code.rst 15.9 KB
Newer Older
Philipp Arras's avatar
Philipp Arras committed
1
.. currentmodule:: nifty5
2

Martin Reinecke's avatar
Martin Reinecke committed
3
=============
4
5
Code Overview
=============
6

Martin Reinecke's avatar
Martin Reinecke committed
7
8
9
10

Executive summary
=================

11
12
The fundamental building blocks required for IFT computations are best
recognized from a large distance, ignoring all technical details.
13

14
15
From such a perspective,

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
16
- IFT problems largely consist of the combination of several high dimensional
17
  *minimization* problems.
Philipp Arras's avatar
Philipp Arras committed
18
- Within NIFTy, *operators* are used to define the characteristic equations and
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
19
  properties of the problems.
Martin Reinecke's avatar
Martin Reinecke committed
20
21
- The equations are built mostly from the application of *linear operators*,
  but there may also be nonlinear functions involved.
22
- The unknowns in the equations represent either continuous physical *fields*,
Martin Reinecke's avatar
Martin Reinecke committed
23
24
25
26
  or they are simply individual measured *data points*.
- Discretized *fields* have geometrical information (like locations and volume
  elements) associated with every entry; this information is called the field's
  *domain*.
27
28
29
30
31
32
33
34
35
36
37
38
39

In the following sections, the concepts briefly presented here will be
discussed in more detail; this is done in reversed order of their introduction,
to avoid forward references.


Domains
=======


Abstract base class
-------------------

Philipp Arras's avatar
Philipp Arras committed
40
One of the fundamental building blocks of the NIFTy5 framework is the *domain*.
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Its required capabilities are expressed by the abstract :class:`Domain` class.
A domain must be able to answer the following queries:

- its total number of data entries (pixels), which is accessible via the
  :attr:`~Domain.size` property
- the shape of the array that is supposed to hold these data entries
  (obtainable by means of the :attr:`~Domain.shape` property)
- equality comparison to another :class:`Domain` instance


Unstructured domains
--------------------

Domains can be either *structured* (i.e. there is geometrical information
associated with them, like position in space and volume factors),
or *unstructured* (meaning that the data points have no associated manifold).

Unstructured domains can be described by instances of NIFTy's
:class:`UnstructuredDomain` class.


Structured domains
------------------

In contrast to unstructured domains, these domains have an assigned geometry.
NIFTy requires them to provide the volume elements of their grid cells.
The additional methods are specified in the abstract class
:class:`StructuredDomain`:

Martin Reinecke's avatar
Martin Reinecke committed
70
- The properties :attr:`~StructuredDomain.scalar_dvol`,
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
  :attr:`~StructuredDomain.dvol`, and  :attr:`~StructuredDomain.total_volume`
  provide information about the domain's pixel volume(s) and its total volume.
- The property :attr:`~StructuredDomain.harmonic` specifies whether a domain
  is harmonic (i.e. describes a frequency space) or not
- Iff the domain is harmonic, the methods
  :meth:`~StructuredDomain.get_k_length_array`,
  :meth:`~StructuredDomain.get_unique_k_lengths`, and
  :meth:`~StructuredDomain.get_fft_smoothing_kernel_function` provide absolute
  distances of the individual grid cells from the origin and assist with
  Gaussian convolution.

NIFTy comes with several concrete subclasses of :class:`StructuredDomain`:

- :class:`RGSpace` represents a regular Cartesian grid with an arbitrary
  number of dimensions, which is supposed to be periodic in each dimension.
- :class:`HPSpace` and :class:`GLSpace` describe pixelisations of the
  2-sphere; their counterpart in harmonic space is :class:`LMSpace`, which
  contains spherical harmonic coefficients.
- :class:`PowerSpace` is used to describe one-dimensional power spectra.

Among these, :class:`RGSpace` can be harmonic or not (depending on constructor arguments), :class:`GLSpace`, :class:`HPSpace`, and :class:`PowerSpace` are
pure position domains (i.e. nonharmonic), and :class:`LMSpace` is always
harmonic.


Combinations of domains
=======================

The fundamental classes described above are often sufficient to specify the
100
101
domain of a field. In some cases, however, it will be necessary to define the
field on a product of elementary domains instead of a single one.
Philipp Arras's avatar
Philipp Arras committed
102
More sophisticated operators also require a set of several such fields.
103
104
105
106
107
Some examples are:

- sky emission depending on location and energy. This could be represented by
  a product of an :class:`HPSpace` (for location) with an :class:`RGSpace`
  (for energy).
Philipp Arras's avatar
Philipp Arras committed
108
- a polarized field, which could be modeled as a product of any structured
Martin Reinecke's avatar
Martin Reinecke committed
109
110
  domain (representing location) with a four-element
  :class:`UnstructuredDomain` holding Stokes I, Q, U and V components.
Martin Reinecke's avatar
Martin Reinecke committed
111
112
113
- a model for the sky emission, which holds both the current realization
  (on a harmonic domain) and a few inferred model parameters (e.g. on an
  unstructured grid).
114
115
116
117
118
119

Consequently, NIFTy defines a class called :class:`DomainTuple` holding
a sequence of :class:`Domain` objects, which is used to specify full field
domains. In principle, a :class:`DomainTuple` can even be empty, which implies
that the field living on it is a scalar.

Martin Reinecke's avatar
Martin Reinecke committed
120
121
122
123
A :class:`DomainTuple` supports iteration and indexing, and also provides the
properties :attr:`~DomainTuple.shape`, :attr:`~DomainTuple.size` in analogy to
the elementary :class:`Domain`.

Torsten Ensslin's avatar
Torsten Ensslin committed
124
An aggregation of several :class:`DomainTuple` s, each member identified by a
Martin Reinecke's avatar
Martin Reinecke committed
125
name, is described by the :class:`MultiDomain` class.
126
127
128
129

Fields
======

Martin Reinecke's avatar
Martin Reinecke committed
130
131
132
Fields on a single DomainTuple
------------------------------

133
134
135
136
137
138
A :class:`Field` object consists of the following components:

- a domain in form of a :class:`DomainTuple` object
- a data type (e.g. numpy.float64)
- an array containing the actual values

Martin Reinecke's avatar
Martin Reinecke committed
139
Usually, the array is stored in the form of a ``numpy.ndarray``, but for very
Martin Reinecke's avatar
Martin Reinecke committed
140
141
142
143
144
145
146
147
148
149
150
151
152
153
resource-intensive tasks NIFTy also provides an alternative storage method to
be used with distributed memory processing.

Fields support a wide range of arithmetic operations, either involving two
fields with equal domains, or a field and a scalar.
Contractions (like summation, integration, minimum/maximum, computation of
statistical moments) can be carried out either over an entire field (producing
a scalar result) or over sub-domains (resulting in a field living on a smaller
domain). Scalar products of two fields can also be computed easily.

There is also a set of convenience functions to generate fields with constant
values or fields filled with random numbers according to a user-specified
distribution.

Martin Reinecke's avatar
Martin Reinecke committed
154
155
156
157
158
Like almost all NIFTy objects, fields are immutable: their value or any other
attribute cannot be modified after construction. To manipulate a field in ways
that are not covered by the provided standard operations, its data content must
be extracted first, then changed, and a new field has to be created from the
result.
Martin Reinecke's avatar
Martin Reinecke committed
159

160
Fields defined on a MultiDomain
Martin Reinecke's avatar
Martin Reinecke committed
161
-------------------------------
Martin Reinecke's avatar
Martin Reinecke committed
162
163

The :class:`MultiField` class can be seen as a dictionary of individual
164
:class:`Field` s, each identified by a name, which is defined on a
Martin Reinecke's avatar
Martin Reinecke committed
165
166
167
168
169
170
171
172
:class:`MultiDomain`.


Operators
=========

All transformations between different NIFTy fields are expressed  (explicitly
or implicitly) in the form of :class:`Operator` objects. The interface of this
Torsten Ensslin's avatar
Torsten Ensslin committed
173
class is very minimalistic: it has a property called :class:`domain` which returns
174
a :class:`DomainTuple` or :class:`MultiDomain` object specifying the structure of the
Torsten Ensslin's avatar
Torsten Ensslin committed
175
:class:`Field` s or :class:`MultiField` s it expects as input, another property :class:`target`
Martin Reinecke's avatar
Martin Reinecke committed
176
177
178
describing its output, and finally an overloaded `apply` method, which can
take

Martin Reinecke's avatar
Martin Reinecke committed
179
- a :class:`Field`/:class:`MultiField` object, in which case it returns the transformed
Torsten Ensslin's avatar
Torsten Ensslin committed
180
181
182
  :class:`Field`/:class:`MultiField`
- a :class:`Linearization` object, in which case it returns the transformed
  :class:`Linearization`
Martin Reinecke's avatar
Martin Reinecke committed
183

Torsten Ensslin's avatar
Torsten Ensslin committed
184
185
This is the interface that all objects derived from :class:`Operator` must implement.
In addition, :class:`Operator` objects can be added/subtracted, multiplied, chained
Martin Reinecke's avatar
Martin Reinecke committed
186
187
(via the :class:`__call__` method and the `@` operator) and support pointwise
application of functions like
Torsten Ensslin's avatar
Torsten Ensslin committed
188
:class:`exp()`, :class:`log()`, :class:`sqrt()`, :class:`conjugate()` etc.
Martin Reinecke's avatar
Martin Reinecke committed
189

190

Martin Reinecke's avatar
Martin Reinecke committed
191
192
193
194
195
196
197
198
199
200
201
Advanced operators
------------------

NIFTy provides a library of more sophisticated operators which are used for more
specific inference problems. Currently these are:

- :class:`AmplitudeOperator`, which returns a smooth power spectrum.
- :class:`InverseGammaOperator`, which models point sources which follow a inverse gamma distribution.
- :class:`CorrelatedField`, which models a diffuse log-normal field. It takes an amplitude operator to specify the correlation structure of the field.


202
203
204
Linear Operators
================

Philipp Arras's avatar
Philipp Arras committed
205
A linear operator (represented by NIFTy5's abstract :class:`LinearOperator`
Martin Reinecke's avatar
Martin Reinecke committed
206
207
class) is derived from `Operator` and can be interpreted as an
(implicitly defined) matrix. Since its operation is linear, it can provide some
Torsten Ensslin's avatar
Torsten Ensslin committed
208
additional functionality which is not available for the more generic :class:`Operator`
Martin Reinecke's avatar
Martin Reinecke committed
209
class.
210

Martin Reinecke's avatar
Martin Reinecke committed
211

Martin Reinecke's avatar
Martin Reinecke committed
212
213
Linear Operator basics
----------------------
Martin Reinecke's avatar
Martin Reinecke committed
214

215
216
There are four basic ways of applying an operator :math:`A` to a field :math:`f`:

Martin Reinecke's avatar
Martin Reinecke committed
217
218
219
220
- direct application: :math:`A\cdot f`
- adjoint application: :math:`A^\dagger \cdot f`
- inverse application: :math:`A^{-1}\cdot f`
- adjoint inverse application: :math:`(A^\dagger)^{-1}\cdot f`
221

Martin Reinecke's avatar
Martin Reinecke committed
222
223
(Because of the linearity, inverse adjoint and adjoint inverse application
are equivalent.)
224

Martin Reinecke's avatar
Martin Reinecke committed
225
226
These different actions of a linear operator ``Op`` on a field ``f`` can be
invoked in various ways:
227
228
229
230
231
232
233
234
235
236
237
238
239

- direct multiplication: ``Op(f)`` or ``Op.times(f)`` or ``Op.apply(f, Op.TIMES)``
- adjoint multiplication: ``Op.adjoint_times(f)`` or ``Op.apply(f, Op.ADJOINT_TIMES)``
- inverse multiplication: ``Op.inverse_times(f)`` or ``Op.apply(f, Op.INVERSE_TIMES)``
- adjoint inverse multiplication: ``Op.adjoint_inverse_times(f)`` or ``Op.apply(f, Op.ADJOINT_INVERSE_TIMES)``

Operator classes defined in NIFTy may implement an arbitrary subset of these
four operations. This subset can be queried using the
:attr:`~LinearOperator.capability` property.

If needed, the set of supported operations can be enhanced by iterative
inversion methods;
for example, an operator defining direct and adjoint multiplication could be
Martin Reinecke's avatar
Martin Reinecke committed
240
enhanced by this approach to support the complete set. This functionality is
241
242
243
244
245
246
247
248
249
250
251
252
253
254
provided by NIFTy's :class:`InversionEnabler` class, which is itself a linear
operator.

Direct multiplication and adjoint inverse multiplication transform a field
living on the operator's :attr:`~LinearOperator.domain` to one living on the operator's :attr:`~LinearOperator.target`, whereas adjoint multiplication
and inverse multiplication transform from :attr:`~LinearOperator.target` to :attr:`~LinearOperator.domain`.

Operators with identical domain and target can be derived from
:class:`EndomorphicOperator`; typical examples for this category are the :class:`ScalingOperator`, which simply multiplies its input by a scalar
value, and :class:`DiagonalOperator`, which multiplies every value of its input
field with potentially different values.

Further operator classes provided by NIFTy are

Martin Reinecke's avatar
Martin Reinecke committed
255
256
- :class:`HarmonicTransformOperator` for transforms from a harmonic domain to
  its counterpart in position space, and their adjoint
257
- :class:`PowerDistributor` for transforms from a :class:`PowerSpace` to
Martin Reinecke's avatar
Martin Reinecke committed
258
  an associated harmonic domain, and their adjoint
259
260
261
262
- :class:`GeometryRemover`, which transforms from structured domains to
  unstructured ones. This is typically needed when building instrument response
  operators.

Martin Reinecke's avatar
Martin Reinecke committed
263

Martin Reinecke's avatar
Martin Reinecke committed
264
265
Syntactic sugar
---------------
Martin Reinecke's avatar
Martin Reinecke committed
266

Martin Reinecke's avatar
Martin Reinecke committed
267
Nifty5 allows simple and intuitive construction of altered and combined
268
269
operators.
As an example, if ``A``, ``B`` and ``C`` are of type :class:`LinearOperator`
Martin Reinecke's avatar
Martin Reinecke committed
270
and ``f1`` and ``f2`` are of type :class:`Field`, writing::
271

Martin Reinecke's avatar
Martin Reinecke committed
272
    X = A(B.inverse(A.adjoint)) + C
273
274
275
276
277
278
279
280
281
282
283
284
285
    f2 = X(f1)

will perform the operation suggested intuitively by the notation, checking
domain compatibility while building the composed operator.
The combined operator infers its domain and target from its constituents,
as well as the set of operations it can support.
The properties :attr:`~LinearOperator.adjoint` and
:attr:`~LinearOperator.inverse` return a new operator which behaves as if it
were the original operator's adjoint or inverse, respectively.


.. _minimization:

Martin Reinecke's avatar
Martin Reinecke committed
286

287
288
289
Minimization
============

Martin Reinecke's avatar
Martin Reinecke committed
290
291
292
293
294
295
Most problems in IFT are solved by (possibly nested) minimizations of
high-dimensional functions, which are often nonlinear.


Energy functionals
------------------
296

Philipp Arras's avatar
Philipp Arras committed
297
In NIFTy5 such functions are represented by objects of type :class:`Energy`.
Martin Reinecke's avatar
Martin Reinecke committed
298
299
These hold the prescription how to calculate the function's
:attr:`~Energy.value`, :attr:`~Energy.gradient` and
Martin Reinecke's avatar
Martin Reinecke committed
300
(optionally) :attr:`~Energy.metric` at any given :attr:`~Energy.position`
Martin Reinecke's avatar
Martin Reinecke committed
301
in parameter space.
Martin Reinecke's avatar
Martin Reinecke committed
302
Function values are floating-point scalars, gradients have the form of fields
Martin Reinecke's avatar
Martin Reinecke committed
303
living on the energy's position domain, and metrics are represented by
Martin Reinecke's avatar
Martin Reinecke committed
304
linear operator objects.
305

Martin Reinecke's avatar
Martin Reinecke committed
306
307
Energies are classes that typically have to be provided by the user when
tackling new IFT problems.
Philipp Arras's avatar
Philipp Arras committed
308
Some examples of concrete energy classes delivered with NIFTy5 are
Martin Reinecke's avatar
Martin Reinecke committed
309
:class:`QuadraticEnergy` (with position-independent metric, mainly used with
Philipp Arras's avatar
Philipp Arras committed
310
conjugate gradient minimization) and :class:`~nifty5.library.WienerFilterEnergy`.
311

Martin Reinecke's avatar
Martin Reinecke committed
312
313
314

Iteration control
-----------------
Martin Reinecke's avatar
Martin Reinecke committed
315
316

Iterative minimization of an energy reqires some means of
Martin Reinecke's avatar
Martin Reinecke committed
317
checking the quality of the current solution estimate and stopping once
Martin Reinecke's avatar
Martin Reinecke committed
318
319
320
it is sufficiently accurate. In case of numerical problems, the iteration needs
to be terminated as well, returning a suitable error description.

Philipp Arras's avatar
Philipp Arras committed
321
In NIFTy5, this functionality is encapsulated in the abstract
Martin Reinecke's avatar
Martin Reinecke committed
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
:class:`IterationController` class, which is provided with the initial energy
object before starting the minimization, and is updated with the improved
energy after every iteration. Based on this information, it can either continue
the minimization or return the current estimate indicating convergence or
failure.

Sensible stopping criteria can vary significantly with the problem being
solved; NIFTy provides one concrete sub-class of :class:`IterationController`
called :class:`GradientNormController`, which should be appropriate in many
circumstances, but users have complete freedom to implement custom sub-classes
for their specific applications.


Minimization algorithms
-----------------------

Martin Reinecke's avatar
Martin Reinecke committed
338
339
340
341
342
343
344
345
346
347
348
349
350
All minimization algorithms in NIFTy inherit from the abstract
:class:`Minimizer` class, which presents a minimalistic interface consisting
only of a :meth:`~Minimizer.__call__` method taking an :class:`Energy` object
and optionally a preconditioning operator, and returning the energy at the
discovered minimum and a status code.

For energies with a quadratic form (i.e. which
can be expressed by means of a :class:`QuadraticEnergy` object), an obvious
choice of algorithm is the :class:`ConjugateGradient` minimizer.

A similar algorithm suited for nonlinear problems is provided by
:class:`NonlinearCG`.

Martin Reinecke's avatar
Martin Reinecke committed
351
Many minimizers for nonlinear problems can be characterized as
Martin Reinecke's avatar
Martin Reinecke committed
352
353
354
355
356
357
358
359
360

- first deciding on a direction for the next step
- then finding a suitable step length along this direction, resulting in the
  next energy estimate.

This family of algorithms is encapsulated in NIFTy's :class:`DescentMinimizer`
class, which currently has three concrete implementations:
:class:`SteepestDescent`, :class:`VL_BFGS`, and :class:`RelaxedNewton`.
Of these algorithms, only :class:`RelaxedNewton` requires the energy object to
Martin Reinecke's avatar
Martin Reinecke committed
361
provide a :attr:`~Energy.metric` property, the others only need energy
Martin Reinecke's avatar
Martin Reinecke committed
362
363
values and gradients.

Martin Reinecke's avatar
Martin Reinecke committed
364
365
366
367
368
The flexibility of NIFTy's design allows using externally provided
minimizers. With only small effort, adapters for two SciPy minimizers were
written; they are available under the names :class:`NewtonCG` and
:class:`L_BFGS_B`.

Martin Reinecke's avatar
Martin Reinecke committed
369
370
371
372

Application to operator inversion
---------------------------------

Martin Reinecke's avatar
Martin Reinecke committed
373
374
375
376
377
378
379
380
381
382
It is important to realize that the machinery presented here cannot only be
used for minimizing IFT Hamiltonians, but also for the numerical inversion of
linear operators, if the desired application mode is not directly available.
A classical example is the information propagator

:math:`D = \left(R^\dagger N^{-1} R + S^{-1}\right)^{-1}`,

which must be applied when calculating a Wiener filter. Only its inverse
application is straightforward; to use it in forward direction, we make use
of NIFTy's :class:`InversionEnabler` class, which internally performs a
Martin Reinecke's avatar
Martin Reinecke committed
383
minimization of a :class:`QuadraticEnergy` by means of the
Martin Reinecke's avatar
Martin Reinecke committed
384
:class:`ConjugateGradient` algorithm.