1 | """ |
---|
2 | Sasview model constructor. |
---|
3 | |
---|
4 | Given a module defining an OpenCL kernel such as sasmodels.models.cylinder, |
---|
5 | create a sasview model class to run that kernel as follows:: |
---|
6 | |
---|
7 | from sasmodels.sasview_model import make_class |
---|
8 | from sasmodels.models import cylinder |
---|
9 | CylinderModel = make_class(cylinder, dtype='single') |
---|
10 | |
---|
11 | The model parameters for sasmodels are different from those in sasview. |
---|
12 | When reloading previously saved models, the parameters should be converted |
---|
13 | using :func:`sasmodels.convert.convert`. |
---|
14 | """ |
---|
15 | |
---|
16 | import math |
---|
17 | from copy import deepcopy |
---|
18 | import warnings |
---|
19 | |
---|
20 | import numpy as np |
---|
21 | |
---|
22 | from . import core |
---|
23 | |
---|
24 | def make_class(model_definition, dtype='single', namestyle='name'): |
---|
25 | """ |
---|
26 | Load the sasview model defined in *kernel_module*. |
---|
27 | |
---|
28 | Returns a class that can be used directly as a sasview model. |
---|
29 | |
---|
30 | Defaults to using the new name for a model. Setting |
---|
31 | *namestyle='oldname'* will produce a class with a name |
---|
32 | compatible with SasView. |
---|
33 | """ |
---|
34 | model = core.load_model(model_definition, dtype=dtype) |
---|
35 | def __init__(self, multfactor=1): |
---|
36 | SasviewModel.__init__(self, model) |
---|
37 | attrs = dict(__init__=__init__) |
---|
38 | ConstructedModel = type(model.info[namestyle], (SasviewModel,), attrs) |
---|
39 | return ConstructedModel |
---|
40 | |
---|
41 | class SasviewModel(object): |
---|
42 | """ |
---|
43 | Sasview wrapper for opencl/ctypes model. |
---|
44 | """ |
---|
45 | def __init__(self, model): |
---|
46 | """Initialization""" |
---|
47 | self._model = model |
---|
48 | |
---|
49 | self.name = model.info['name'] |
---|
50 | self.oldname = model.info['oldname'] |
---|
51 | self.description = model.info['description'] |
---|
52 | self.category = None |
---|
53 | self.multiplicity_info = None |
---|
54 | self.is_multifunc = False |
---|
55 | |
---|
56 | ## interpret the parameters |
---|
57 | ## TODO: reorganize parameter handling |
---|
58 | self.details = dict() |
---|
59 | self.params = dict() |
---|
60 | self.dispersion = dict() |
---|
61 | partype = model.info['partype'] |
---|
62 | for name, units, default, limits, _, _ in model.info['parameters']: |
---|
63 | self.params[name] = default |
---|
64 | self.details[name] = [units] + limits |
---|
65 | |
---|
66 | for name in partype['pd-2d']: |
---|
67 | self.dispersion[name] = { |
---|
68 | 'width': 0, |
---|
69 | 'npts': 35, |
---|
70 | 'nsigmas': 3, |
---|
71 | 'type': 'gaussian', |
---|
72 | } |
---|
73 | |
---|
74 | self.orientation_params = ( |
---|
75 | partype['orientation'] |
---|
76 | + [n + '.width' for n in partype['orientation']] |
---|
77 | + partype['magnetic']) |
---|
78 | self.magnetic_params = partype['magnetic'] |
---|
79 | self.fixed = [n + '.width' for n in partype['pd-2d']] |
---|
80 | self.non_fittable = [] |
---|
81 | |
---|
82 | ## independent parameter name and unit [string] |
---|
83 | self.input_name = model.info.get("input_name", "Q") |
---|
84 | self.input_unit = model.info.get("input_unit", "A^{-1}") |
---|
85 | self.output_name = model.info.get("output_name", "Intensity") |
---|
86 | self.output_unit = model.info.get("output_unit", "cm^{-1}") |
---|
87 | |
---|
88 | ## _persistency_dict is used by sas.perspectives.fitting.basepage |
---|
89 | ## to store dispersity reference. |
---|
90 | ## TODO: _persistency_dict to persistency_dict throughout sasview |
---|
91 | self._persistency_dict = {} |
---|
92 | |
---|
93 | ## New fields introduced for opencl rewrite |
---|
94 | self.cutoff = 1e-5 |
---|
95 | |
---|
96 | def __str__(self): |
---|
97 | """ |
---|
98 | :return: string representation |
---|
99 | """ |
---|
100 | return self.name |
---|
101 | |
---|
102 | def is_fittable(self, par_name): |
---|
103 | """ |
---|
104 | Check if a given parameter is fittable or not |
---|
105 | |
---|
106 | :param par_name: the parameter name to check |
---|
107 | """ |
---|
108 | return par_name.lower() in self.fixed |
---|
109 | #For the future |
---|
110 | #return self.params[str(par_name)].is_fittable() |
---|
111 | |
---|
112 | |
---|
113 | # pylint: disable=no-self-use |
---|
114 | def getProfile(self): |
---|
115 | """ |
---|
116 | Get SLD profile |
---|
117 | |
---|
118 | : return: (z, beta) where z is a list of depth of the transition points |
---|
119 | beta is a list of the corresponding SLD values |
---|
120 | """ |
---|
121 | return None, None |
---|
122 | |
---|
123 | def setParam(self, name, value): |
---|
124 | """ |
---|
125 | Set the value of a model parameter |
---|
126 | |
---|
127 | :param name: name of the parameter |
---|
128 | :param value: value of the parameter |
---|
129 | |
---|
130 | """ |
---|
131 | # Look for dispersion parameters |
---|
132 | toks = name.split('.') |
---|
133 | if len(toks) == 2: |
---|
134 | for item in self.dispersion.keys(): |
---|
135 | if item.lower() == toks[0].lower(): |
---|
136 | for par in self.dispersion[item]: |
---|
137 | if par.lower() == toks[1].lower(): |
---|
138 | self.dispersion[item][par] = value |
---|
139 | return |
---|
140 | else: |
---|
141 | # Look for standard parameter |
---|
142 | for item in self.params.keys(): |
---|
143 | if item.lower() == name.lower(): |
---|
144 | self.params[item] = value |
---|
145 | return |
---|
146 | |
---|
147 | raise ValueError("Model does not contain parameter %s" % name) |
---|
148 | |
---|
149 | def getParam(self, name): |
---|
150 | """ |
---|
151 | Set the value of a model parameter |
---|
152 | |
---|
153 | :param name: name of the parameter |
---|
154 | |
---|
155 | """ |
---|
156 | # Look for dispersion parameters |
---|
157 | toks = name.split('.') |
---|
158 | if len(toks) == 2: |
---|
159 | for item in self.dispersion.keys(): |
---|
160 | if item.lower() == toks[0].lower(): |
---|
161 | for par in self.dispersion[item]: |
---|
162 | if par.lower() == toks[1].lower(): |
---|
163 | return self.dispersion[item][par] |
---|
164 | else: |
---|
165 | # Look for standard parameter |
---|
166 | for item in self.params.keys(): |
---|
167 | if item.lower() == name.lower(): |
---|
168 | return self.params[item] |
---|
169 | |
---|
170 | raise ValueError("Model does not contain parameter %s" % name) |
---|
171 | |
---|
172 | def getParamList(self): |
---|
173 | """ |
---|
174 | Return a list of all available parameters for the model |
---|
175 | """ |
---|
176 | param_list = self.params.keys() |
---|
177 | # WARNING: Extending the list with the dispersion parameters |
---|
178 | param_list.extend(self.getDispParamList()) |
---|
179 | return param_list |
---|
180 | |
---|
181 | def getDispParamList(self): |
---|
182 | """ |
---|
183 | Return a list of all available parameters for the model |
---|
184 | """ |
---|
185 | # TODO: fix test so that parameter order doesn't matter |
---|
186 | ret = ['%s.%s' % (d.lower(), p) |
---|
187 | for d in self._model.info['partype']['pd-2d'] |
---|
188 | for p in ('npts', 'nsigmas', 'width')] |
---|
189 | #print(ret) |
---|
190 | return ret |
---|
191 | |
---|
192 | def clone(self): |
---|
193 | """ Return a identical copy of self """ |
---|
194 | return deepcopy(self) |
---|
195 | |
---|
196 | def run(self, x=0.0): |
---|
197 | """ |
---|
198 | Evaluate the model |
---|
199 | |
---|
200 | :param x: input q, or [q,phi] |
---|
201 | |
---|
202 | :return: scattering function P(q) |
---|
203 | |
---|
204 | **DEPRECATED**: use calculate_Iq instead |
---|
205 | """ |
---|
206 | if isinstance(x, (list, tuple)): |
---|
207 | # pylint: disable=unpacking-non-sequence |
---|
208 | q, phi = x |
---|
209 | return self.calculate_Iq([q * math.cos(phi)], |
---|
210 | [q * math.sin(phi)])[0] |
---|
211 | else: |
---|
212 | return self.calculate_Iq([float(x)])[0] |
---|
213 | |
---|
214 | |
---|
215 | def runXY(self, x=0.0): |
---|
216 | """ |
---|
217 | Evaluate the model in cartesian coordinates |
---|
218 | |
---|
219 | :param x: input q, or [qx, qy] |
---|
220 | |
---|
221 | :return: scattering function P(q) |
---|
222 | |
---|
223 | **DEPRECATED**: use calculate_Iq instead |
---|
224 | """ |
---|
225 | if isinstance(x, (list, tuple)): |
---|
226 | return self.calculate_Iq([float(x[0])], [float(x[1])])[0] |
---|
227 | else: |
---|
228 | return self.calculate_Iq([float(x)])[0] |
---|
229 | |
---|
230 | def evalDistribution(self, qdist): |
---|
231 | r""" |
---|
232 | Evaluate a distribution of q-values. |
---|
233 | |
---|
234 | :param qdist: array of q or a list of arrays [qx,qy] |
---|
235 | |
---|
236 | * For 1D, a numpy array is expected as input |
---|
237 | |
---|
238 | :: |
---|
239 | |
---|
240 | evalDistribution(q) |
---|
241 | |
---|
242 | where *q* is a numpy array. |
---|
243 | |
---|
244 | * For 2D, a list of *[qx,qy]* is expected with 1D arrays as input |
---|
245 | |
---|
246 | :: |
---|
247 | |
---|
248 | qx = [ qx[0], qx[1], qx[2], ....] |
---|
249 | qy = [ qy[0], qy[1], qy[2], ....] |
---|
250 | |
---|
251 | If the model is 1D only, then |
---|
252 | |
---|
253 | .. math:: |
---|
254 | |
---|
255 | q = \sqrt{q_x^2+q_y^2} |
---|
256 | |
---|
257 | """ |
---|
258 | if isinstance(qdist, (list, tuple)): |
---|
259 | # Check whether we have a list of ndarrays [qx,qy] |
---|
260 | qx, qy = qdist |
---|
261 | partype = self._model.info['partype'] |
---|
262 | if not partype['orientation'] and not partype['magnetic']: |
---|
263 | return self.calculate_Iq(np.sqrt(qx ** 2 + qy ** 2)) |
---|
264 | else: |
---|
265 | return self.calculate_Iq(qx, qy) |
---|
266 | |
---|
267 | elif isinstance(qdist, np.ndarray): |
---|
268 | # We have a simple 1D distribution of q-values |
---|
269 | return self.calculate_Iq(qdist) |
---|
270 | |
---|
271 | else: |
---|
272 | raise TypeError("evalDistribution expects q or [qx, qy], not %r" |
---|
273 | % type(qdist)) |
---|
274 | |
---|
275 | def calculate_Iq(self, *args): |
---|
276 | """ |
---|
277 | Calculate Iq for one set of q with the current parameters. |
---|
278 | |
---|
279 | If the model is 1D, use *q*. If 2D, use *qx*, *qy*. |
---|
280 | |
---|
281 | This should NOT be used for fitting since it copies the *q* vectors |
---|
282 | to the card for each evaluation. |
---|
283 | """ |
---|
284 | q_vectors = [np.asarray(q) for q in args] |
---|
285 | fn = self._model(q_vectors) |
---|
286 | pars = [self.params[v] for v in fn.fixed_pars] |
---|
287 | pd_pars = [self._get_weights(p) for p in fn.pd_pars] |
---|
288 | result = fn(pars, pd_pars, self.cutoff) |
---|
289 | fn.input.release() |
---|
290 | fn.release() |
---|
291 | return result |
---|
292 | |
---|
293 | def calculate_ER(self): |
---|
294 | """ |
---|
295 | Calculate the effective radius for P(q)*S(q) |
---|
296 | |
---|
297 | :return: the value of the effective radius |
---|
298 | """ |
---|
299 | ER = self._model.info.get('ER', None) |
---|
300 | if ER is None: |
---|
301 | return 1.0 |
---|
302 | else: |
---|
303 | values, weights = self._dispersion_mesh() |
---|
304 | fv = ER(*values) |
---|
305 | #print(values[0].shape, weights.shape, fv.shape) |
---|
306 | return np.sum(weights * fv) / np.sum(weights) |
---|
307 | |
---|
308 | def calculate_VR(self): |
---|
309 | """ |
---|
310 | Calculate the volf ratio for P(q)*S(q) |
---|
311 | |
---|
312 | :return: the value of the volf ratio |
---|
313 | """ |
---|
314 | VR = self._model.info.get('VR', None) |
---|
315 | if VR is None: |
---|
316 | return 1.0 |
---|
317 | else: |
---|
318 | values, weights = self._dispersion_mesh() |
---|
319 | whole, part = VR(*values) |
---|
320 | return np.sum(weights * part) / np.sum(weights * whole) |
---|
321 | |
---|
322 | def set_dispersion(self, parameter, dispersion): |
---|
323 | """ |
---|
324 | Set the dispersion object for a model parameter |
---|
325 | |
---|
326 | :param parameter: name of the parameter [string] |
---|
327 | :param dispersion: dispersion object of type Dispersion |
---|
328 | """ |
---|
329 | if parameter.lower() in (s.lower() for s in self.params.keys()): |
---|
330 | # TODO: Store the disperser object directly in the model. |
---|
331 | # The current method of creating one on the fly whenever it is |
---|
332 | # needed is kind of funky. |
---|
333 | # Note: can't seem to get disperser parameters from sasview |
---|
334 | # (1) Could create a sasview model that has not yet # been |
---|
335 | # converted, assign the disperser to one of its polydisperse |
---|
336 | # parameters, then retrieve the disperser parameters from the |
---|
337 | # sasview model. (2) Could write a disperser parameter retriever |
---|
338 | # in sasview. (3) Could modify sasview to use sasmodels.weights |
---|
339 | # dispersers. |
---|
340 | # For now, rely on the fact that the sasview only ever uses |
---|
341 | # new dispersers in the set_dispersion call and create a new |
---|
342 | # one instead of trying to assign parameters. |
---|
343 | from . import weights |
---|
344 | disperser = weights.dispersers[dispersion.__class__.__name__] |
---|
345 | dispersion = weights.models[disperser]() |
---|
346 | self.dispersion[parameter] = dispersion.get_pars() |
---|
347 | else: |
---|
348 | raise ValueError("%r is not a dispersity or orientation parameter") |
---|
349 | |
---|
350 | def _dispersion_mesh(self): |
---|
351 | """ |
---|
352 | Create a mesh grid of dispersion parameters and weights. |
---|
353 | |
---|
354 | Returns [p1,p2,...],w where pj is a vector of values for parameter j |
---|
355 | and w is a vector containing the products for weights for each |
---|
356 | parameter set in the vector. |
---|
357 | """ |
---|
358 | pars = self._model.info['partype']['volume'] |
---|
359 | return core.dispersion_mesh([self._get_weights(p) for p in pars]) |
---|
360 | |
---|
361 | def _get_weights(self, par): |
---|
362 | """ |
---|
363 | Return dispersion weights |
---|
364 | :param par parameter name |
---|
365 | """ |
---|
366 | from . import weights |
---|
367 | |
---|
368 | relative = self._model.info['partype']['pd-rel'] |
---|
369 | limits = self._model.info['limits'] |
---|
370 | dis = self.dispersion[par] |
---|
371 | value, weight = weights.get_weights( |
---|
372 | dis['type'], dis['npts'], dis['width'], dis['nsigmas'], |
---|
373 | self.params[par], limits[par], par in relative) |
---|
374 | return value, weight / np.sum(weight) |
---|
375 | |
---|