energy_and_model_tests.py 4.4 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
# 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) 2013-2018 Max-Planck-Society
#
# NIFTy is being developed at the Max-Planck-Institut fuer Astrophysik
# and financially supported by the Studienstiftung des deutschen Volkes.

19
20
from __future__ import (absolute_import, division, print_function)
from builtins import *
21
import numpy as np
Martin Reinecke's avatar
Martin Reinecke committed
22
from ..sugar import from_random
Martin Reinecke's avatar
tweak    
Martin Reinecke committed
23
24
from ..minimization.energy import Energy
from ..models.model import Model
25

Martin Reinecke's avatar
Martin Reinecke committed
26
__all__ = ["check_value_gradient_consistency",
Martin Reinecke's avatar
Martin Reinecke committed
27
           "check_value_gradient_metric_consistency"]
28

Philipp Arras's avatar
Philipp Arras committed
29

30
31
32
33
34
def _get_acceptable_model(M):
    val = M.value
    if not np.isfinite(val.sum()):
        raise ValueError('Initial Model value must be finite')
    dir = from_random("normal", M.position.domain)
35
    dirder = M.jacobian(dir)
36
    dir = dir * val * (1e-5/dirder.norm())
Philipp Arras's avatar
Philipp Arras committed
37
    # Find a step length that leads to a "reasonable" Model
38
39
40
41
42
43
44
    for i in range(50):
        try:
            M2 = M.at(M.position+dir)
            if np.isfinite(M2.value.sum()) and abs(M2.value.sum()) < 1e20:
                break
        except FloatingPointError:
            pass
45
        dir = dir*0.5
46
47
48
49
    else:
        raise ValueError("could not find a reasonable initial step")
    return M2

50

Martin Reinecke's avatar
Martin Reinecke committed
51
def _get_acceptable_energy(E):
52
53
    val = E.value
    if not np.isfinite(val):
54
        raise ValueError('Initial Energy must be finite')
55
    dir = from_random("normal", E.position.domain)
56
    dirder = E.gradient.vdot(dir)
57
    dir = dir * (np.abs(val)/np.abs(dirder)*1e-5)
Philipp Arras's avatar
Philipp Arras committed
58
    # Find a step length that leads to a "reasonable" energy
Martin Reinecke's avatar
Martin Reinecke committed
59
60
61
62
63
64
65
    for i in range(50):
        try:
            E2 = E.at(E.position+dir)
            if np.isfinite(E2.value) and abs(E2.value) < 1e20:
                break
        except FloatingPointError:
            pass
66
        dir = dir*0.5
Martin Reinecke's avatar
Martin Reinecke committed
67
68
69
70
71
    else:
        raise ValueError("could not find a reasonable initial step")
    return E2


Martin Reinecke's avatar
Martin Reinecke committed
72
def check_value_gradient_consistency(E, tol=1e-8, ntries=100):
73
    for _ in range(ntries):
74
75
76
77
        if isinstance(E, Energy):
            E2 = _get_acceptable_energy(E)
        else:
            E2 = _get_acceptable_model(E)
78
        val = E.value
Martin Reinecke's avatar
Martin Reinecke committed
79
        dir = E2.position - E.position
Martin Reinecke's avatar
tweak    
Martin Reinecke committed
80
        Enext = E2
Martin Reinecke's avatar
Martin Reinecke committed
81
        dirnorm = dir.norm()
82
        for i in range(50):
Martin Reinecke's avatar
Martin Reinecke committed
83
            Emid = E.at(E.position + 0.5*dir)
84
85
86
            if isinstance(E, Energy):
                dirder = Emid.gradient.vdot(dir)/dirnorm
            else:
87
                dirder = Emid.jacobian(dir)/dirnorm
88
            numgrad = (E2.value-val)/dirnorm
89
            if isinstance(E, Model):
90
                xtol = tol * dirder.norm() / np.sqrt(dirder.size)
91
                if (abs(numgrad-dirder) < xtol).all():
92
93
94
                    break
            else:
                xtol = tol*Emid.gradient_norm
95
                if abs(numgrad-dirder) < xtol:
96
                    break
97
            dir = dir*0.5
Martin Reinecke's avatar
Martin Reinecke committed
98
            dirnorm *= 0.5
Martin Reinecke's avatar
Martin Reinecke committed
99
            E2 = Emid
100
        else:
Martin Reinecke's avatar
Martin Reinecke committed
101
            raise ValueError("gradient and value seem inconsistent")
Martin Reinecke's avatar
tweak    
Martin Reinecke committed
102
        E = Enext
Martin Reinecke's avatar
Martin Reinecke committed
103
104


Martin Reinecke's avatar
Martin Reinecke committed
105
def check_value_gradient_metric_consistency(E, tol=1e-8, ntries=100):
106
    if isinstance(E, Model):
Martin Reinecke's avatar
Martin Reinecke committed
107
        raise ValueError('Models have no metric, thus it cannot be tested.')
Martin Reinecke's avatar
Martin Reinecke committed
108
109
    for _ in range(ntries):
        E2 = _get_acceptable_energy(E)
110
        val = E.value
Martin Reinecke's avatar
Martin Reinecke committed
111
        dir = E2.position - E.position
Martin Reinecke's avatar
tweak    
Martin Reinecke committed
112
        Enext = E2
Martin Reinecke's avatar
Martin Reinecke committed
113
        dirnorm = dir.norm()
114
        for i in range(50):
115
116
            Emid = E.at(E.position + 0.5*dir)
            dirder = Emid.gradient.vdot(dir)/dirnorm
Martin Reinecke's avatar
Martin Reinecke committed
117
            dgrad = Emid.metric(dir)/dirnorm
Martin Reinecke's avatar
Martin Reinecke committed
118
119
120
            xtol = tol*Emid.gradient_norm
            if abs((E2.value-val)/dirnorm - dirder) < xtol and \
               (abs((E2.gradient-E.gradient)/dirnorm-dgrad) < xtol).all():
121
                break
122
            dir = dir*0.5
Martin Reinecke's avatar
Martin Reinecke committed
123
            dirnorm *= 0.5
124
            E2 = Emid
125
        else:
Martin Reinecke's avatar
Martin Reinecke committed
126
            raise ValueError("gradient, value and metric seem inconsistent")
Martin Reinecke's avatar
tweak    
Martin Reinecke committed
127
        E = Enext