extra.py 8.92 KB
Newer Older
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
15
#
16
# NIFTy is being developed at the Max-Planck-Institut fuer Astrophysik.
17
18

import numpy as np
19
from numpy.testing import assert_
Philipp Arras's avatar
Philipp Arras committed
20

21
from .domain_tuple import DomainTuple
Martin Reinecke's avatar
fix  
Martin Reinecke committed
22
23
from .field import Field
from .linearization import Linearization
24
from .multi_domain import MultiDomain
25
from .multi_field import MultiField
26
from .operators.linear_operator import LinearOperator
Martin Reinecke's avatar
fix  
Martin Reinecke committed
27
from .sugar import from_random
28

Martin Reinecke's avatar
Martin Reinecke committed
29
__all__ = ["consistency_check", "check_jacobian_consistency"]
30

Philipp Arras's avatar
Philipp Arras committed
31

Martin Reinecke's avatar
Martin Reinecke committed
32
33
34
35
36
37
38
39
def _assert_allclose(f1, f2, atol, rtol):
    if isinstance(f1, Field):
        return np.testing.assert_allclose(f1.local_data, f2.local_data,
                                          atol=atol, rtol=rtol)
    for key, val in f1.items():
        _assert_allclose(val, f2[key], atol=atol, rtol=rtol)


40
41
def _adjoint_implementation(op, domain_dtype, target_dtype, atol, rtol,
                            only_r_linear):
Martin Reinecke's avatar
Martin Reinecke committed
42
43
44
45
46
47
48
    needed_cap = op.TIMES | op.ADJOINT_TIMES
    if (op.capability & needed_cap) != needed_cap:
        return
    f1 = from_random("normal", op.domain, dtype=domain_dtype)
    f2 = from_random("normal", op.target, dtype=target_dtype)
    res1 = f1.vdot(op.adjoint_times(f2))
    res2 = op.times(f1).vdot(f2)
49
50
    if only_r_linear:
        res1, res2 = res1.real, res2.real
Martin Reinecke's avatar
Martin Reinecke committed
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
    np.testing.assert_allclose(res1, res2, atol=atol, rtol=rtol)


def _inverse_implementation(op, domain_dtype, target_dtype, atol, rtol):
    needed_cap = op.TIMES | op.INVERSE_TIMES
    if (op.capability & needed_cap) != needed_cap:
        return
    foo = from_random("normal", op.target, dtype=target_dtype)
    res = op(op.inverse_times(foo))
    _assert_allclose(res, foo, atol=atol, rtol=rtol)

    foo = from_random("normal", op.domain, dtype=domain_dtype)
    res = op.inverse_times(op(foo))
    _assert_allclose(res, foo, atol=atol, rtol=rtol)


67
68
69
70
def _full_implementation(op, domain_dtype, target_dtype, atol, rtol,
                         only_r_linear):
    _adjoint_implementation(op, domain_dtype, target_dtype, atol, rtol,
                            only_r_linear)
Martin Reinecke's avatar
Martin Reinecke committed
71
72
73
    _inverse_implementation(op, domain_dtype, target_dtype, atol, rtol)


74
def _check_linearity(op, domain_dtype, atol, rtol):
Martin Reinecke's avatar
Martin Reinecke committed
75
76
77
    needed_cap = op.TIMES
    if (op.capability & needed_cap) != needed_cap:
        return
78
79
    fld1 = from_random("normal", op.domain, dtype=domain_dtype)
    fld2 = from_random("normal", op.domain, dtype=domain_dtype)
Martin Reinecke's avatar
Martin Reinecke committed
80
    alpha = np.random.random()  # FIXME: this can break badly with MPI!
81
82
83
84
85
    val1 = op(alpha*fld1+fld2)
    val2 = alpha*op(fld1)+op(fld2)
    _assert_allclose(val1, val2, atol=atol, rtol=rtol)


86
87
88
89
90
91
92
93
94
95
96
97
def _actual_domain_check(op, domain_dtype=None, inp=None):
    needed_cap = op.TIMES
    if (op.capability & needed_cap) != needed_cap:
        return
    if domain_dtype is not None:
        inp = from_random("normal", op.domain, dtype=domain_dtype)
    elif inp is None:
        raise ValueError('Need to specify either dtype or inp')
    assert_(inp.domain is op.domain)
    assert_(op(inp).domain is op.target)


Philipp Arras's avatar
Philipp Arras committed
98
def _actual_domain_check_nonlinear(op, loc):
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
    assert isinstance(loc, (Field, MultiField))
    assert_(loc.domain is op.domain)
    lin = Linearization.make_var(loc, False)
    reslin = op(lin)
    assert_(lin.domain is op.domain)
    assert_(lin.target is op.domain)
    assert_(lin.val.domain is lin.domain)

    assert_(reslin.domain is op.domain)
    assert_(reslin.target is op.target)
    assert_(reslin.val.domain is reslin.target)

    assert_(reslin.target is op.target)
    assert_(reslin.jac.domain is reslin.domain)
    assert_(reslin.jac.target is reslin.target)
    _actual_domain_check(reslin.jac, inp=loc)
Philipp Arras's avatar
Philipp Arras committed
115
    _actual_domain_check(reslin.jac.adjoint, inp=reslin.jac(loc))
116
117


118
119
120
def _domain_check(op):
    for dd in [op.domain, op.target]:
        if not isinstance(dd, (DomainTuple, MultiDomain)):
Martin Reinecke's avatar
Martin Reinecke committed
121
122
123
            raise TypeError(
                'The domain and the target of an operator need to',
                'be instances of either DomainTuple or MultiDomain.')
124
125


Martin Reinecke's avatar
Martin Reinecke committed
126
def consistency_check(op, domain_dtype=np.float64, target_dtype=np.float64,
127
                      atol=0, rtol=1e-7, only_r_linear=False):
Reimar H Leike's avatar
Reimar H Leike committed
128
129
130
131
    """
    Checks an operator for algebraic consistency of its capabilities.

    Checks whether times(), adjoint_times(), inverse_times() and
Philipp Arras's avatar
Philipp Arras committed
132
    adjoint_inverse_times() (if in capability list) is implemented
Reimar H Leike's avatar
Reimar H Leike committed
133
    consistently. Additionally, it checks whether the operator is linear.
Philipp Arras's avatar
Philipp Arras committed
134
135
136
137
138

    Parameters
    ----------
    op : LinearOperator
        Operator which shall be checked.
Reimar H Leike's avatar
Reimar H Leike committed
139
    domain_dtype : dtype
Philipp Arras's avatar
Philipp Arras committed
140
141
        The data type of the random vectors in the operator's domain. Default
        is `np.float64`.
Reimar H Leike's avatar
Reimar H Leike committed
142
    target_dtype : dtype
Philipp Arras's avatar
Philipp Arras committed
143
144
145
        The data type of the random vectors in the operator's target. Default
        is `np.float64`.
    atol : float
Martin Reinecke's avatar
Martin Reinecke committed
146
147
        Absolute tolerance for the check. If rtol is specified,
        then satisfying any tolerance will let the check pass.
Reimar H Leike's avatar
Reimar H Leike committed
148
        Default: 0.
Philipp Arras's avatar
Philipp Arras committed
149
    rtol : float
Martin Reinecke's avatar
Martin Reinecke committed
150
151
        Relative tolerance for the check. If atol is specified,
        then satisfying any tolerance will let the check pass.
Reimar H Leike's avatar
Reimar H Leike committed
152
        Default: 0.
153
154
155
    only_r_linear: bool
        set to True if the operator is only R-linear, not C-linear.
        This will relax the adjointness test accordingly.
Philipp Arras's avatar
Philipp Arras committed
156
    """
157
158
    if not isinstance(op, LinearOperator):
        raise TypeError('This test tests only linear operators.')
159
    _domain_check(op)
160
    _actual_domain_check(op, domain_dtype)
Philipp Arras's avatar
Philipp Arras committed
161
162
    _actual_domain_check(op.adjoint, target_dtype)
    _actual_domain_check(op.inverse, target_dtype)
163
    _actual_domain_check(op.adjoint.inverse, domain_dtype)
164
    _check_linearity(op, domain_dtype, atol, rtol)
Martin Reinecke's avatar
Martin Reinecke committed
165
166
167
    _check_linearity(op.adjoint, target_dtype, atol, rtol)
    _check_linearity(op.inverse, target_dtype, atol, rtol)
    _check_linearity(op.adjoint.inverse, domain_dtype, atol, rtol)
168
169
170
171
172
173
    _full_implementation(op, domain_dtype, target_dtype, atol, rtol,
                         only_r_linear)
    _full_implementation(op.adjoint, target_dtype, domain_dtype, atol, rtol,
                         only_r_linear)
    _full_implementation(op.inverse, target_dtype, domain_dtype, atol, rtol,
                         only_r_linear)
Martin Reinecke's avatar
Martin Reinecke committed
174
    _full_implementation(op.adjoint.inverse, domain_dtype, target_dtype, atol,
175
                         rtol, only_r_linear)
Martin Reinecke's avatar
Martin Reinecke committed
176
177


Martin Reinecke's avatar
Martin Reinecke committed
178
def _get_acceptable_location(op, loc, lin):
Martin Reinecke's avatar
Martin Reinecke committed
179
    if not np.isfinite(lin.val.sum()):
Martin Reinecke's avatar
Martin Reinecke committed
180
181
182
183
        raise ValueError('Initial value must be finite')
    dir = from_random("normal", loc.domain)
    dirder = lin.jac(dir)
    if dirder.norm() == 0:
Martin Reinecke's avatar
Martin Reinecke committed
184
        dir = dir * (lin.val.norm()*1e-5)
Martin Reinecke's avatar
Martin Reinecke committed
185
    else:
Martin Reinecke's avatar
Martin Reinecke committed
186
        dir = dir * (lin.val.norm()*1e-5/dirder.norm())
Martin Reinecke's avatar
Martin Reinecke committed
187
188
189
190
    # Find a step length that leads to a "reasonable" location
    for i in range(50):
        try:
            loc2 = loc+dir
191
            lin2 = op(Linearization.make_var(loc2, lin.want_metric))
Martin Reinecke's avatar
Martin Reinecke committed
192
193
194
195
196
197
198
199
200
            if np.isfinite(lin2.val.sum()) and abs(lin2.val.sum()) < 1e20:
                break
        except FloatingPointError:
            pass
        dir = dir*0.5
    else:
        raise ValueError("could not find a reasonable initial step")
    return loc2, lin2

Martin Reinecke's avatar
Martin Reinecke committed
201

Martin Reinecke's avatar
Martin Reinecke committed
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
def check_jacobian_consistency(op, loc, tol=1e-8, ntries=100):
    """
    Checks the Jacobian of an operator against its finite difference
    approximation.

    Computes the Jacobian with finite differences and compares it to the
    implemented Jacobian.

    Parameters
    ----------
    op : Operator
        Operator which shall be checked.
    loc : Field or MultiField
        An Field or MultiField instance which has the same domain
        as op. The location at which the gradient is checked
    tol : float
        Tolerance for the check.
    """
220
    _domain_check(op)
221
    _actual_domain_check_nonlinear(op, loc)
Martin Reinecke's avatar
Martin Reinecke committed
222
    for _ in range(ntries):
223
        lin = op(Linearization.make_var(loc))
Martin Reinecke's avatar
Martin Reinecke committed
224
        loc2, lin2 = _get_acceptable_location(op, loc, lin)
Martin Reinecke's avatar
Martin Reinecke committed
225
        dir = loc2-loc
Martin Reinecke's avatar
Martin Reinecke committed
226
227
228
229
        locnext = loc2
        dirnorm = dir.norm()
        for i in range(50):
            locmid = loc + 0.5*dir
230
            linmid = op(Linearization.make_var(locmid))
Martin Reinecke's avatar
Martin Reinecke committed
231
232
            dirder = linmid.jac(dir)
            numgrad = (lin2.val-lin.val)
Martin Reinecke's avatar
Martin Reinecke committed
233
            xtol = tol * dirder.norm() / np.sqrt(dirder.size)
Martin Reinecke's avatar
Martin Reinecke committed
234
            if (abs(numgrad-dirder) <= xtol).all():
Martin Reinecke's avatar
Martin Reinecke committed
235
236
237
                break
            dir = dir*0.5
            dirnorm *= 0.5
Martin Reinecke's avatar
Martin Reinecke committed
238
            loc2, lin2 = locmid, linmid
Martin Reinecke's avatar
Martin Reinecke committed
239
240
241
        else:
            raise ValueError("gradient and value seem inconsistent")
        loc = locnext