critical_power_energy.py 4.86 KB
Newer Older
1

2
from nifty.energies.energy import Energy
3
4
from nifty.operators.smoothness_operator import SmoothnessOperator
from nifty.library.critical_filter import CriticalPowerCurvature
5
from nifty.energies.memoization import memo
Jakob Knollmueller's avatar
Jakob Knollmueller committed
6

7
from nifty.sugar import generate_posterior_sample
8
from nifty import Field, exp
9

10

11
class CriticalPowerEnergy(Energy):
Jakob Knollmueller's avatar
Jakob Knollmueller committed
12
    """The Energy of the power spectrum according to the critical filter.
13

14
15
16
17
    It describes the energy of the logarithmic amplitudes of the power spectrum
    of a Gaussian random field under reconstruction uncertainty with smoothness
    and inverse gamma prior. It is used to infer the correlation structure of a
    correlated signal.
18
19
20

    Parameters
    ----------
Jakob Knollmueller's avatar
Jakob Knollmueller committed
21
22
23
24
25
26
27
28
29
    position : Field,
        The current position of this energy.
    m : Field,
        The map whichs power spectrum has to be inferred
    D : EndomorphicOperator,
        The curvature of the Gaussian encoding the posterior covariance.
        If not specified, the map is assumed to be no reconstruction.
        default : None
    alpha : float
30
31
        The spectral prior of the inverse gamma distribution. 1.0 corresponds
        to non-informative.
Jakob Knollmueller's avatar
Jakob Knollmueller committed
32
33
        default : 1.0
    q : float
34
35
36
37
38
        The cutoff parameter of the inverse gamma distribution. 0.0 corresponds
        to non-informative.
        default : 0.0
    smoothness_prior : float
        Controls the strength of the smoothness prior
Jakob Knollmueller's avatar
Jakob Knollmueller committed
39
40
41
42
        default : 0.0
    logarithmic : boolean
        Whether smoothness acts on linear or logarithmic scale.
    samples : integer
43
44
        Number of samples used for the estimation of the uncertainty
        corrections.
Jakob Knollmueller's avatar
Jakob Knollmueller committed
45
46
47
48
49
        default : 3
    w : Field
        The contribution from the map with or without uncertainty. It is used
        to pass on the result of the costly sampling during the minimization.
        default : None
50
51
52
    inverter : ConjugateGradient
        The inversion strategy to invert the curvature and to generate samples.
        default : None
53
54
    """

55
56
    # ---Overwritten properties and methods---

57
58
59
    def __init__(self, position, m, D=None, alpha=1.0, q=0.,
                 smoothness_prior=0., logarithmic=True, samples=3, w=None):
        super(CriticalPowerEnergy, self).__init__(position=position)
60
        self.m = m
61
62
        self.D = D
        self.samples = samples
63
64
        self.alpha = Field(self.position.domain, val=alpha)
        self.q = Field(self.position.domain, val=q)
65
        self.T = SmoothnessOperator(domain=self.position.domain[0],
66
                                    strength=smoothness_prior,
Jakob Knollmueller's avatar
Jakob Knollmueller committed
67
                                    logarithmic=logarithmic)
68
        self.rho = self.position.domain[0].rho
69
        self._w = w if w is not None else None
70

71
72
    # ---Mandatory properties and methods---

73
    def at(self, position):
74
75
        return self.__class__(position, self.m, D=self.D, alpha=self.alpha,
                              q=self.q, smoothness_prior=self.smoothness_prior,
76
                              logarithmic=self.logarithmic,
77
                              w=self.w, samples=self.samples)
78
79
80

    @property
    def value(self):
81
        energy = self._theta.sum()
82
83
        energy += self.position.vdot(self._rho_prime, bare=True)
        energy += 0.5 * self.position.vdot(self._Tt)
84
85
86
87
        return energy.real

    @property
    def gradient(self):
88
        gradient = -self._theta.weight(-1)
89
        gradient += (self._rho_prime).weight(-1)
90
        gradient += self._Tt
91
        gradient.val = gradient.val.real
92
93
94
95
        return gradient

    @property
    def curvature(self):
96
97
        curvature = CriticalPowerCurvature(theta=self._theta.weight(-1),
                                           T=self.T)
98
99
        return curvature

100
101
102
103
104
105
106
107
108
109
    # ---Added properties and methods---

    @property
    def logarithmic(self):
        return self.T.logarithmic

    @property
    def smoothness_prior(self):
        return self.T.strength

110
111
112
113
114
115
116
117
118
    @property
    def w(self):
        if self._w is None:
            w = Field(domain=self.position.domain, val=0., dtype=self.m.dtype)
            if self.D is not None:
                for i in range(self.samples):
                    posterior_sample = generate_posterior_sample(
                                                            self.m, self.D)
                    projected_sample = posterior_sample.power_analyze(
Martin Reinecke's avatar
Martin Reinecke committed
119
                     binbounds=self.position.domain[0].binbounds)
120
121
122
123
                    w += (projected_sample) * self.rho
                w /= float(self.samples)
            else:
                w = self.m.power_analyze(
Martin Reinecke's avatar
Martin Reinecke committed
124
                     binbounds=self.position.domain[0].binbounds)
125
126
127
                w *= self.rho
            self._w = w
        return self._w
128

129
130
131
    @property
    @memo
    def _theta(self):
132
        return exp(-self.position) * (self.q + self.w / 2.)
133
134
135
136
137
138
139
140
141
142

    @property
    @memo
    def _rho_prime(self):
        return self.alpha - 1. + self.rho / 2.

    @property
    @memo
    def _Tt(self):
        return self.T(self.position)