There is a maintenance of MPCDF Gitlab on Thursday, April 22st 2020, 9:00 am CEST - Expect some service interruptions during this time

correlated_fields.py 15.4 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
    assert np.all(sig > 0)
    return Adder(from_global_data(domain, mean)) @ (
Philipp Haim's avatar
Fixes  
Philipp Haim committed
69
        DiagonalOperator(from_global_data(domain,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:
Philipp Haim's avatar
Fixes  
Philipp Haim committed
85 86 87 88
            print(self._sc.shape)
            print(x.shape)
            print(x[self._last].shape)
            res = x - np.tensordot(x[self._last], self._sc, axes = 0)
Philipp Frank's avatar
Philipp Frank committed
89
        else:
90
            #NOTE Why not x.copy()?
Philipp Frank's avatar
Philipp Frank committed
91
            res = np.zeros(x.shape,dtype=x.dtype)
Philipp Frank's avatar
Philipp Frank committed
92
            res += x
93
            res[self._last] -= (x*self._sc).sum(axis = self._space)
Philipp Frank's avatar
Philipp Frank committed
94 95
        return from_global_data(self._tgt(mode),res)

96
def _make_slope_Operator(smooth,loglogavgslope, space = 0):
Philipp Frank's avatar
Philipp Frank committed
97
    tg = smooth.target
98
    logkl = _log_k_lengths(tg[space])
Philipp Frank's avatar
Philipp Frank committed
99 100
    logkl -= logkl[0]
    logkl = np.insert(logkl, 0, 0)
Philipp Haim's avatar
Fixes  
Philipp Haim committed
101
    noslope = smooth
102
    noslope = _SlopeRemover(tg,logkl, space) @ smooth
Philipp Frank's avatar
Philipp Frank committed
103 104 105
    # FIXME Move to tests
    consistency_check(_SlopeRemover(tg,logkl))

106
    expander = ContractionOperator(tg, spaces = space).adjoint
Philipp Haim's avatar
Fixes  
Philipp Haim committed
107
    _t = DiagonalOperator(from_global_data(tg[space], logkl), tg, spaces = space)
108
    return _t @ expander @ loglogavgslope + noslope
Philipp Arras's avatar
Philipp Arras committed
109 110 111 112 113

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

class _TwoLogIntegrations(LinearOperator):
114
    def __init__(self, target, space = None):
Philipp Arras's avatar
Philipp Arras committed
115
        self._target = makeDomain(target)
116 117 118 119 120
        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
121
        self._capability = self.TIMES | self.ADJOINT_TIMES
122
        logk_lengths = _log_k_lengths(self._target[space])
Philipp Arras's avatar
Philipp Arras committed
123 124 125 126
        self._logvol = logk_lengths[1:] - logk_lengths[:-1]

    def apply(self, x, mode):
        self._check_input(x, mode)
127 128 129 130

        #Maybe make class properties
        axis = self._target.axes[self._space][0]
        sl = (slice(None),)*axis
Philipp Haim's avatar
Fixes  
Philipp Haim committed
131
        extender_sl = (None,)*axis + (slice(None),) + (None,)*(self._target.axes[-1][-1] - axis)
132 133 134 135 136
        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
137 138 139
        if mode == self.TIMES:
            x = x.to_global_data()
            res = np.empty(self._target.shape)
140 141 142
            res[first] = 0
            res[second] = 0
            res[from_third] = np.cumsum(x[second], axis = axis)
Philipp Haim's avatar
Fixes  
Philipp Haim committed
143
            res[from_third] = (res[from_third] + res[no_border])/2*self._logvol[extender_sl] + x[first]
144
            res[from_third] = np.cumsum(res[from_third], axis = axis)
Philipp Arras's avatar
Philipp Arras committed
145 146 147
        else:
            x = x.to_global_data_rw()
            res = np.zeros(self._domain.shape)
148 149
            x[from_third] = np.cumsum(x[from_third][reverse], axis = axis)[reverse]
            res[first] += x[from_third]
Philipp Haim's avatar
Fixes  
Philipp Haim committed
150
            x[from_third] *= (self._logvol/2.)[extender_sl]
151 152 153
            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
154 155 156


class _Normalization(Operator):
157
    def __init__(self, domain, space = 0):
Philipp Arras's avatar
Philipp Arras committed
158
        self._domain = self._target = makeDomain(domain)
159 160 161 162
        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
163
        # TODO Does not work on sphere yet
164 165 166 167
        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
168 169
        # FIXME Move to tests
        consistency_check(self._specsum)
Philipp Arras's avatar
Philipp Arras committed
170 171 172 173 174 175 176

    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
177
        return self._specsum(self._mode_multiplicity*spec)**(-0.5)*amp
Philipp Arras's avatar
Philipp Arras committed
178 179 180


class _SpecialSum(EndomorphicOperator):
181
    def __init__(self, domain, space = 0):
Philipp Arras's avatar
Philipp Arras committed
182 183
        self._domain = makeDomain(domain)
        self._capability = self.TIMES | self.ADJOINT_TIMES
184 185
        self._contractor = ContractionOperator(domain, space)
        self._zero_mode = (slice(None),)*domain.axes[space][0] + (0,)
Philipp Arras's avatar
Philipp Arras committed
186 187 188

    def apply(self, x, mode):
        self._check_input(x, mode)
189
        return self._contractor.adjoint(self._contractor(x))
Philipp Arras's avatar
Philipp Arras committed
190 191


192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211
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
212 213 214
class CorrelatedFieldMaker:
    def __init__(self):
        self._amplitudes = []
215
        self._spaces = []
Philipp Arras's avatar
Philipp Arras committed
216 217

    def add_fluctuations_from_ops(self, target, fluctuations, flexibility,
218
                                  asperity, loglogavgslope, key, space = 0):
Philipp Arras's avatar
Philipp Arras committed
219 220 221 222 223 224 225 226 227 228 229
        """
        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)
230
        assert isinstance(target[space], PowerSpace)
Philipp Arras's avatar
Philipp Arras committed
231

232
        twolog = _TwoLogIntegrations(target, space)
233
        dt = twolog._logvol
Philipp Haim's avatar
Fixes  
Philipp Haim committed
234 235 236
        axis = target.axes[space][0]
        sl = (slice(None),)*axis
        extender_sl = (None,)*axis + (slice(None),) + (None,)*(target.axes[-1][-1] - axis)
237 238 239 240
        first = sl + (0,)
        second = sl + (1,)
        expander = ContractionOperator(twolog.domain, spaces = space).adjoint
        
Philipp Haim's avatar
Fixes  
Philipp Haim committed
241
        sqrt_t = np.zeros(twolog.domain[space].shape)
242
        sqrt_t[first] = sqrt_t[second] = np.sqrt(dt)
Philipp Haim's avatar
Fixes  
Philipp Haim committed
243
        sqrt_t = from_global_data(twolog.domain[space], sqrt_t)
244 245 246
        sqrt_t = DiagonalOperator(sqrt_t, twolog.domain, spaces = space)
        sigmasq = sqrt_t @ expander @ flexibility

Philipp Haim's avatar
Fixes  
Philipp Haim committed
247
        dist = np.zeros(twolog.domain[space].shape)
248
        dist[first] += 1.
Philipp Haim's avatar
Fixes  
Philipp Haim committed
249
        dist = from_global_data(twolog.domain[space], dist)
250
        dist = DiagonalOperator(dist, twolog.domain, spaces = space)
Philipp Arras's avatar
Philipp Arras committed
251

252
        shift = np.ones(twolog.domain.shape)
Philipp Haim's avatar
Fixes  
Philipp Haim committed
253
        shift[first] = (dt**2/12.)[extender_sl]
254 255
        shift = from_global_data(twolog.domain, shift)
        scale = sigmasq*(Adder(shift) @ dist @ expander @ asperity).sqrt()
Philipp Arras's avatar
Philipp Arras committed
256 257

        smooth = twolog @ (scale*ducktape(scale.target, None, key))
Philipp Haim's avatar
Fixes  
Philipp Haim committed
258
        smoothslope = _make_slope_Operator(smooth,loglogavgslope, space)
Philipp Frank's avatar
Philipp Frank committed
259
        
Philipp Arras's avatar
Philipp Arras committed
260 261 262
        # move to tests
        assert_allclose(
            smooth(from_random('normal', smooth.domain)).val[0:2], 0)
Philipp Arras's avatar
Philipp Arras committed
263
        consistency_check(twolog)
Philipp Arras's avatar
Philipp Arras committed
264 265
        check_jacobian_consistency(smooth, from_random('normal',
                                                       smooth.domain))
Philipp Arras's avatar
Philipp Arras committed
266 267
        check_jacobian_consistency(smoothslope,
                                   from_random('normal', smoothslope.domain))
Philipp Arras's avatar
Philipp Arras committed
268 269
        # end move to tests

Philipp Haim's avatar
Fixes  
Philipp Haim committed
270
        normal_ampl = _Normalization(target, space) @ smoothslope
Philipp Arras's avatar
Philipp Arras committed
271
        vol = target[0].harmonic_partner.get_default_codomain().total_volume
Philipp Haim's avatar
Fixes  
Philipp Haim committed
272
        arr = np.zeros(target[space].shape)
Philipp Arras's avatar
Philipp Arras committed
273
        arr[1:] = vol
Philipp Haim's avatar
Fixes  
Philipp Haim committed
274 275 276
        expander = ContractionOperator(target, spaces = space).adjoint
        expander = DiagonalOperator(from_global_data(target[space], arr)
                , target, spaces = space) @ expander
Philipp Arras's avatar
Philipp Arras committed
277 278 279 280 281 282 283 284 285 286
        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 Haim's avatar
Fixes  
Philipp Haim committed
287
        op = _Normalization(target, space)
Philipp Arras's avatar
Philipp Arras committed
288
        check_jacobian_consistency(op, from_random('normal', op.domain))
Philipp Arras's avatar
Philipp Arras committed
289 290 291
        # End move to tests

        self._amplitudes.append(ampl)
292
        self._spaces.append(space)
Philipp Arras's avatar
Philipp Arras committed
293 294 295 296

    def add_fluctuations(self, target, fluctuations_mean, fluctuations_stddev,
                         flexibility_mean, flexibility_stddev, asperity_mean,
                         asperity_stddev, loglogavgslope_mean,
297
                         loglogavgslope_stddev, prefix, space = 0):
Philipp Arras's avatar
Philipp Arras committed
298 299
        prefix = str(prefix)

300 301 302 303 304 305 306
        parameter_domain = list(makeDomain(target))
        del parameter_domain[space]
        if parameter_domain != []:
            parameter_domain = makeDomain(parameter_domain)
        else:
            parameter_domain = DomainTuple.scalar_domain()

307
        fluct = _lognormal_moment_matching(fluctuations_mean, fluctuations_stddev,
308
                        prefix + 'fluctuations', parameter_domain, space = space)
Philipp Arras's avatar
Philipp Arras committed
309
        flex = _lognormal_moment_matching(flexibility_mean, flexibility_stddev,
310
                        prefix + 'flexibility', parameter_domain, space = space)
Philipp Arras's avatar
Philipp Arras committed
311
        asp = _lognormal_moment_matching(asperity_mean, asperity_stddev,
312
                        prefix + 'asperity', parameter_domain, space = space)
313
        avgsl = _normal(loglogavgslope_mean, loglogavgslope_stddev,
314 315
                        prefix + 'loglogavgslope', parameter_domain, space = space)

Philipp Haim's avatar
Fixes  
Philipp Haim committed
316
        return self.add_fluctuations_from_ops(target, fluct, flex, asp, avgsl,
317
                                       prefix + 'spectrum', space)
Philipp Arras's avatar
Philipp Arras committed
318 319 320 321 322 323 324 325 326 327 328 329

    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
        """
330
        prefix = str(prefix)
Philipp Arras's avatar
Philipp Arras committed
331
        if offset is not None:
332
            offset = float(offset) 
333 334 335 336 337 338 339 340 341
        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
342

343 344 345 346 347 348 349 350
        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
351 352
        azm = _lognormal_moment_matching(offset_amplitude_mean,
                                         offset_amplitude_stddev,
353 354
                                         prefix + 'zeromode', parameter_domain,
                                         space = tuple(self._spaces))
355

Philipp Arras's avatar
Philipp Arras committed
356 357 358
        foo = np.ones(hspace.shape)
        foo[zeroind] = 0

359 360 361 362 363 364 365 366
        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
367
        for i in range(1, len(self._amplitudes)):
368 369 370
            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
371 372 373 374 375 376 377 378 379 380 381 382 383

        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