sugar.py 17.9 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
# 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/>.
Theo Steininger's avatar
Theo Steininger committed
13
#
14
# Copyright(C) 2013-2019 Max-Planck-Society
Theo Steininger's avatar
Theo Steininger committed
15
#
16
# NIFTy is being developed at the Max-Planck-Institut fuer Astrophysik.
17

18
import sys
19
from time import time
20

21
import numpy as np
22

23
from .logger import logger
Martin Reinecke's avatar
stage 1    
Martin Reinecke committed
24
from . import utilities
25
from .domain_tuple import DomainTuple
Martin Reinecke's avatar
Martin Reinecke committed
26
from .domains.power_space import PowerSpace
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
27
from .field import Field
Martin Reinecke's avatar
Martin Reinecke committed
28
29
from .multi_domain import MultiDomain
from .multi_field import MultiField
Philipp Arras's avatar
Philipp Arras committed
30
from .operators.block_diagonal_operator import BlockDiagonalOperator
Martin Reinecke's avatar
Martin Reinecke committed
31
from .operators.diagonal_operator import DiagonalOperator
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
32
from .operators.distributors import PowerDistributor
33
from .operators.operator import Operator
34
from .operators.scaling_operator import ScalingOperator
Lukas Platz's avatar
Lukas Platz committed
35
from .plot import Plot
36
37
from . import pointwise

38

Martin Reinecke's avatar
step 1    
Martin Reinecke committed
39
40
__all__ = ['PS_field', 'power_analyze', 'create_power_operator',
           'create_harmonic_smoothing_operator', 'from_random',
Martin Reinecke's avatar
Martin Reinecke committed
41
           'full', 'makeField',
Martin Reinecke's avatar
Martin Reinecke committed
42
           'makeDomain', 'get_signal_variance', 'makeOp', 'domain_union',
Philipp Arras's avatar
Philipp Arras committed
43
           'get_default_codomain', 'single_plot', 'exec_time',
44
           'calculate_position'] + list(pointwise.ptw_dict.keys())
45

46

47
def PS_field(pspace, func):
Martin Reinecke's avatar
Martin Reinecke committed
48
49
50
51
52
53
54
55
    """Convenience function sampling a power spectrum

    Parameters
    ----------
    pspace : PowerSpace
        space at whose `k_lengths` the power spectrum function is evaluated
    func : function taking and returning a numpy.ndarray(float)
        the power spectrum function
Martin Reinecke's avatar
Martin Reinecke committed
56

Martin Reinecke's avatar
Martin Reinecke committed
57
58
    Returns
    -------
Martin Reinecke's avatar
Martin Reinecke committed
59
60
    Field
        A field defined on (pspace,) containing the computed function values
Martin Reinecke's avatar
Martin Reinecke committed
61
    """
Martin Reinecke's avatar
Martin Reinecke committed
62
63
    if not isinstance(pspace, PowerSpace):
        raise TypeError
Martin Reinecke's avatar
stage 1    
Martin Reinecke committed
64
    data = func(pspace.k_lengths)
65
    return Field(DomainTuple.make(pspace), data)
Martin Reinecke's avatar
Martin Reinecke committed
66

Martin Reinecke's avatar
Martin Reinecke committed
67

68
69
70
71
def get_signal_variance(spec, space):
    """
    Computes how much a field with a given power spectrum will vary in space

72
    This is a small helper function that computes the expected variance
73
74
75
76
77
78
79
80
    of a harmonically transformed sample of this power spectrum.

    Parameters
    ---------
    spec: method
        a method that takes one k-value and returns the power spectrum at that
        location
    space: PowerSpace or any harmonic Domain
Martin Reinecke's avatar
Martin Reinecke committed
81
82
83
84
        If this function is given a harmonic domain, it creates the naturally
        binned PowerSpace to that domain.
        The field, for which the signal variance is then computed, is assumed
        to have this PowerSpace as naturally binned PowerSpace
85
86
87
88
    """
    if space.harmonic:
        space = PowerSpace(space)
    if not isinstance(space, PowerSpace):
Martin Reinecke's avatar
Martin Reinecke committed
89
90
        raise ValueError(
            "space must be either a harmonic space or Power space.")
91
92
93
    field = PS_field(space, spec)
    dist = PowerDistributor(space.harmonic_partner, space)
    k_field = dist(field)
Martin Reinecke's avatar
Martin Reinecke committed
94
    return k_field.weight(2).s_sum()
95

96

97
98
def _single_power_analyze(field, idx, binbounds):
    power_domain = PowerSpace(field.domain[idx], binbounds)
Martin Reinecke's avatar
Martin Reinecke committed
99
100
    pd = PowerDistributor(field.domain, power_domain, idx)
    return pd.adjoint_times(field.weight(1)).weight(-1)  # divides by bin size
101
102


Martin Reinecke's avatar
Martin Reinecke committed
103
104
# MR FIXME: this function is not well suited for analyzing more than one
# subdomain at once, because it allows only one set of binbounds.
105
106
def power_analyze(field, spaces=None, binbounds=None,
                  keep_phase_information=False):
107
    """Computes the power spectrum for a subspace of `field`.
108
109

    Creates a PowerSpace for the space addressed by `spaces` with the given
110
    binning and computes the power spectrum as a :class:`Field` over this
111
    PowerSpace. This can only be done if the subspace to  be analyzed is a
Martin Reinecke's avatar
Martin Reinecke committed
112
113
    harmonic space. The resulting field has the same units as the square of the
    initial field.
114
115
116

    Parameters
    ----------
Philipp Arras's avatar
Fixups    
Philipp Arras committed
117
    field : Field
118
        The field to be analyzed
Martin Reinecke's avatar
Martin Reinecke committed
119
120
121
    spaces : None or int or tuple of int, optional
        The indices of subdomains for which the power spectrum shall be
        computed.
Martin Reinecke's avatar
Martin Reinecke committed
122
        If None, all subdomains will be converted.
123
        (default : None).
Martin Reinecke's avatar
Martin Reinecke committed
124
    binbounds : None or array-like, optional
125
        Inner bounds of the bins (default : None).
Martin Reinecke's avatar
Martin Reinecke committed
126
127
        if binbounds is None : bins are inferred.
    keep_phase_information : bool, optional
128
        If False, return a real-valued result containing the power spectrum
129
        of `field`.
130
        If True, return a complex-valued result whose real component
131
132
133
        contains the power spectrum computed from the real part of `field`,
        and whose imaginary component contains the power
        spectrum computed from the imaginary part of `field`.
134
135
136
137
138
139
        The absolute value of this result should be identical to the output
        of power_analyze with keep_phase_information=False.
        (default : False).

    Returns
    -------
Philipp Arras's avatar
Fixups    
Philipp Arras committed
140
    Field
141
        The output object. Its domain is a PowerSpace and it contains
Martin Reinecke's avatar
Martin Reinecke committed
142
        the power spectrum of `field`.
143
144
145
146
    """

    for sp in field.domain:
        if not sp.harmonic and not isinstance(sp, PowerSpace):
Martin Reinecke's avatar
Martin Reinecke committed
147
148
            logger.warning("WARNING: Field has a space in `domain` which is "
                           "neither harmonic nor a PowerSpace.")
149

150
    spaces = utilities.parse_spaces(spaces, len(field.domain))
151
152
153
154

    if len(spaces) == 0:
        raise ValueError("No space for analysis specified.")

Martin Reinecke's avatar
Martin Reinecke committed
155
    field_real = not utilities.iscomplextype(field.dtype)
156
157
158
    if (not field_real) and keep_phase_information:
        raise ValueError("cannot keep phase from real-valued input Field")

159
160
161
    if keep_phase_information:
        parts = [field.real*field.real, field.imag*field.imag]
    else:
162
163
164
165
        if field_real:
            parts = [field**2]
        else:
            parts = [field.real*field.real + field.imag*field.imag]
166
167

    for space_index in spaces:
Martin Reinecke's avatar
Martin Reinecke committed
168
        parts = [_single_power_analyze(part, space_index, binbounds)
169
170
171
172
173
                 for part in parts]

    return parts[0] + 1j*parts[1] if keep_phase_information else parts[0]


Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
174
def _create_power_field(domain, power_spectrum):
Philipp Arras's avatar
Philipp Arras committed
175
    if not callable(power_spectrum):  # we have a Field defined on a PowerSpace
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
176
177
178
179
180
181
182
        if not isinstance(power_spectrum, Field):
            raise TypeError("Field object expected")
        if len(power_spectrum.domain) != 1:
            raise ValueError("exactly one domain required")
        if not isinstance(power_spectrum.domain[0], PowerSpace):
            raise TypeError("PowerSpace required")
        power_domain = power_spectrum.domain[0]
183
        fp = power_spectrum
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
184
185
    else:
        power_domain = PowerSpace(domain)
186
        fp = PS_field(power_domain, power_spectrum)
187

Martin Reinecke's avatar
Martin Reinecke committed
188
    return PowerDistributor(domain, power_domain)(fp)
189

190

191
def create_power_operator(domain, power_spectrum, space=None):
192
    """Creates a diagonal operator with the given power spectrum.
193

Philipp Arras's avatar
Philipp Arras committed
194
    Constructs a diagonal operator that is defined on the specified domain.
195

196
197
    Parameters
    ----------
Martin Reinecke's avatar
Martin Reinecke committed
198
    domain : Domain, tuple of Domain or DomainTuple
Philipp Arras's avatar
Philipp Arras committed
199
        Domain on which the power operator shall be defined.
Martin Reinecke's avatar
Martin Reinecke committed
200
201
    power_spectrum : callable or Field
        An object that contains the power spectrum as a function of k.
Martin Reinecke's avatar
Martin Reinecke committed
202
    space : int
Martin Reinecke's avatar
Martin Reinecke committed
203
        the domain index on which the power operator will work
Theo Steininger's avatar
Theo Steininger committed
204

205
206
    Returns
    -------
Martin Reinecke's avatar
Martin Reinecke committed
207
208
    DiagonalOperator
        An operator that implements the given power spectrum.
209
    """
Martin Reinecke's avatar
Martin Reinecke committed
210
    domain = DomainTuple.make(domain)
Martin Reinecke's avatar
Martin Reinecke committed
211
    space = utilities.infer_space(domain, space)
Martin Reinecke's avatar
Martin Reinecke committed
212
213
    field = _create_power_field(domain[space], power_spectrum)
    return DiagonalOperator(field, domain, space)
214

215

216
def create_harmonic_smoothing_operator(domain, space, sigma):
Martin Reinecke's avatar
Martin Reinecke committed
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
    """Creates an operator which smoothes a subspace of a harmonic domain.

    Parameters
    ----------
    domain: DomainTuple
        The total domain and target of the operator
    space : int
        the index of the subspace on which the operator acts.
        This must be a harmonic space
    sigma : float
        The sigma of the Gaussian smoothing kernel

    Returns
    -------
    DiagonalOperator
        The requested smoothing operator
    """
234
235
236
    kfunc = domain[space].get_fft_smoothing_kernel_function(sigma)
    return DiagonalOperator(kfunc(domain[space].get_k_length_array()), domain,
                            space)
Martin Reinecke's avatar
step 1    
Martin Reinecke committed
237
238
239


def full(domain, val):
Martin Reinecke's avatar
Martin Reinecke committed
240
241
242
243
244
245
246
247
248
249
250
    """Convenience function creating Fields/MultiFields with uniform values.

    Parameters
    ----------
    domain : Domainoid
        the intended domain of the output field
    val : scalar value
        the uniform value to be placed into all entries of the result

    Returns
    -------
Martin Reinecke's avatar
Martin Reinecke committed
251
252
    Field or MultiField
        The newly created uniform field
Martin Reinecke's avatar
Martin Reinecke committed
253
    """
Martin Reinecke's avatar
step 1    
Martin Reinecke committed
254
255
256
257
258
259
    if isinstance(domain, (dict, MultiDomain)):
        return MultiField.full(domain, val)
    return Field.full(domain, val)


def from_random(random_type, domain, dtype=np.float64, **kwargs):
Martin Reinecke's avatar
Martin Reinecke committed
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
    """Convenience function creating Fields/MultiFields with random values.

    Parameters
    ----------
    random_type : 'pm1', 'normal', or 'uniform'
            The random distribution to use.
    domain : Domainoid
        the intended domain of the output field
    dtype : type
        data type of the output field (e.g. numpy.float64)
    **kwargs : additional parameters for the random distribution
        ('mean' and 'std' for 'normal', 'low' and 'high' for 'uniform')

    Returns
    -------
Martin Reinecke's avatar
Martin Reinecke committed
275
276
    Field or MultiField
        The newly created random field
277
278
279
280
281
282
283

    Notes
    -----
    The ordering of the keys within a multi-domain WILL change the resulting
    :class:`MultiField`. Note, since keys in a :class:`MultiDomain` are sorted,
    renaming keys may also impacts the ordering of the generated random numbers
    within the returned :class:`MultiField`.
Martin Reinecke's avatar
Martin Reinecke committed
284
    """
Martin Reinecke's avatar
step 1    
Martin Reinecke committed
285
286
287
288
289
    if isinstance(domain, (dict, MultiDomain)):
        return MultiField.from_random(random_type, domain, dtype, **kwargs)
    return Field.from_random(random_type, domain, dtype, **kwargs)


Martin Reinecke's avatar
Martin Reinecke committed
290
def makeField(domain, arr):
Martin Reinecke's avatar
Martin Reinecke committed
291
292
293
294
295
296
297
298
299
300
301
302
    """Convenience function creating Fields/MultiFields from Numpy arrays or
    dicts of Numpy arrays.

    Parameters
    ----------
    domain : Domainoid
        the intended domain of the output field
    arr : Numpy array if `domain` corresponds to a `DomainTuple`,
          dictionary of Numpy arrays if `domain` corresponds to a `MultiDomain`

    Returns
    -------
Martin Reinecke's avatar
Martin Reinecke committed
303
304
    Field or MultiField
        The newly created random field
Martin Reinecke's avatar
Martin Reinecke committed
305
    """
Martin Reinecke's avatar
step 1    
Martin Reinecke committed
306
    if isinstance(domain, (dict, MultiDomain)):
Martin Reinecke's avatar
Martin Reinecke committed
307
308
        return MultiField.from_raw(domain, arr)
    return Field.from_raw(domain, arr)
Martin Reinecke's avatar
step 1    
Martin Reinecke committed
309
310
311


def makeDomain(domain):
Martin Reinecke's avatar
Martin Reinecke committed
312
313
314
315
    """Convenience function creating DomainTuples/MultiDomains Domainoids.

    Parameters
    ----------
Martin Reinecke's avatar
Martin Reinecke committed
316
    domain : Domainoid (can be DomainTuple, MultiDomain, dict, Domain or list of Domains)
Martin Reinecke's avatar
Martin Reinecke committed
317
318
319
320
        the description of the requested (multi-)domain

    Returns
    -------
Martin Reinecke's avatar
Martin Reinecke committed
321
322
    DomainTuple or MultiDomain
        The newly created domain object
Martin Reinecke's avatar
Martin Reinecke committed
323
    """
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
324
    if isinstance(domain, (MultiDomain, dict)):
Martin Reinecke's avatar
step 1    
Martin Reinecke committed
325
326
        return MultiDomain.make(domain)
    return DomainTuple.make(domain)
327
328


329
def makeOp(input, dom=None):
Martin Reinecke's avatar
Martin Reinecke committed
330
331
332
333
334
335
    """Converts a Field or MultiField to a diagonal operator.

    Parameters
    ----------
    input : None, Field or MultiField
        - if None, None is returned.
336
337
        - if Field on scalar-domain, a ScalingOperator with the coefficient
            given by the Field is returned.
Martin Reinecke's avatar
Martin Reinecke committed
338
339
340
341
342
        - if Field, a DiagonalOperator with the coefficients given by this
            Field is returned.
        - if MultiField, a BlockDiagonalOperator with entries given by this
            MultiField is returned.

343
344
345
    dom : DomainTuple or MultiDomain
        if `input` is a scalar, this is used as the operator's domain

Martin Reinecke's avatar
Martin Reinecke committed
346
347
348
349
    Notes
    -----
    No volume factors are applied.
    """
Martin Reinecke's avatar
Martin Reinecke committed
350
351
    if input is None:
        return None
352
    if np.isscalar(input):
Rouven Lemmerz's avatar
Typo    
Rouven Lemmerz committed
353
        if not isinstance(dom, (DomainTuple, MultiDomain)):
354
            raise TypeError("need proper `dom` argument")
355
        return ScalingOperator(dom, input)
356
357
358
    if dom is not None:
        if not dom == input.domain:
            raise ValueError("domain mismatch")
359
    if input.domain is DomainTuple.scalar_domain():
360
        return ScalingOperator(input.domain, input.val[()])
Martin Reinecke's avatar
Martin Reinecke committed
361
362
363
    if isinstance(input, Field):
        return DiagonalOperator(input)
    if isinstance(input, MultiField):
Martin Reinecke's avatar
Martin Reinecke committed
364
        return BlockDiagonalOperator(
Martin Reinecke's avatar
fix    
Martin Reinecke committed
365
            input.domain, {key: makeOp(val) for key, val in input.items()})
Martin Reinecke's avatar
Martin Reinecke committed
366
367
    raise NotImplementedError

Martin Reinecke's avatar
more    
Martin Reinecke committed
368
369

def domain_union(domains):
Martin Reinecke's avatar
Martin Reinecke committed
370
371
372
373
374
375
376
377
    """Computes the union of multiple DomainTuples/MultiDomains.

    Parameters
    ----------
    domains : list of DomainTuple or MultiDomain
        - if DomainTuple, all entries must be equal
        - if MultiDomain, there must not be any conflicting components
    """
Martin Reinecke's avatar
more    
Martin Reinecke committed
378
    if isinstance(domains[0], DomainTuple):
Martin Reinecke's avatar
Martin Reinecke committed
379
        if any(dom != domains[0] for dom in domains[1:]):
Martin Reinecke's avatar
more    
Martin Reinecke committed
380
381
382
383
            raise ValueError("domain mismatch")
        return domains[0]
    return MultiDomain.union(domains)

Martin Reinecke's avatar
more    
Martin Reinecke committed
384

385
386
387
388
389
390
391
392
393
394
# Pointwise functions

_current_module = sys.modules[__name__]

for f in pointwise.ptw_dict.keys():
    def func(f):
        def func2(x, *args, **kwargs):
           return x.ptw(f, *args, **kwargs)
        return func2
    setattr(_current_module, f, func(f))
Martin Reinecke's avatar
Martin Reinecke committed
395
396


397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
def get_default_codomain(domainoid, space=None):
    """For `RGSpace`, returns the harmonic partner domain.
    For `DomainTuple`, returns a copy of the object in which the domain
    indexed by `space` is substituted by its harmonic partner domain.
    In this case, if `space` is None, it is set to 0 if the `DomainTuple`
    contains exactly one domain.

    Parameters
    ----------
    domain: `RGSpace` or `DomainTuple`
        Domain for which to constuct the default harmonic partner
    space: int
        Optional index of the subdomain to be replaced by its default
        codomain. `domain[space]` must be of class `RGSpace`.
    """
    from .domains.rg_space import RGSpace
413
414
415
    from .domains.hp_space import HPSpace
    from .domains.gl_space import GLSpace
    from .domains.lm_space import LMSpace
416
417
418
419
420
421
    if isinstance(domainoid, RGSpace):
        return domainoid.get_default_codomain()
    if not isinstance(domainoid, DomainTuple):
        raise TypeError(
            'Works only on RGSpaces and DomainTuples containing those')
    space = utilities.infer_space(domainoid, space)
422
423
    if not isinstance(domainoid[space], (RGSpace, HPSpace, GLSpace, LMSpace)):
        raise TypeError("can only codomain structrued spaces")
424
425
426
    ret = [dom for dom in domainoid]
    ret[space] = domainoid[space].get_default_codomain()
    return DomainTuple.make(ret)
Lukas Platz's avatar
Lukas Platz committed
427
428
429
430
431
432
433
434
435
436
437


def single_plot(field, **kwargs):
    """Creates a single plot using `Plot`.
    Keyword arguments are passed to both `Plot.add` and `Plot.output`.
    """
    p = Plot()
    p.add(field, **kwargs)
    if 'title' in kwargs:
        del(kwargs['title'])
    p.output(**kwargs)
438
439
440
441


def exec_time(obj, want_metric=True):
    """Times the execution time of an operator or an energy."""
Philipp Arras's avatar
Philipp Arras committed
442
443
444
    from .linearization import Linearization
    from .minimization.energy import Energy
    from .operators.energy_operators import EnergyOperator
445
446
447
    if isinstance(obj, Energy):
        t0 = time()
        obj.at(0.99*obj.position)
448
        logger.info('Energy.at(): {}'.format(time() - t0))
449
450
451

        t0 = time()
        obj.value
452
        logger.info('Energy.value: {}'.format(time() - t0))
453
454
        t0 = time()
        obj.gradient
455
        logger.info('Energy.gradient: {}'.format(time() - t0))
456
457
        t0 = time()
        obj.metric
458
        logger.info('Energy.metric: {}'.format(time() - t0))
459
460
461

        t0 = time()
        obj.apply_metric(obj.position)
462
        logger.info('Energy.apply_metric: {}'.format(time() - t0))
463
464
465

        t0 = time()
        obj.metric(obj.position)
466
        logger.info('Energy.metric(position): {}'.format(time() - t0))
467
468
469
470
471
    elif isinstance(obj, Operator):
        want_metric = bool(want_metric)
        pos = from_random('normal', obj.domain)
        t0 = time()
        obj(pos)
472
        logger.info('Operator call with field: {}'.format(time() - t0))
473
474
475
476

        lin = Linearization.make_var(pos, want_metric=want_metric)
        t0 = time()
        res = obj(lin)
477
        logger.info('Operator call with linearization: {}'.format(time() - t0))
478
479
480
481

        if isinstance(obj, EnergyOperator):
            t0 = time()
            res.gradient
482
            logger.info('Gradient evaluation: {}'.format(time() - t0))
483
484
485
486

            if want_metric:
                t0 = time()
                res.metric(pos)
487
                logger.info('Metric apply: {}'.format(time() - t0))
488
489
    else:
        raise TypeError
Philipp Arras's avatar
Philipp Arras committed
490
491
492
493
494
495
496
497
498
499
500
501
502


def calculate_position(operator, output):
    """Finds approximate preimage of an operator for a given output."""
    from .minimization.descent_minimizers import NewtonCG
    from .minimization.iteration_controllers import GradientNormController
    from .minimization.metric_gaussian_kl import MetricGaussianKL
    from .operators.scaling_operator import ScalingOperator
    from .operators.energy_operators import GaussianEnergy, StandardHamiltonian
    if not isinstance(operator, Operator):
        raise TypeError
    if output.domain != operator.target:
        raise TypeError
503
504
505
506
    if isinstance(output, MultiField):
        cov = 1e-3*max([vv.max() for vv in output.val.values()])**2
    else:
        cov = 1e-3*output.val.max()**2
507
    invcov = ScalingOperator(output.domain, cov).inverse
Philipp Arras's avatar
Fixup    
Philipp Arras committed
508
    d = output + invcov.draw_sample_with_dtype(dtype=output.dtype, from_inverse=True)
Philipp Arras's avatar
Philipp Arras committed
509
    lh = GaussianEnergy(d, invcov) @ operator
Philipp Arras's avatar
Philipp Arras committed
510
511
512
513
514
515
516
517
518
    H = StandardHamiltonian(
        lh, ic_samp=GradientNormController(iteration_limit=200))
    pos = 0.1*from_random('normal', operator.domain)
    minimizer = NewtonCG(GradientNormController(iteration_limit=10))
    for ii in range(3):
        kl = MetricGaussianKL(pos, H, 3, mirror_samples=True)
        kl, _ = minimizer(kl)
        pos = kl.position
    return pos