with the
    prior being a Gaussian with unit covariance.
291

Philipp Arras's avatar
Philipp Arras committed
292
    Let the likelihood energy be :math:`E_{lh}`. Then this operator computes:
293

Philipp Arras's avatar
Philipp Arras committed
294
295
    .. math ::
         H(f) = 0.5 f^\\dagger f + E_{lh}(f):
296

Martin Reinecke's avatar
Martin Reinecke committed
297
    Other field priors can be represented via transformations of a white
298
299
    Gaussian field into a field with the desired prior probability structure.

Martin Reinecke's avatar
Martin Reinecke committed
300
    By implementing prior information this way, the field prior is represented
301
302
303
    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
304
305
306
307
308
309
310
311
    The metric of this operator can be used as covariance for drawing Gaussian
    samples.

    Parameters
    ----------
    lh : EnergyOperator
        The likelihood energy.
    ic_samp : IterationController
312
        Tells an internal :class:`SamplingEnabler` which convergence criterion
Philipp Arras's avatar
Philipp Arras committed
313
314
315
316
317
318
        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
319
    `<https://arxiv.org/abs/1812.04403>`_
Martin Reinecke's avatar
Martin Reinecke committed
320
    """
Philipp Arras's avatar
Philipp Arras committed
321

Martin Reinecke's avatar
Martin Reinecke committed
322
323
324
325
    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
326
        self._domain = lh.domain
Martin Reinecke's avatar
Martin Reinecke committed
327
328

    def apply(self, x):
329
        self._check_input(x)
Philipp Arras's avatar
Philipp Arras committed
330
331
332
        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
333
        else:
334
            lhx, prx = self._lh(x), self._prior(x)
335
336
            mtr = SamplingEnabler(lhx.metric, prx.metric,
                                  self._ic_samp)
Philipp Arras's avatar
Philipp Arras committed
337
            return (lhx + prx).add_metric(mtr)
Martin Reinecke's avatar
Martin Reinecke committed
338

Philipp Arras's avatar
Philipp Arras committed
339
340
341
    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
342
        return 'StandardHamiltonian:\n' + utilities.indent(subs)
Philipp Arras's avatar
Philipp Arras committed
343

Martin Reinecke's avatar
Martin Reinecke committed
344

Martin Reinecke's avatar
Martin Reinecke committed
345
class AveragedEnergy(EnergyOperator):
Philipp Arras's avatar
Docs    
Philipp Arras committed
346
    """Averages an energy over samples.
Martin Reinecke's avatar
Martin Reinecke committed
347

348
349
350
    Parameters
    ----------
    h: Hamiltonian
Philipp Arras's avatar
Philipp Arras committed
351
       The energy to be averaged.
Martin Reinecke's avatar
Martin Reinecke committed
352
    res_samples : iterable of Fields
Torsten Ensslin's avatar
Torsten Ensslin committed
353
354
       Set of residual sample points to be added to mean field for
       approximate estimation of the KL.
355

Philipp Arras's avatar
Docs    
Philipp Arras committed
356
357
358
359
360
    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
361

Philipp Arras's avatar
Docs    
Philipp Arras committed
362
363
364
    - :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
365
    """
Martin Reinecke's avatar
Martin Reinecke committed
366
367
368

    def __init__(self, h, res_samples):
        self._h = h
Martin Reinecke's avatar
Martin Reinecke committed
369
        self._domain = h.domain
Martin Reinecke's avatar
Martin Reinecke committed
370
371
372
        self._res_samples = tuple(res_samples)

    def apply(self, x):
373
        self._check_input(x)
Philipp Arras's avatar
Philipp Arras committed
374
375
        mymap = map(lambda v: self._h(x + v), self._res_samples)
        return utilities.my_sum(mymap)*(1./len(self._res_samples))