field.py 29.3 KB
Newer Older
csongor's avatar
csongor committed
1
2
3
from __future__ import division
import numpy as np

Theo Steininger's avatar
Theo Steininger committed
4
5
from keepers import Versionable,\
                    Loggable
Jait Dixit's avatar
Jait Dixit committed
6

7
from d2o import distributed_data_object,\
8
    STRATEGIES as DISTRIBUTION_STRATEGIES
csongor's avatar
csongor committed
9

10
from nifty.config import nifty_configuration as gc
csongor's avatar
csongor committed
11

12
from nifty.domain_object import DomainObject
13

14
from nifty.spaces.power_space import PowerSpace
csongor's avatar
csongor committed
15

csongor's avatar
csongor committed
16
import nifty.nifty_utilities as utilities
17
18
from nifty.random import Random

csongor's avatar
csongor committed
19

Jait Dixit's avatar
Jait Dixit committed
20
class Field(Loggable, Versionable, object):
Theo Steininger's avatar
Theo Steininger committed
21
    # ---Initialization methods---
22

23
    def __init__(self, domain=None, val=None, dtype=None,
24
                 distribution_strategy=None, copy=False):
csongor's avatar
csongor committed
25

26
        self.domain = self._parse_domain(domain=domain, val=val)
27
        self.domain_axes = self._get_axes_tuple(self.domain)
csongor's avatar
csongor committed
28

Theo Steininger's avatar
Theo Steininger committed
29
        self.dtype = self._infer_dtype(dtype=dtype,
Jait Dixit's avatar
Jait Dixit committed
30
                                       val=val,
31
                                       domain=self.domain)
32

33
34
35
        self.distribution_strategy = self._parse_distribution_strategy(
                                distribution_strategy=distribution_strategy,
                                val=val)
csongor's avatar
csongor committed
36

37
38
39
40
        if val is None:
            self._val = None
        else:
            self.set_val(new_val=val, copy=copy)
csongor's avatar
csongor committed
41

42
    def _parse_domain(self, domain, val=None):
43
        if domain is None:
44
45
46
47
            if isinstance(val, Field):
                domain = val.domain
            else:
                domain = ()
48
        elif isinstance(domain, DomainObject):
49
            domain = (domain,)
50
51
52
        elif not isinstance(domain, tuple):
            domain = tuple(domain)

csongor's avatar
csongor committed
53
        for d in domain:
54
            if not isinstance(d, DomainObject):
55
56
                raise TypeError(
                    "Given domain contains something that is not a "
57
                    "DomainObject instance.")
csongor's avatar
csongor committed
58
59
        return domain

Theo Steininger's avatar
Theo Steininger committed
60
61
62
63
64
65
66
67
68
69
    def _get_axes_tuple(self, things_with_shape, start=0):
        i = start
        axes_list = []
        for thing in things_with_shape:
            l = []
            for j in range(len(thing.shape)):
                l += [i]
                i += 1
            axes_list += [tuple(l)]
        return tuple(axes_list)
70

71
    def _infer_dtype(self, dtype, val, domain):
csongor's avatar
csongor committed
72
        if dtype is None:
73
74
75
            if isinstance(val, Field) or \
               isinstance(val, distributed_data_object):
                dtype = val.dtype
Theo Steininger's avatar
Theo Steininger committed
76
77
78
79
80
            dtype_tuple = (np.dtype(gc['default_field_dtype']),)
        else:
            dtype_tuple = (np.dtype(dtype),)
        if domain is not None:
            dtype_tuple += tuple(np.dtype(sp.dtype) for sp in domain)
csongor's avatar
csongor committed
81

Theo Steininger's avatar
Theo Steininger committed
82
        dtype = reduce(lambda x, y: np.result_type(x, y), dtype_tuple)
83

Theo Steininger's avatar
Theo Steininger committed
84
        return dtype
85

86
87
    def _parse_distribution_strategy(self, distribution_strategy, val):
        if distribution_strategy is None:
88
            if isinstance(val, distributed_data_object):
89
                distribution_strategy = val.distribution_strategy
90
            elif isinstance(val, Field):
91
                distribution_strategy = val.distribution_strategy
92
            else:
93
                self.logger.info("Datamodel set to default!")
94
                distribution_strategy = gc['default_distribution_strategy']
95
        elif distribution_strategy not in DISTRIBUTION_STRATEGIES['global']:
96
97
98
            raise ValueError(
                    "distribution_strategy must be a global-type "
                    "strategy.")
99
        return distribution_strategy
100
101

    # ---Factory methods---
102

103
    @classmethod
104
    def from_random(cls, random_type, domain=None, dtype=None,
105
                    distribution_strategy=None, **kwargs):
106
        # create a initially empty field
107
        f = cls(domain=domain, dtype=dtype,
108
                distribution_strategy=distribution_strategy)
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143

        # now use the processed input in terms of f in order to parse the
        # random arguments
        random_arguments = cls._parse_random_arguments(random_type=random_type,
                                                       f=f,
                                                       **kwargs)

        # extract the distributed_dato_object from f and apply the appropriate
        # random number generator to it
        sample = f.get_val(copy=False)
        generator_function = getattr(Random, random_type)
        sample.apply_generator(
            lambda shape: generator_function(dtype=f.dtype,
                                             shape=shape,
                                             **random_arguments))
        return f

    @staticmethod
    def _parse_random_arguments(random_type, f, **kwargs):

        if random_type == "pm1":
            random_arguments = {}

        elif random_type == "normal":
            mean = kwargs.get('mean', 0)
            std = kwargs.get('std', 1)
            random_arguments = {'mean': mean,
                                'std': std}

        elif random_type == "uniform":
            low = kwargs.get('low', 0)
            high = kwargs.get('high', 1)
            random_arguments = {'low': low,
                                'high': high}

csongor's avatar
csongor committed
144
        else:
145
146
            raise KeyError(
                "unsupported random key '" + str(random_type) + "'.")
csongor's avatar
csongor committed
147

148
        return random_arguments
csongor's avatar
csongor committed
149

150
151
152
153
154
155
156
157
158
    # ---Powerspectral methods---

    def power_analyze(self, spaces=None, log=False, nbin=None, binbounds=None,
                      real_signal=True):
        # assert that all spaces in `self.domain` are either harmonic or
        # power_space instances
        for sp in self.domain:
            if not sp.harmonic and not isinstance(sp, PowerSpace):
                raise AttributeError(
159
                    "Field has a space in `domain` which is neither "
160
161
162
                    "harmonic nor a PowerSpace.")

        # check if the `spaces` input is valid
163
164
165
166
167
        spaces = utilities.cast_axis_to_tuple(spaces, len(self.domain))
        if spaces is None:
            if len(self.domain) == 1:
                spaces = (0,)
            else:
168
169
170
                raise ValueError(
                    "Field has multiple spaces as domain "
                    "but `spaces` is None.")
171
172

        if len(spaces) == 0:
173
174
            raise ValueError(
                "No space for analysis specified.")
175
        elif len(spaces) > 1:
176
177
            raise ValueError(
                "Conversion of only one space at a time is allowed.")
178
179
180
181

        space_index = spaces[0]

        if not self.domain[space_index].harmonic:
182
183
            raise ValueError(
                "The analyzed space must be harmonic.")
184

185
186
187
188
189
190
        # Create the target PowerSpace instance:
        # If the associated signal-space field was real, we extract the
        # hermitian and anti-hermitian parts of `self` and put them
        # into the real and imaginary parts of the power spectrum.
        # If it was complex, all the power is put into a real power spectrum.

191
192
193
194
        distribution_strategy = \
            self.val.get_axes_local_distribution_strategy(
                self.domain_axes[space_index])

195
196
197
198
199
        if real_signal:
            power_dtype = np.dtype('complex')
        else:
            power_dtype = np.dtype('float')

200
201
        harmonic_domain = self.domain[space_index]
        power_domain = PowerSpace(harmonic_domain=harmonic_domain,
202
                                  distribution_strategy=distribution_strategy,
203
204
                                  log=log, nbin=nbin, binbounds=binbounds,
                                  dtype=power_dtype)
205

206
        # extract pindex and rho from power_domain
207
208
        pindex = power_domain.pindex
        rho = power_domain.rho
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226

        if real_signal:
            hermitian_part, anti_hermitian_part = \
                harmonic_domain.hermitian_decomposition(
                                            self.val,
                                            axes=self.domain_axes[space_index])

            [hermitian_power, anti_hermitian_power] = \
                [self._calculate_power_spectrum(
                                            x=part,
                                            pindex=pindex,
                                            rho=rho,
                                            axes=self.domain_axes[space_index])
                 for part in [hermitian_part, anti_hermitian_part]]

            power_spectrum = hermitian_power + 1j * anti_hermitian_power
        else:
            power_spectrum = self._calculate_power_spectrum(
227
228
229
230
231
232
233
234
235
                                            x=self.val,
                                            pindex=pindex,
                                            rho=rho,
                                            axes=self.domain_axes[space_index])

        # create the result field and put power_spectrum into it
        result_domain = list(self.domain)
        result_domain[space_index] = power_domain

236
237
238
        result_field = self.copy_empty(
                   domain=result_domain,
                   distribution_strategy=power_spectrum.distribution_strategy)
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
        result_field.set_val(new_val=power_spectrum, copy=False)

        return result_field

    def _calculate_power_spectrum(self, x, pindex, rho, axes=None):
        fieldabs = abs(x)
        fieldabs **= 2

        if axes is not None:
            pindex = self._shape_up_pindex(
                                    pindex=pindex,
                                    target_shape=x.shape,
                                    target_strategy=x.distribution_strategy,
                                    axes=axes)
        power_spectrum = pindex.bincount(weights=fieldabs,
                                         axis=axes)
        if axes is not None:
            new_rho_shape = [1, ] * len(power_spectrum.shape)
            new_rho_shape[axes[0]] = len(rho)
            rho = rho.reshape(new_rho_shape)
        power_spectrum /= rho

        power_spectrum **= 0.5
        return power_spectrum

    def _shape_up_pindex(self, pindex, target_shape, target_strategy, axes):
        if pindex.distribution_strategy not in \
                DISTRIBUTION_STRATEGIES['global']:
267
            raise ValueError("pindex's distribution strategy must be "
268
269
270
271
272
273
                             "global-type")

        if pindex.distribution_strategy in DISTRIBUTION_STRATEGIES['slicing']:
            if ((0 not in axes) or
                    (target_strategy is not pindex.distribution_strategy)):
                raise ValueError(
274
                    "A slicing distributor shall not be reshaped to "
275
276
277
278
279
280
281
282
283
284
285
286
287
                    "something non-sliced.")

        semiscaled_shape = [1, ] * len(target_shape)
        for i in axes:
            semiscaled_shape[i] = target_shape[i]
        local_data = pindex.get_local_data(copy=False)
        semiscaled_local_data = local_data.reshape(semiscaled_shape)
        result_obj = pindex.copy_empty(global_shape=target_shape,
                                       distribution_strategy=target_strategy)
        result_obj.set_full_data(semiscaled_local_data, copy=False)

        return result_obj

288
289
    def power_synthesize(self, spaces=None, real_signal=True,
                         mean=None, std=None):
290
        # assert that all spaces in `self.domain` are either of signal-type or
291
292
        # power_space instances
        for sp in self.domain:
293
            if not sp.harmonic and not isinstance(sp, PowerSpace):
294
                raise AttributeError(
295
                    "Field has a space in `domain` which is neither "
296
297
                    "harmonic nor a PowerSpace.")

298
299
300
301
302
303
        # check if the `spaces` input is valid
        spaces = utilities.cast_axis_to_tuple(spaces, len(self.domain))
        if spaces is None:
            if len(self.domain) == 1:
                spaces = (0,)
            else:
304
305
306
                raise ValueError(
                    "Field has multiple spaces as domain "
                    "but `spaces` is None.")
307
308

        if len(spaces) == 0:
309
310
            raise ValueError(
                "No space for synthesis specified.")
311
        elif len(spaces) > 1:
312
313
            raise ValueError(
                "Conversion of only one space at a time is allowed.")
314
315
316
317

        power_space_index = spaces[0]
        power_domain = self.domain[power_space_index]
        if not isinstance(power_domain, PowerSpace):
318
319
            raise ValueError(
                "A PowerSpace is needed for field synthetization.")
320
321
322
323
324
325
326
327
328
329
330
331
332
333

        # create the result domain
        result_domain = list(self.domain)
        harmonic_domain = power_domain.harmonic_domain
        result_domain[power_space_index] = harmonic_domain

        # create random samples: one or two, depending on whether the
        # power spectrum is real or complex

        if issubclass(power_domain.dtype.type, np.complexfloating):
            result_list = [None, None]
        else:
            result_list = [None]

334
335
        result_list = [self.__class__.from_random(
                             'normal',
336
337
338
                             mean=mean,
                             std=std,
                             domain=result_domain,
339
340
                             dtype=harmonic_domain.dtype,
                             distribution_strategy=self.distribution_strategy)
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
                       for x in result_list]

        # from now on extract the values from the random fields for further
        # processing without killing the fields.
        # if the signal-space field should be real, hermitianize the field
        # components
        if real_signal:
            result_val_list = [harmonic_domain.hermitian_decomposition(
                                    x.val,
                                    axes=x.domain_axes[power_space_index])[0]
                               for x in result_list]
        else:
            result_val_list = [x.val for x in result_list]

        # weight the random fields with the power spectrum
        # therefore get the pindex from the power space
        pindex = power_domain.pindex
        # take the local data from pindex. This data must be compatible to the
        # local data of the field given the slice of the PowerSpace
        local_distribution_strategy = \
            result_list[0].val.get_axes_local_distribution_strategy(
                result_list[0].domain_axes[power_space_index])

        if pindex.distribution_strategy is not local_distribution_strategy:
365
            self.logger.warn(
366
                "The distribution_stragey of pindex does not fit the "
367
368
369
370
371
372
                "slice_local distribution strategy of the synthesized field.")

        # Now use numpy advanced indexing in order to put the entries of the
        # power spectrum into the appropriate places of the pindex array.
        # Do this for every 'pindex-slice' in parallel using the 'slice(None)'s
        local_pindex = pindex.get_local_data(copy=False)
373
        full_spec = self.val.get_full_data()
374
375
376
377
378

        local_blow_up = [slice(None)]*len(self.shape)
        local_blow_up[self.domain_axes[power_space_index][0]] = local_pindex

        # here, the power_spectrum is distributed into the new shape
379
        local_rescaler = full_spec[local_blow_up]
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400

        # apply the rescaler to the random fields
        result_val_list[0].apply_scalar_function(
                                            lambda x: x * local_rescaler.real,
                                            inplace=True)

        if issubclass(power_domain.dtype.type, np.complexfloating):
            result_val_list[1].apply_scalar_function(
                                            lambda x: x * local_rescaler.imag,
                                            inplace=True)

        # store the result into the fields
        [x.set_val(new_val=y, copy=False) for x, y in
            zip(result_list, result_val_list)]

        if issubclass(power_domain.dtype.type, np.complexfloating):
            result = result_list[0] + 1j*result_list[1]
        else:
            result = result_list[0]

        return result
401

Theo Steininger's avatar
Theo Steininger committed
402
    # ---Properties---
403

Theo Steininger's avatar
Theo Steininger committed
404
    def set_val(self, new_val=None, copy=False):
405
406
        new_val = self.cast(new_val)
        if copy:
Theo Steininger's avatar
Theo Steininger committed
407
408
            new_val = new_val.copy()
        self._val = new_val
409
        return self
csongor's avatar
csongor committed
410

411
    def get_val(self, copy=False):
412
413
414
        if self._val is None:
            self.set_val(None)

415
        if copy:
Theo Steininger's avatar
Theo Steininger committed
416
            return self._val.copy()
417
        else:
Theo Steininger's avatar
Theo Steininger committed
418
            return self._val
csongor's avatar
csongor committed
419

Theo Steininger's avatar
Theo Steininger committed
420
421
    @property
    def val(self):
422
        return self.get_val(copy=False)
csongor's avatar
csongor committed
423

Theo Steininger's avatar
Theo Steininger committed
424
425
    @val.setter
    def val(self, new_val):
426
        self.set_val(new_val=new_val, copy=False)
csongor's avatar
csongor committed
427

428
429
    @property
    def shape(self):
430
        shape_tuple = tuple(sp.shape for sp in self.domain)
431
432
433
434
        try:
            global_shape = reduce(lambda x, y: x + y, shape_tuple)
        except TypeError:
            global_shape = ()
csongor's avatar
csongor committed
435

436
        return global_shape
csongor's avatar
csongor committed
437

438
439
    @property
    def dim(self):
440
        dim_tuple = tuple(sp.dim for sp in self.domain)
Theo Steininger's avatar
Theo Steininger committed
441
442
443
444
        try:
            return reduce(lambda x, y: x * y, dim_tuple)
        except TypeError:
            return 0
csongor's avatar
csongor committed
445

446
447
    @property
    def dof(self):
Theo Steininger's avatar
Theo Steininger committed
448
449
450
451
452
453
454
455
        dof = self.dim
        if issubclass(self.dtype.type, np.complexfloating):
            dof *= 2
        return dof

    @property
    def total_volume(self):
        volume_tuple = tuple(sp.total_volume for sp in self.domain)
456
        try:
Theo Steininger's avatar
Theo Steininger committed
457
            return reduce(lambda x, y: x * y, volume_tuple)
458
        except TypeError:
Theo Steininger's avatar
Theo Steininger committed
459
            return 0
460

Theo Steininger's avatar
Theo Steininger committed
461
    # ---Special unary/binary operations---
462

csongor's avatar
csongor committed
463
464
465
    def cast(self, x=None, dtype=None):
        if dtype is None:
            dtype = self.dtype
466
467
        else:
            dtype = np.dtype(dtype)
468

469
470
        casted_x = x

471
        for ind, sp in enumerate(self.domain):
472
            casted_x = sp.pre_cast(casted_x,
473
474
475
                                   axes=self.domain_axes[ind])

        casted_x = self._actual_cast(casted_x, dtype=dtype)
476
477

        for ind, sp in enumerate(self.domain):
478
479
            casted_x = sp.post_cast(casted_x,
                                    axes=self.domain_axes[ind])
480

481
        return casted_x
csongor's avatar
csongor committed
482

Theo Steininger's avatar
Theo Steininger committed
483
    def _actual_cast(self, x, dtype=None):
484
        if isinstance(x, Field):
csongor's avatar
csongor committed
485
486
487
488
489
            x = x.get_val()

        if dtype is None:
            dtype = self.dtype

490
        return_x = distributed_data_object(
491
492
493
                            global_shape=self.shape,
                            dtype=dtype,
                            distribution_strategy=self.distribution_strategy)
494
495
        return_x.set_full_data(x, copy=False)
        return return_x
Theo Steininger's avatar
Theo Steininger committed
496

497
    def copy(self, domain=None, dtype=None, distribution_strategy=None):
Theo Steininger's avatar
Theo Steininger committed
498
        copied_val = self.get_val(copy=True)
499
500
501
502
        new_field = self.copy_empty(
                                domain=domain,
                                dtype=dtype,
                                distribution_strategy=distribution_strategy)
Theo Steininger's avatar
Theo Steininger committed
503
504
        new_field.set_val(new_val=copied_val, copy=False)
        return new_field
csongor's avatar
csongor committed
505

506
    def copy_empty(self, domain=None, dtype=None, distribution_strategy=None):
Theo Steininger's avatar
Theo Steininger committed
507
508
        if domain is None:
            domain = self.domain
csongor's avatar
csongor committed
509
        else:
Theo Steininger's avatar
Theo Steininger committed
510
            domain = self._parse_domain(domain)
csongor's avatar
csongor committed
511

Theo Steininger's avatar
Theo Steininger committed
512
513
514
515
        if dtype is None:
            dtype = self.dtype
        else:
            dtype = np.dtype(dtype)
csongor's avatar
csongor committed
516

517
518
        if distribution_strategy is None:
            distribution_strategy = self.distribution_strategy
csongor's avatar
csongor committed
519

Theo Steininger's avatar
Theo Steininger committed
520
521
522
523
524
525
526
527
528
529
        fast_copyable = True
        try:
            for i in xrange(len(self.domain)):
                if self.domain[i] is not domain[i]:
                    fast_copyable = False
                    break
        except IndexError:
            fast_copyable = False

        if (fast_copyable and dtype == self.dtype and
530
                distribution_strategy == self.distribution_strategy):
Theo Steininger's avatar
Theo Steininger committed
531
532
533
534
            new_field = self._fast_copy_empty()
        else:
            new_field = Field(domain=domain,
                              dtype=dtype,
535
                              distribution_strategy=distribution_strategy)
Theo Steininger's avatar
Theo Steininger committed
536
        return new_field
csongor's avatar
csongor committed
537

Theo Steininger's avatar
Theo Steininger committed
538
539
540
541
542
543
544
    def _fast_copy_empty(self):
        # make an empty field
        new_field = EmptyField()
        # repair its class
        new_field.__class__ = self.__class__
        # copy domain, codomain and val
        for key, value in self.__dict__.items():
545
            if key != '_val':
Theo Steininger's avatar
Theo Steininger committed
546
547
548
549
550
551
                new_field.__dict__[key] = value
            else:
                new_field.__dict__[key] = self.val.copy_empty()
        return new_field

    def weight(self, power=1, inplace=False, spaces=None):
552
        if inplace:
csongor's avatar
csongor committed
553
554
555
556
            new_field = self
        else:
            new_field = self.copy_empty()

557
        new_val = self.get_val(copy=False)
csongor's avatar
csongor committed
558

csongor's avatar
csongor committed
559
        if spaces is None:
Theo Steininger's avatar
Theo Steininger committed
560
561
562
            spaces = range(len(self.domain))
        else:
            spaces = utilities.cast_axis_to_tuple(spaces, len(self.domain))
csongor's avatar
csongor committed
563

564
        for ind, sp in enumerate(self.domain):
Theo Steininger's avatar
Theo Steininger committed
565
566
567
568
569
            if ind in spaces:
                new_val = sp.weight(new_val,
                                    power=power,
                                    axes=self.domain_axes[ind],
                                    inplace=inplace)
570
571

        new_field.set_val(new_val=new_val, copy=False)
csongor's avatar
csongor committed
572
573
        return new_field

Theo Steininger's avatar
Theo Steininger committed
574
575
576
577
578
579
580
    def dot(self, x=None, bare=False):
        if isinstance(x, Field):
            try:
                assert len(x.domain) == len(self.domain)
                for index in xrange(len(self.domain)):
                    assert x.domain[index] == self.domain[index]
            except AssertionError:
581
582
                raise ValueError(
                    "domains are incompatible.")
Theo Steininger's avatar
Theo Steininger committed
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
            # extract the data from x and try to dot with this
            x = x.get_val(copy=False)

        # Compute the dot respecting the fact of discrete/continous spaces
        if bare:
            y = self
        else:
            y = self.weight(power=1)

        y = y.get_val(copy=False)

        # Cast the input in order to cure dtype and shape differences
        x = self.cast(x)

        dotted = x.conjugate() * y

        return dotted.sum()

601
    def norm(self, q=2):
csongor's avatar
csongor committed
602
603
604
605
606
607
608
609
610
611
612
613
614
615
        """
            Computes the Lq-norm of the field values.

            Parameters
            ----------
            q : scalar
                Parameter q of the Lq-norm (default: 2).

            Returns
            -------
            norm : scalar
                The Lq-norm of the field values.

        """
616
        if q == 2:
617
            return (self.dot(x=self)) ** (1 / 2)
csongor's avatar
csongor committed
618
        else:
619
            return self.dot(x=self ** (q - 1)) ** (1 / q)
csongor's avatar
csongor committed
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635

    def conjugate(self, inplace=False):
        """
            Computes the complex conjugate of the field.

            Returns
            -------
            cc : field
                The complex conjugated field.

        """
        if inplace:
            work_field = self
        else:
            work_field = self.copy_empty()

636
        new_val = self.get_val(copy=False)
Theo Steininger's avatar
Theo Steininger committed
637
        new_val = new_val.conjugate()
638
        work_field.set_val(new_val=new_val, copy=False)
csongor's avatar
csongor committed
639
640
641

        return work_field

Theo Steininger's avatar
Theo Steininger committed
642
    # ---General unary/contraction methods---
643

Theo Steininger's avatar
Theo Steininger committed
644
645
    def __pos__(self):
        return self.copy()
646

Theo Steininger's avatar
Theo Steininger committed
647
648
649
650
    def __neg__(self):
        return_field = self.copy_empty()
        new_val = -self.get_val(copy=False)
        return_field.set_val(new_val, copy=False)
csongor's avatar
csongor committed
651
652
        return return_field

Theo Steininger's avatar
Theo Steininger committed
653
654
655
656
657
    def __abs__(self):
        return_field = self.copy_empty()
        new_val = abs(self.get_val(copy=False))
        return_field.set_val(new_val, copy=False)
        return return_field
csongor's avatar
csongor committed
658

659
    def _contraction_helper(self, op, spaces):
Theo Steininger's avatar
Theo Steininger committed
660
661
662
663
664
        # build a list of all axes
        if spaces is None:
            spaces = xrange(len(self.domain))
        else:
            spaces = utilities.cast_axis_to_tuple(spaces, len(self.domain))
csongor's avatar
csongor committed
665

666
        axes_list = tuple(self.domain_axes[sp_index] for sp_index in spaces)
667
668

        try:
Theo Steininger's avatar
Theo Steininger committed
669
            axes_list = reduce(lambda x, y: x+y, axes_list)
670
        except TypeError:
Theo Steininger's avatar
Theo Steininger committed
671
            axes_list = ()
csongor's avatar
csongor committed
672

Theo Steininger's avatar
Theo Steininger committed
673
674
675
        # perform the contraction on the d2o
        data = self.get_val(copy=False)
        data = getattr(data, op)(axis=axes_list)
csongor's avatar
csongor committed
676

Theo Steininger's avatar
Theo Steininger committed
677
678
679
        # check if the result is scalar or if a result_field must be constr.
        if np.isscalar(data):
            return data
csongor's avatar
csongor committed
680
        else:
Theo Steininger's avatar
Theo Steininger committed
681
682
683
            return_domain = tuple(self.domain[i]
                                  for i in xrange(len(self.domain))
                                  if i not in spaces)
684

Theo Steininger's avatar
Theo Steininger committed
685
686
687
688
            return_field = Field(domain=return_domain,
                                 val=data,
                                 copy=False)
            return return_field
csongor's avatar
csongor committed
689

690
691
    def sum(self, spaces=None):
        return self._contraction_helper('sum', spaces)
csongor's avatar
csongor committed
692

693
694
    def prod(self, spaces=None):
        return self._contraction_helper('prod', spaces)
csongor's avatar
csongor committed
695

696
697
    def all(self, spaces=None):
        return self._contraction_helper('all', spaces)
csongor's avatar
csongor committed
698

699
700
    def any(self, spaces=None):
        return self._contraction_helper('any', spaces)
csongor's avatar
csongor committed
701

702
703
    def min(self, spaces=None):
        return self._contraction_helper('min', spaces)
csongor's avatar
csongor committed
704

705
706
    def nanmin(self, spaces=None):
        return self._contraction_helper('nanmin', spaces)
csongor's avatar
csongor committed
707

708
709
    def max(self, spaces=None):
        return self._contraction_helper('max', spaces)
csongor's avatar
csongor committed
710

711
712
    def nanmax(self, spaces=None):
        return self._contraction_helper('nanmax', spaces)
csongor's avatar
csongor committed
713

714
715
    def mean(self, spaces=None):
        return self._contraction_helper('mean', spaces)
csongor's avatar
csongor committed
716

717
718
    def var(self, spaces=None):
        return self._contraction_helper('var', spaces)
csongor's avatar
csongor committed
719

720
721
    def std(self, spaces=None):
        return self._contraction_helper('std', spaces)
csongor's avatar
csongor committed
722

Theo Steininger's avatar
Theo Steininger committed
723
    # ---General binary methods---
csongor's avatar
csongor committed
724

Theo Steininger's avatar
Theo Steininger committed
725
    def _binary_helper(self, other, op, inplace=False):
csongor's avatar
csongor committed
726
        # if other is a field, make sure that the domains match
727
        if isinstance(other, Field):
Theo Steininger's avatar
Theo Steininger committed
728
729
730
731
732
            try:
                assert len(other.domain) == len(self.domain)
                for index in xrange(len(self.domain)):
                    assert other.domain[index] == self.domain[index]
            except AssertionError:
733
734
                raise ValueError(
                    "domains are incompatible.")
Theo Steininger's avatar
Theo Steininger committed
735
            other = other.get_val(copy=False)
csongor's avatar
csongor committed
736

Theo Steininger's avatar
Theo Steininger committed
737
738
        self_val = self.get_val(copy=False)
        return_val = getattr(self_val, op)(other)
csongor's avatar
csongor committed
739
740
741
742

        if inplace:
            working_field = self
        else:
743
            working_field = self.copy_empty(dtype=return_val.dtype)
csongor's avatar
csongor committed
744

Theo Steininger's avatar
Theo Steininger committed
745
        working_field.set_val(return_val, copy=False)
csongor's avatar
csongor committed
746
747
748
        return working_field

    def __add__(self, other):
Theo Steininger's avatar
Theo Steininger committed
749
        return self._binary_helper(other, op='__add__')
750

751
    def __radd__(self, other):
Theo Steininger's avatar
Theo Steininger committed
752
        return self._binary_helper(other, op='__radd__')
csongor's avatar
csongor committed
753
754

    def __iadd__(self, other):
Theo Steininger's avatar
Theo Steininger committed
755
        return self._binary_helper(other, op='__iadd__', inplace=True)
csongor's avatar
csongor committed
756
757

    def __sub__(self, other):
Theo Steininger's avatar
Theo Steininger committed
758
        return self._binary_helper(other, op='__sub__')
csongor's avatar
csongor committed
759
760

    def __rsub__(self, other):
Theo Steininger's avatar
Theo Steininger committed
761
        return self._binary_helper(other, op='__rsub__')
csongor's avatar
csongor committed
762
763

    def __isub__(self, other):
Theo Steininger's avatar
Theo Steininger committed
764
        return self._binary_helper(other, op='__isub__', inplace=True)
csongor's avatar
csongor committed
765
766

    def __mul__(self, other):
Theo Steininger's avatar
Theo Steininger committed
767
        return self._binary_helper(other, op='__mul__')
768

769
    def __rmul__(self, other):
Theo Steininger's avatar
Theo Steininger committed
770
        return self._binary_helper(other, op='__rmul__')
csongor's avatar
csongor committed
771
772

    def __imul__(self, other):
Theo Steininger's avatar
Theo Steininger committed
773
        return self._binary_helper(other, op='__imul__', inplace=True)
csongor's avatar
csongor committed
774
775

    def __div__(self, other):
Theo Steininger's avatar
Theo Steininger committed
776
        return self._binary_helper(other, op='__div__')
csongor's avatar
csongor committed
777
778

    def __rdiv__(self, other):
Theo Steininger's avatar
Theo Steininger committed
779
        return self._binary_helper(other, op='__rdiv__')
csongor's avatar
csongor committed
780
781

    def __idiv__(self, other):
Theo Steininger's avatar
Theo Steininger committed
782
        return self._binary_helper(other, op='__idiv__', inplace=True)
783

csongor's avatar
csongor committed
784
    def __pow__(self, other):
Theo Steininger's avatar
Theo Steininger committed
785
        return self._binary_helper(other, op='__pow__')
csongor's avatar
csongor committed
786
787

    def __rpow__(self, other):
Theo Steininger's avatar
Theo Steininger committed
788
        return self._binary_helper(other, op='__rpow__')
csongor's avatar
csongor committed
789
790

    def __ipow__(self, other):
Theo Steininger's avatar
Theo Steininger committed
791
        return self._binary_helper(other, op='__ipow__', inplace=True)
csongor's avatar
csongor committed
792
793

    def __lt__(self, other):
Theo Steininger's avatar
Theo Steininger committed
794
        return self._binary_helper(other, op='__lt__')
csongor's avatar
csongor committed
795
796

    def __le__(self, other):
Theo Steininger's avatar
Theo Steininger committed
797
        return self._binary_helper(other, op='__le__')
csongor's avatar
csongor committed
798
799
800
801
802

    def __ne__(self, other):
        if other is None:
            return True
        else:
Theo Steininger's avatar
Theo Steininger committed
803
            return self._binary_helper(other, op='__ne__')
csongor's avatar
csongor committed
804
805
806
807
808

    def __eq__(self, other):
        if other is None:
            return False
        else:
Theo Steininger's avatar
Theo Steininger committed
809
            return self._binary_helper(other, op='__eq__')
csongor's avatar
csongor committed
810
811

    def __ge__(self, other):
Theo Steininger's avatar
Theo Steininger committed
812
        return self._binary_helper(other, op='__ge__')
csongor's avatar
csongor committed
813
814

    def __gt__(self, other):
Theo Steininger's avatar
Theo Steininger committed
815
816
817
818
819
820
821
822
823
824
825
826
827
        return self._binary_helper(other, op='__gt__')

    def __repr__(self):
        return "<nifty_core.field>"

    def __str__(self):
        minmax = [self.min(), self.max()]
        mean = self.mean()
        return "nifty_core.field instance\n- domain      = " + \
               repr(self.domain) + \
               "\n- val         = " + repr(self.get_val()) + \
               "\n  - min.,max. = " + str(minmax) + \
               "\n  - mean = " + str(mean)
csongor's avatar
csongor committed
828

Jait Dixit's avatar
Jait Dixit committed
829
830
831
    # ---Serialization---

    def _to_hdf5(self, hdf5_group):
Theo Steininger's avatar
Theo Steininger committed
832
833
834
        hdf5_group.attrs['dtype'] = self.dtype.name
        hdf5_group.attrs['distribution_strategy'] = self.distribution_strategy
        hdf5_group.attrs['domain_axes'] = str(self.domain_axes)
835
        hdf5_group['num_domain'] = len(self.domain)
Jait Dixit's avatar
Jait Dixit committed
836

Theo Steininger's avatar
Theo Steininger committed
837
        ret_dict = {'val': self.val}
Jait Dixit's avatar
Jait Dixit committed
838
839
840
841
842
843
844

        for i in range(len(self.domain)):
            ret_dict['s_' + str(i)] = self.domain[i]

        return ret_dict

    @classmethod
Theo Steininger's avatar
Theo Steininger committed
845
    def _from_hdf5(cls, hdf5_group, repository):
Jait Dixit's avatar
Jait Dixit committed
846
847
848
849
850
851
        # create empty field
        new_field = EmptyField()
        # reset class
        new_field.__class__ = cls
        # set values
        temp_domain = []
852
        for i in range(hdf5_group['num_domain'][()]):
Theo Steininger's avatar
Theo Steininger committed
853
            temp_domain.append(repository.get('s_' + str(i), hdf5_group))
Jait Dixit's avatar
Jait Dixit committed
854
855
        new_field.domain = tuple(temp_domain)

Theo Steininger's avatar
Theo Steininger committed
856
857
858
859
860
        exec('new_field.domain_axes = ' + hdf5_group.attrs['domain_axes'])
        new_field._val = repository.get('val', hdf5_group)
        new_field.dtype = np.dtype(hdf5_group.attrs['dtype'])
        new_field.distribution_strategy =\
            hdf5_group.attrs['distribution_strategy']
Jait Dixit's avatar
Jait Dixit committed
861
862

        return new_field
863

Theo Steininger's avatar
Theo Steininger committed
864

865
class EmptyField(Field):
csongor's avatar
csongor committed
866
867
    def __init__(self):
        pass