variational_models.py 7.71 KB
Newer Older
Philipp Arras's avatar
Philipp Arras committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
# 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/>.
#
# Copyright(C) 2013-2021 Max-Planck-Society
#
# NIFTy is being developed at the Max-Planck-Institut fuer Astrophysik.

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

20
21
from ..domain_tuple import DomainTuple
from ..domains.unstructured_domain import UnstructuredDomain
Philipp Arras's avatar
Philipp Arras committed
22
23
24
25
26
27
28
29
from ..field import Field
from ..linearization import Linearization
from ..multi_field import MultiField
from ..operators.einsum import MultiLinearEinsum
from ..operators.energy_operators import EnergyOperator
from ..operators.linear_operator import LinearOperator
from ..operators.multifield2vector import Multifield2Vector
from ..operators.sandwich_operator import SandwichOperator
Philipp Frank's avatar
Philipp Frank committed
30
from ..operators.simple_linear_operators import FieldAdapter
Philipp Frank's avatar
cleanup    
Philipp Frank committed
31
from ..sugar import full, makeField, makeDomain, from_random, is_fieldlike
Philipp Frank's avatar
Philipp Frank committed
32
from ..minimization.energy_adapter import StochasticEnergyAdapter
Philipp Frank's avatar
cleanup    
Philipp Frank committed
33
34
35
from ..utilities import myassert


Philipp Frank's avatar
Philipp Frank committed
36
class MeanFieldVI:
Philipp Arras's avatar
Docs    
Philipp Arras committed
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
    """Collect the operators required for Gaussian meanfield variational
    inference.

    Parameters
    ----------
    position :
        FIXME
    hamiltonian :
        FIXME
    n_samples :
        FIXME
    mirror_samples :
        FIXME
    initial_sig :
        FIXME
    comm :
        FIXME
    nanisinf :
        FIXME
    """
    def __init__(self, position, hamiltonian, n_samples, mirror_samples,
Philipp Frank's avatar
Philipp Frank committed
58
                 initial_sig=1, comm=None, nanisinf=False):
Philipp Arras's avatar
Docs    
Philipp Arras committed
59
        Flat = Multifield2Vector(position.domain)
Philipp Frank's avatar
Philipp Frank committed
60
        self._std = FieldAdapter(Flat.target, 'std').absolute()
Philipp Frank's avatar
Philipp Frank committed
61
        latent = FieldAdapter(Flat.target,'latent')
Philipp Frank's avatar
Philipp Frank committed
62
63
64
65
66
        self._mean = FieldAdapter(Flat.target, 'mean')
        self._generator = Flat.adjoint(self._mean + self._std * latent)
        self._entropy = GaussianEntropy(self._std.target) @ self._std
        self._mean = Flat.adjoint @ self._mean
        self._std = Flat.adjoint @ self._std
Philipp Arras's avatar
Docs    
Philipp Arras committed
67
        pos = {'mean': Flat(position)}
Philipp Frank's avatar
Philipp Frank committed
68
        if is_fieldlike(initial_sig):
Philipp Frank's avatar
Philipp Frank committed
69
            pos['std'] = Flat(initial_sig)
Philipp Frank's avatar
Philipp Frank committed
70
        else:
Philipp Frank's avatar
Philipp Frank committed
71
            pos['std'] = full(Flat.target, initial_sig)
Philipp Frank's avatar
Philipp Frank committed
72
        pos = MultiField.from_dict(pos)
Philipp Frank's avatar
Philipp Frank committed
73
74
75
76
        op = hamiltonian(self._generator) + self._entropy
        self._KL = StochasticEnergyAdapter.make(pos, op, ['latent',], n_samples,
                                    mirror_samples, nanisinf=nanisinf, comm=comm)
        self._samdom = latent.domain
Philipp Frank's avatar
Philipp Frank committed
77
78

    @property
Philipp Frank's avatar
Philipp Frank committed
79
    def mean(self):
Philipp Arras's avatar
Philipp Arras committed
80
        return self._mean.force(self._KL.position)
Philipp Frank's avatar
Philipp Frank committed
81
82
83

    @property
    def std(self):
Philipp Arras's avatar
Philipp Arras committed
84
        return self._std.force(self._KL.position)
Philipp Frank's avatar
Philipp Frank committed
85
86
87

    @property
    def entropy(self):
Philipp Arras's avatar
Philipp Arras committed
88
        return self._entropy.force(self._KL.position)
Philipp Frank's avatar
Philipp Frank committed
89
90
91
92
93

    def draw_sample(self):
        _, op = self._generator.simplify_for_constant_input(
                from_random(self._samdom))
        return op(self._KL.position)
Philipp Frank's avatar
Philipp Frank committed
94
95
96
97

    def minimize(self, minimizer):
        self._KL, _ = minimizer(self._KL)

Philipp Arras's avatar
Docs    
Philipp Arras committed
98

Philipp Frank's avatar
Philipp Frank committed
99
class FullCovarianceVI:
Philipp Arras's avatar
Docs    
Philipp Arras committed
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
    """Collect the operators required for Gaussian full-covariance variational
    inference.

    Parameters
    ----------
    position :
        FIXME
    hamiltonian :
        FIXME
    n_samples :
        FIXME
    mirror_samples :
        FIXME
    initial_sig :
        FIXME
    comm :
        FIXME
    nanisinf :
        FIXME
    """
Philipp Frank's avatar
Philipp Frank committed
120
    def __init__(self, position, hamiltonian, n_samples, mirror_samples,
Philipp Frank's avatar
Philipp Frank committed
121
                initial_sig=1, comm=None, nanisinf=False):
Philipp Frank's avatar
Philipp Frank committed
122
123
124
        Flat = Multifield2Vector(position.domain)
        flat_domain = Flat.target[0]
        mat_space = DomainTuple.make((flat_domain,flat_domain))
Philipp Frank's avatar
Philipp Frank committed
125
        lat = FieldAdapter(Flat.target,'latent')
Philipp Frank's avatar
cleanup    
Philipp Frank committed
126
127
        LT = LowerTriangularInserter(mat_space)
        tri = FieldAdapter(LT.domain, 'cov')
Philipp Frank's avatar
Philipp Frank committed
128
        mean = FieldAdapter(flat_domain,'mean')
129
        cov = LT @ tri
Philipp Frank's avatar
Philipp Frank committed
130
131
132
133
134
135
        matmul_setup = lat.adjoint @ lat + cov.ducktape_left('co')
        MatMult = MultiLinearEinsum(matmul_setup.target,'ij,j->i',
                                    key_order=('co','latent'))

        self._generator = Flat.adjoint @ (mean + MatMult @ matmul_setup)

Philipp Frank's avatar
cleanup    
Philipp Frank committed
136
        diag_cov = (DiagonalSelector(cov.target) @ cov).absolute()
Philipp Frank's avatar
Philipp Frank committed
137
138
139
        self._entropy = GaussianEntropy(diag_cov.target) @ diag_cov
        diag_tri = np.diag(np.full(flat_domain.shape[0], initial_sig))
        pos = MultiField.from_dict(
Philipp Frank's avatar
cleanup    
Philipp Frank committed
140
141
                {'mean': Flat(position),
                 'cov': LT.adjoint(makeField(mat_space, diag_tri))})
Philipp Frank's avatar
Philipp Frank committed
142
143
144
145
146
147
148
149
        op = hamiltonian(self._generator) + self._entropy
        self._KL = StochasticEnergyAdapter.make(pos, op, ['latent',], n_samples,
                                    mirror_samples, nanisinf=nanisinf, comm=comm)
        self._mean = Flat.adjoint @ mean
        self._samdom = lat.domain

    @property
    def mean(self):
Philipp Arras's avatar
Philipp Arras committed
150
        return self._mean.force(self._KL.position)
Philipp Frank's avatar
Philipp Frank committed
151
152

    @property
Philipp Frank's avatar
Philipp Frank committed
153
    def entropy(self):
Philipp Arras's avatar
Philipp Arras committed
154
        return self._entropy.force(self._KL.position)
Philipp Frank's avatar
Philipp Frank committed
155
156
157
158
159

    def draw_sample(self):
        _, op = self._generator.simplify_for_constant_input(
                from_random(self._samdom))
        return op(self._KL.position)
Philipp Frank's avatar
Philipp Frank committed
160
161
162

    def minimize(self, minimizer):
        self._KL, _ = minimizer(self._KL)
163
164
165


class GaussianEntropy(EnergyOperator):
Philipp Arras's avatar
Docs    
Philipp Arras committed
166
167
    """Entropy of a Gaussian distribution given the diagonal of a triangular
    decomposition of the covariance.
Jakob Knollmüller's avatar
Jakob Knollmüller committed
168
169
170
171

    Parameters
    ----------
    domain: Domain
Philipp Arras's avatar
Philipp Arras committed
172
173
174
        The domain of the diagonal.
    """

175
176
177
178
179
    def __init__(self, domain):
        self._domain = domain

    def apply(self, x):
        self._check_input(x)
Philipp Arras's avatar
Philipp Arras committed
180
        res = -0.5*(2*np.pi*np.e*x**2).log().sum()
181
        if not isinstance(x, Linearization):
Jakob Knollmüller's avatar
tests    
Jakob Knollmüller committed
182
            return res
183
184
        if not x.want_metric:
            return res
Philipp Arras's avatar
Philipp Arras committed
185
186
        # FIXME not sure about metric
        return res.add_metric(SandwichOperator.make(res.jac))
187
188


Philipp Frank's avatar
cleanup    
Philipp Frank committed
189
class LowerTriangularInserter(LinearOperator):
Philipp Arras's avatar
Docs    
Philipp Arras committed
190
    """Insert the entries of a lower triangular matrix into a matrix.
Philipp Arras's avatar
Philipp Arras committed
191

Jakob Knollmüller's avatar
Jakob Knollmüller committed
192
193
194
    Parameters
    ----------
    target: Domain
Philipp Arras's avatar
Philipp Arras committed
195
196
197
        A two-dimensional domain with NxN entries.
    """

Philipp Frank's avatar
cleanup    
Philipp Frank committed
198
199
200
201
202
203
    def __init__(self, target):
        myassert(len(target.shape) == 2)
        myassert(target.shape[0] == target.shape[1])
        self._target = makeDomain(target)
        ndof = (target.shape[0]*(target.shape[0]+1))//2
        self._domain = makeDomain(UnstructuredDomain(ndof))
Philipp Arras's avatar
Philipp Arras committed
204
        self._indices = np.tril_indices(target.shape[0])
205
206
207
        self._capability = self.TIMES | self.ADJOINT_TIMES

    def apply(self, x, mode):
Philipp Arras's avatar
Philipp Arras committed
208
209
        self._check_input(x, mode)
        x = x.val
210
        if mode == self.TIMES:
Philipp Arras's avatar
Philipp Arras committed
211
212
213
214
215
216
            res = np.zeros(self._target.shape)
            res[self._indices] = x
        else:
            res = x[self._indices].reshape(self._domain.shape)
        return makeField(self._tgt(mode), res)

217
218

class DiagonalSelector(LinearOperator):
Philipp Arras's avatar
Philipp Arras committed
219
    """Extract the diagonal of a two-dimensional field.
Jakob Knollmüller's avatar
Jakob Knollmüller committed
220
221
222
223

    Parameters
    ----------
    domain: Domain
Philipp Frank's avatar
cleanup    
Philipp Frank committed
224
        The two-dimensional domain of the input field. Must be of shape NxN.
Philipp Arras's avatar
Philipp Arras committed
225
226
    """

Philipp Frank's avatar
cleanup    
Philipp Frank committed
227
228
229
230
231
    def __init__(self, domain):
        myassert(len(domain.shape) == 2)
        myassert(domain.shape[0] == domain.shape[1])
        self._domain = makeDomain(domain)
        self._target = makeDomain(UnstructuredDomain(domain.shape[0]))
232
233
        self._capability = self.TIMES | self.ADJOINT_TIMES

Philipp Arras's avatar
Philipp Arras committed
234
235
    def apply(self, x, mode):
        self._check_input(x, mode)
Philipp Frank's avatar
cleanup    
Philipp Frank committed
236
        return makeField(self._tgt(mode), np.diag(x.val))