correlated_fields.py 14.8 KB
Newer Older
Martin Reinecke's avatar
Martin Reinecke committed
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
Philipp Arras's avatar
Philipp Arras committed
15
# Authors: Philipp Frank, Philipp Arras
Martin Reinecke's avatar
Martin Reinecke committed
16
#
17
# NIFTy is being developed at the Max-Planck-Institut fuer Astrophysik.
Philipp Arras's avatar
Philipp Arras committed
18

Philipp Arras's avatar
Philipp Arras committed
19
import numpy as np
20
from functools import reduce
Philipp Arras's avatar
Philipp Arras committed
21
from numpy.testing import assert_allclose
22

Philipp Arras's avatar
Philipp Arras committed
23
from ..domain_tuple import DomainTuple
Philipp Arras's avatar
Philipp Arras committed
24 25
from ..domains.power_space import PowerSpace
from ..domains.unstructured_domain import UnstructuredDomain
Philipp Arras's avatar
Philipp Arras committed
26
from ..extra import check_jacobian_consistency, consistency_check
27
from ..field import Field
Philipp Arras's avatar
Philipp Arras committed
28
from ..multi_domain import MultiDomain
Philipp Arras's avatar
Philipp Arras committed
29
from ..operators.adder import Adder
30
from ..operators.contraction_operator import ContractionOperator
Philipp Arras's avatar
Philipp Arras committed
31
from ..operators.distributors import PowerDistributor
Philipp Arras's avatar
Philipp Arras committed
32
from ..operators.endomorphic_operator import EndomorphicOperator
Martin Reinecke's avatar
Martin Reinecke committed
33
from ..operators.harmonic_operators import HarmonicTransformOperator
Philipp Arras's avatar
Philipp Arras committed
34
from ..operators.linear_operator import LinearOperator
35
from ..operators.diagonal_operator import DiagonalOperator
Philipp Arras's avatar
Philipp Arras committed
36 37
from ..operators.operator import Operator
from ..operators.simple_linear_operators import VdotOperator, ducktape
Philipp Arras's avatar
Philipp Arras committed
38
from ..operators.value_inserter import ValueInserter
39 40
from ..sugar import from_global_data, from_random, full, makeDomain, get_default_codomain

41
def _reshaper(x, shape):
42
    x = np.array(x)
43

44 45 46 47 48 49 50 51 52 53
    if x.shape == shape:
        return np.asfarray(x)
    elif x.shape in [(), (1,)]:
        return np.full(shape, x, dtype=np.float)
    else:
        raise TypeError("Shape of parameters cannot be interpreted")

def _lognormal_moment_matching(mean, sig, key,
        domain = DomainTuple.scalar_domain(), space = 0):
    domain = makeDomain(domain)
54
    mean, sig = (_reshaper(param, domain.shape) for param in (mean, sig))
55 56 57
    key = str(key)
    assert np.all(mean > 0)
    assert np.all(sig > 0)
Philipp Arras's avatar
Philipp Arras committed
58 59
    logsig = np.sqrt(np.log((sig/mean)**2 + 1))
    logmean = np.log(mean) - logsig**2/2
60
    return _normal(logmean, logsig, key, domain).exp()
Philipp Arras's avatar
Philipp Arras committed
61 62


63 64 65
def _normal(mean, sig, key,
        domain = DomainTuple.scalar_domain(), space = 0):
    domain = makeDomain(domain)
66
    mean, sig = (_reshaper(param, domain.shape) for param in (mean, sig))
67 68 69
    assert np.all(sig > 0)
    return Adder(from_global_data(domain, mean)) @ (
        sig*ducktape(domain, None, key))
Philipp Arras's avatar
Philipp Arras committed
70 71


Philipp Frank's avatar
Philipp Frank committed
72
class _SlopeRemover(EndomorphicOperator):
73
    def __init__(self, domain, cooridinates, space = 0):
Philipp Frank's avatar
Philipp Frank committed
74
        self._domain = makeDomain(domain)
75
        self._sc = cooridinates / float(cooridinates[-1])
Philipp Arras's avatar
Philipp Arras committed
76

77 78
        self._space = space
        self._last = (slice(None),)*self._domain.axes[space][0] + (-1,)
Philipp Frank's avatar
Philipp Frank committed
79
        self._capability = self.TIMES | self.ADJOINT_TIMES
Philipp Arras's avatar
Philipp Arras committed
80

Philipp Frank's avatar
Philipp Frank committed
81 82 83 84
    def apply(self,x,mode):
        self._check_input(x,mode)
        x = x.to_global_data()
        if mode == self.TIMES:
85
            res = x - x[self._last] * self._sc
Philipp Frank's avatar
Philipp Frank committed
86
        else:
87
            #NOTE Why not x.copy()?
Philipp Frank's avatar
Philipp Frank committed
88
            res = np.zeros(x.shape,dtype=x.dtype)
Philipp Frank's avatar
Philipp Frank committed
89
            res += x
90
            res[self._last] -= (x*self._sc).sum(axis = self._space)
Philipp Frank's avatar
Philipp Frank committed
91 92
        return from_global_data(self._tgt(mode),res)

93
def _make_slope_Operator(smooth,loglogavgslope, space = 0):
Philipp Frank's avatar
Philipp Frank committed
94
    tg = smooth.target
95
    logkl = _log_k_lengths(tg[space])
Philipp Frank's avatar
Philipp Frank committed
96 97
    logkl -= logkl[0]
    logkl = np.insert(logkl, 0, 0)
98
    noslope = _SlopeRemover(tg,logkl, space) @ smooth
Philipp Frank's avatar
Philipp Frank committed
99 100 101
    # FIXME Move to tests
    consistency_check(_SlopeRemover(tg,logkl))

102 103 104
    expander = ContractionOperator(tg, spaces = space).adjoint
    _t = DiagonalOperator(from_global_data(tg, logkl), tg, spaces = space)
    return _t @ expander @ loglogavgslope + noslope
Philipp Arras's avatar
Philipp Arras committed
105 106 107 108 109

def _log_k_lengths(pspace):
    return np.log(pspace.k_lengths[1:])

class _TwoLogIntegrations(LinearOperator):
110
    def __init__(self, target, space = None):
Philipp Arras's avatar
Philipp Arras committed
111
        self._target = makeDomain(target)
112 113 114 115 116
        assert isinstance(self.target[space], PowerSpace)
        dom = list(self._target)
        dom[space] = UnstructuredDomain((2, self.target[space].shape[0]-2))
        self._domain = makeDomain(dom)
        self._space = space
Philipp Arras's avatar
Philipp Arras committed
117
        self._capability = self.TIMES | self.ADJOINT_TIMES
118
        logk_lengths = _log_k_lengths(self._target[space])
Philipp Arras's avatar
Philipp Arras committed
119 120 121 122
        self._logvol = logk_lengths[1:] - logk_lengths[:-1]

    def apply(self, x, mode):
        self._check_input(x, mode)
123 124 125 126 127 128 129 130 131

        #Maybe make class properties
        axis = self._target.axes[self._space][0]
        sl = (slice(None),)*axis
        first = sl + (0,)
        second = sl + (1,)
        from_third = sl + (slice(2,None),)
        no_border = sl + (slice(1,-1),)
        reverse = sl + (slice(None,None,-1),)
Philipp Arras's avatar
Philipp Arras committed
132 133 134
        if mode == self.TIMES:
            x = x.to_global_data()
            res = np.empty(self._target.shape)
135 136 137 138 139
            res[first] = 0
            res[second] = 0
            res[from_third] = np.cumsum(x[second], axis = axis)
            res[from_third] = (res[from_third] + res[no_border])/2*self._logvol + x[first]
            res[from_third] = np.cumsum(res[from_third], axis = axis)
Philipp Arras's avatar
Philipp Arras committed
140 141 142
        else:
            x = x.to_global_data_rw()
            res = np.zeros(self._domain.shape)
143 144 145 146 147 148
            x[from_third] = np.cumsum(x[from_third][reverse], axis = axis)[reverse]
            res[first] += x[from_third]
            x[from_third] *= self._logvol/2.
            x[no_border] += x[from_third]
            res[second] += np.cumsum(x[from_third][reverse], axis = axis)[reverse]
        return from_global_data(self._tgt(mode), res)
Philipp Arras's avatar
Philipp Arras committed
149 150 151


class _Normalization(Operator):
152
    def __init__(self, domain, space = 0):
Philipp Arras's avatar
Philipp Arras committed
153
        self._domain = self._target = makeDomain(domain)
154 155 156 157
        hspace = list(self._domain)
        hspace[space] = hspace[space].harmonic_partner
        hspace = makeDomain(hspace)
        pd = PowerDistributor(hspace, power_space=self._domain[space], space = space)
Philipp Arras's avatar
Philipp Arras committed
158
        # TODO Does not work on sphere yet
159 160 161 162
        mode_multiplicity = pd.adjoint(full(pd.target, 1.)).to_global_data_rw()
        mode_multiplicity[0] = 0
        self._mode_multiplicity = from_global_data(self._domain, mode_multiplicity)
        self._specsum = _SpecialSum(self._domain, space)
Philipp Arras's avatar
Philipp Arras committed
163 164
        # FIXME Move to tests
        consistency_check(self._specsum)
Philipp Arras's avatar
Philipp Arras committed
165 166 167 168 169 170 171

    def apply(self, x):
        self._check_input(x)
        amp = x.exp()
        spec = (2*x).exp()
        # FIXME This normalizes also the zeromode which is supposed to be left
        # untouched by this operator
172
        return self._specsum(self._mode_multiplicity*spec)**(-0.5)*amp
Philipp Arras's avatar
Philipp Arras committed
173 174 175


class _SpecialSum(EndomorphicOperator):
176
    def __init__(self, domain, space = 0):
Philipp Arras's avatar
Philipp Arras committed
177 178
        self._domain = makeDomain(domain)
        self._capability = self.TIMES | self.ADJOINT_TIMES
179 180
        self._contractor = ContractionOperator(domain, space)
        self._zero_mode = (slice(None),)*domain.axes[space][0] + (0,)
Philipp Arras's avatar
Philipp Arras committed
181 182 183

    def apply(self, x, mode):
        self._check_input(x, mode)
184
        return self._contractor.adjoint(self._contractor(x))
Philipp Arras's avatar
Philipp Arras committed
185 186


187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206
class _slice_extractor(LinearOperator):
    #FIXME it should be tested if the the domain and target are consistent with the slice
    def __init__(self, domain, target, sl):
        self._domain = makeDomain(domain)
        self._target = makeDomain(target)
        self._sl = sl
        self._capability = self.TIMES | self.ADJOINT_TIMES

    def apply(self, x, mode):
        self._check_input(x, mode)
        x = x.to_global_data()
        if mode == self.TIMES:
            res = x[self._sl]
            res = res.reshape(self._target.shape)
        else:
            res = np.zeros(self._domain.shape)
            res[self._sl] = x
        return from_global_data(self._tgt(mode), res)
    

Philipp Arras's avatar
Philipp Arras committed
207 208 209
class CorrelatedFieldMaker:
    def __init__(self):
        self._amplitudes = []
210
        self._spaces = []
Philipp Arras's avatar
Philipp Arras committed
211 212

    def add_fluctuations_from_ops(self, target, fluctuations, flexibility,
213
                                  asperity, loglogavgslope, key, space = 0):
Philipp Arras's avatar
Philipp Arras committed
214 215 216 217 218 219 220 221 222 223 224
        """
        fluctuations > 0
        flexibility > 0
        asperity > 0
        loglogavgslope probably negative
        """
        assert isinstance(fluctuations, Operator)
        assert isinstance(flexibility, Operator)
        assert isinstance(asperity, Operator)
        assert isinstance(loglogavgslope, Operator)
        target = makeDomain(target)
225
        assert isinstance(target[space], PowerSpace)
Philipp Arras's avatar
Philipp Arras committed
226

227
        twolog = _TwoLogIntegrations(target, space)
228 229 230 231 232 233 234 235 236 237 238 239
        dt = twolog._logvol
        sl = (slice(None),)*target.axes[space][0]
        first = sl + (0,)
        second = sl + (1,)
        expander = ContractionOperator(twolog.domain, spaces = space).adjoint
        
        sqrt_t = np.zeros(twolog.domain.shape)
        sqrt_t[first] = sqrt_t[second] = np.sqrt(dt)
        sqrt_t = from_global_data(twolog.domain, sqrt_t)
        sqrt_t = DiagonalOperator(sqrt_t, twolog.domain, spaces = space)
        sigmasq = sqrt_t @ expander @ flexibility

Philipp Arras's avatar
Philipp Arras committed
240
        dist = np.zeros(twolog.domain.shape)
241
        dist[first] += 1.
Philipp Arras's avatar
Philipp Arras committed
242
        dist = from_global_data(twolog.domain, dist)
243
        dist = DiagonalOperator(dist, twolog.domain, spaces = space)
Philipp Arras's avatar
Philipp Arras committed
244

245 246 247 248
        shift = np.ones(twolog.domain.shape)
        shift[first] = dt**2/12.
        shift = from_global_data(twolog.domain, shift)
        scale = sigmasq*(Adder(shift) @ dist @ expander @ asperity).sqrt()
Philipp Arras's avatar
Philipp Arras committed
249 250

        smooth = twolog @ (scale*ducktape(scale.target, None, key))
Philipp Frank's avatar
Philipp Frank committed
251 252
        smoothslope = _make_slope_Operator(smooth,loglogavgslope)
        
Philipp Arras's avatar
Philipp Arras committed
253 254 255
        # move to tests
        assert_allclose(
            smooth(from_random('normal', smooth.domain)).val[0:2], 0)
Philipp Arras's avatar
Philipp Arras committed
256
        consistency_check(twolog)
Philipp Arras's avatar
Philipp Arras committed
257 258
        check_jacobian_consistency(smooth, from_random('normal',
                                                       smooth.domain))
Philipp Arras's avatar
Philipp Arras committed
259 260
        check_jacobian_consistency(smoothslope,
                                   from_random('normal', smoothslope.domain))
Philipp Arras's avatar
Philipp Arras committed
261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277
        # end move to tests

        normal_ampl = _Normalization(target) @ smoothslope
        vol = target[0].harmonic_partner.get_default_codomain().total_volume
        arr = np.zeros(target.shape)
        arr[1:] = vol
        expander = VdotOperator(from_global_data(target, arr)).adjoint
        mask = np.zeros(target.shape)
        mask[0] = vol
        adder = Adder(from_global_data(target, mask))
        ampl = adder @ ((expander @ fluctuations)*normal_ampl)

        # Move to tests
        # FIXME This test fails but it is not relevant for the final result
        # assert_allclose(
        #     normal_ampl(from_random('normal', normal_ampl.domain)).val[0], 1)
        assert_allclose(ampl(from_random('normal', ampl.domain)).val[0], vol)
Philipp Arras's avatar
Philipp Arras committed
278 279
        op = _Normalization(target)
        check_jacobian_consistency(op, from_random('normal', op.domain))
Philipp Arras's avatar
Philipp Arras committed
280 281 282
        # End move to tests

        self._amplitudes.append(ampl)
283
        self._spaces.append(space)
Philipp Arras's avatar
Philipp Arras committed
284 285 286 287

    def add_fluctuations(self, target, fluctuations_mean, fluctuations_stddev,
                         flexibility_mean, flexibility_stddev, asperity_mean,
                         asperity_stddev, loglogavgslope_mean,
288
                         loglogavgslope_stddev, prefix, space = 0):
Philipp Arras's avatar
Philipp Arras committed
289 290
        prefix = str(prefix)

291 292 293 294 295 296 297
        parameter_domain = list(makeDomain(target))
        del parameter_domain[space]
        if parameter_domain != []:
            parameter_domain = makeDomain(parameter_domain)
        else:
            parameter_domain = DomainTuple.scalar_domain()

298
        fluct = _lognormal_moment_matching(fluctuations_mean, fluctuations_stddev,
299
                        prefix + 'fluctuations', parameter_domain, space = space)
Philipp Arras's avatar
Philipp Arras committed
300
        flex = _lognormal_moment_matching(flexibility_mean, flexibility_stddev,
301
                        prefix + 'flexibility', parameter_domain, space = space)
Philipp Arras's avatar
Philipp Arras committed
302
        asp = _lognormal_moment_matching(asperity_mean, asperity_stddev,
303
                        prefix + 'asperity', parameter_domain, space = space)
304
        avgsl = _normal(loglogavgslope_mean, loglogavgslope_stddev,
305 306
                        prefix + 'loglogavgslope', parameter_domain, space = space)

Philipp Arras's avatar
Philipp Arras committed
307
        self.add_fluctuations_from_ops(target, fluct, flex, asp, avgsl,
308
                                       prefix + 'spectrum', space)
Philipp Arras's avatar
Philipp Arras committed
309 310 311 312 313 314 315 316 317 318 319 320

    def finalize_from_op(self, zeromode):
        raise NotImplementedError

    def finalize(self,
                 offset_amplitude_mean,
                 offset_amplitude_stddev,
                 prefix,
                 offset=None):
        """
        offset vs zeromode: volume factor
        """
321
        prefix = str(prefix)
Philipp Arras's avatar
Philipp Arras committed
322
        if offset is not None:
323
            offset = float(offset) 
324 325 326 327 328 329 330 331 332
        hspace = []
        zeroind = ()
        for amp, space in zip(self._amplitudes, self._spaces):
            dd =  list(amp.target)
            dd[space] = dd[space].harmonic_partner
            hspace.extend(dd)
            zeroind += (slice(None),)*space + (0,)*len(dd[space].shape)
        hspace = makeDomain(hspace)
        spaces = np.cumsum(self._spaces) + np.arange(len(self._spaces))
Philipp Arras's avatar
Philipp Arras committed
333

334 335 336 337 338 339 340 341
        parameter_domain = list(makeDomain(hspace))
        for space in self._spaces:
            del parameter_domain[space]
        if parameter_domain != []:
            parameter_domain = makeDomain(parameter_domain)
        else:
            parameter_domain = DomainTuple.scalar_domain()

Philipp Arras's avatar
Philipp Arras committed
342 343
        azm = _lognormal_moment_matching(offset_amplitude_mean,
                                         offset_amplitude_stddev,
344 345
                                         prefix + 'zeromode', parameter_domain,
                                         space = tuple(self._spaces))
346

Philipp Arras's avatar
Philipp Arras committed
347 348 349
        foo = np.ones(hspace.shape)
        foo[zeroind] = 0

350 351 352 353 354 355 356 357
        ZeroModeInserter = _slice_extractor(hspace, azm.target, zeroind).adjoint

        azm = Adder(from_global_data(hspace, foo)) @ ZeroModeInserter @ azm

        #NOTE ht and pd operator able to act on several spaces might be nice
        ht = HarmonicTransformOperator(hspace, space = spaces[0])
        pd = PowerDistributor(hspace, 
                self._amplitudes[0].target[spaces[0]], spaces[0])
Philipp Arras's avatar
Philipp Arras committed
358
        for i in range(1, len(self._amplitudes)):
359 360 361
            ht = HarmonicTransformOperator(ht.target, space = spaces[i]) @ ht
            pd = pd @ PowerDistributor( pd.domain, 
                    self._amplitudes[i].target[spaces[i]], space = spaces[i])
Philipp Arras's avatar
Philipp Arras committed
362 363 364 365 366 367 368 369 370 371 372 373 374

        a = ContractionOperator(pd.domain,
                                spaces[1:]).adjoint(self._amplitudes[0])
        for i in range(1, len(self._amplitudes)):
            a = a*(ContractionOperator(pd.domain, spaces[:i] + spaces[
                (i + 1):]).adjoint(self._amplitudes[i]))

        A = pd @ a
        return ht(azm*A*ducktape(hspace, None, prefix + 'xi'))

    @property
    def amplitudes(self):
        return self._amplitudes