source: sasmodels/sasmodels/product.py @ 117c02a

core_shell_microgelsmagnetic_modelticket-1257-vesicle-productticket_1156ticket_1265_superballticket_822_more_unit_tests
Last change on this file since 117c02a was 117c02a, checked in by Torin Cooper-Bennun <torin.cooper-bennun@…>, 18 months ago

cherry-picking lazy results implementation from beta_approx_lazy_results, beta_approx_new_R_eff branches

  • Property mode set to 100644
File size: 15.0 KB
Line 
1"""
2Product model
3-------------
4
5The product model multiplies the structure factor by the form factor,
6modulated by the effective radius of the form.  The resulting model
7has a attributes of both the model description (with parameters, etc.)
8and the module evaluator (with call, release, etc.).
9
10To use it, first load form factor P and structure factor S, then create
11*make_product_info(P, S)*.
12"""
13from __future__ import print_function, division
14
15from collections import OrderedDict
16
17from copy import copy
18import numpy as np  # type: ignore
19
20from .modelinfo import ParameterTable, ModelInfo, Parameter
21from .kernel import KernelModel, Kernel
22from .details import make_details, dispersion_mesh
23
24# pylint: disable=unused-import
25try:
26    from typing import Tuple, Callable
27except ImportError:
28    pass
29else:
30    from .modelinfo import ParameterSet
31# pylint: enable=unused-import
32
33# TODO: make estimates available to constraints
34#ESTIMATED_PARAMETERS = [
35#    ["est_radius_effective", "A", 0.0, [0, np.inf], "", "Estimated effective radius"],
36#    ["est_volume_ratio", "", 1.0, [0, np.inf], "", "Estimated volume ratio"],
37#]
38
39ER_ID = "radius_effective"
40VF_ID = "volfraction"
41BETA_DEFINITION = ("beta_mode", ["P*S", "P*(1+beta*(S-1))"], 0, (None, None), "",
42                   "Structure factor dispersion calculation mode")
43def make_extra_pars(p_info):
44    pars = []
45    if p_info.have_Fq:
46        par = Parameter("structure_factor_mode", ["P*S","P*(1+beta*(S-1))"], 0, (None, None), "",
47                        "Structure factor calculation")
48        pars.append(par)
49    return pars
50
51# TODO: core_shell_sphere model has suppressed the volume ratio calculation
52# revert it after making VR and ER available at run time as constraints.
53def make_product_info(p_info, s_info):
54    # type: (ModelInfo, ModelInfo) -> ModelInfo
55    """
56    Create info block for product model.
57    """
58    # Make sure effective radius is the first parameter and
59    # make sure volume fraction is the second parameter of the
60    # structure factor calculator.  Structure factors should not
61    # have any magnetic parameters
62    if not len(s_info.parameters.kernel_parameters) >= 2:
63        raise TypeError("S needs {} and {} as its first parameters".format(ER_ID, VF_ID))
64    if not s_info.parameters.kernel_parameters[0].id == ER_ID:
65        raise TypeError("S needs {} as first parameter".format(ER_ID))
66    if not s_info.parameters.kernel_parameters[1].id == VF_ID:
67        raise TypeError("S needs {} as second parameter".format(VF_ID))
68    if not s_info.parameters.magnetism_index == []:
69        raise TypeError("S should not have SLD parameters")
70    p_id, p_name, p_pars = p_info.id, p_info.name, p_info.parameters
71    s_id, s_name, s_pars = s_info.id, s_info.name, s_info.parameters
72
73    # Create list of parameters for the combined model.  Skip the first
74    # parameter of S, which we verified above is effective radius.  If there
75    # are any names in P that overlap with those in S, modify the name in S
76    # to distinguish it.
77    p_set = set(p.id for p in p_pars.kernel_parameters)
78    s_list = [(_tag_parameter(par) if par.id in p_set else par)
79              for par in s_pars.kernel_parameters[1:]]
80    # Check if still a collision after renaming.  This could happen if for
81    # example S has volfrac and P has both volfrac and volfrac_S.
82    if any(p.id in p_set for p in s_list):
83        raise TypeError("name collision: P has P.name and P.name_S while S has S.name")
84
85    translate_name = dict((old.id, new.id) for old, new
86                          in zip(s_pars.kernel_parameters[1:], s_list))
87    combined_pars = p_pars.kernel_parameters + s_list + make_extra_pars(p_info)
88    parameters = ParameterTable(combined_pars)
89    parameters.max_pd = p_pars.max_pd + s_pars.max_pd
90    def random():
91        combined_pars = p_info.random()
92        s_names = set(par.id for par in s_pars.kernel_parameters[1:])
93        combined_pars.update((translate_name[k], v)
94                             for k, v in s_info.random().items()
95                             if k in s_names)
96        return combined_pars
97
98    model_info = ModelInfo()
99    model_info.id = '@'.join((p_id, s_id))
100    model_info.name = '@'.join((p_name, s_name))
101    model_info.filename = None
102    model_info.title = 'Product of %s and %s'%(p_name, s_name)
103    model_info.description = model_info.title
104    model_info.docs = model_info.title
105    model_info.category = "custom"
106    model_info.parameters = parameters
107    model_info.random = random
108    #model_info.single = p_info.single and s_info.single
109    model_info.structure_factor = False
110    model_info.variant_info = None
111    #model_info.tests = []
112    #model_info.source = []
113    # Iq, Iqxy, form_volume, ER, VR and sesans
114    # Remember the component info blocks so we can build the model
115    model_info.composition = ('product', [p_info, s_info])
116    model_info.control = p_info.control
117    model_info.hidden = p_info.hidden
118    if getattr(p_info, 'profile', None) is not None:
119        profile_pars = set(p.id for p in p_info.parameters.kernel_parameters)
120        def profile(**kwargs):
121            # extract the profile args
122            kwargs = dict((k, v) for k, v in kwargs.items() if k in profile_pars)
123            return p_info.profile(**kwargs)
124    else:
125        profile = None
126    model_info.profile = profile
127    model_info.profile_axes = p_info.profile_axes
128
129    # TODO: delegate random to p_info, s_info
130    #model_info.random = lambda: {}
131
132    ## Show the parameter table
133    #from .compare import get_pars, parlist
134    #print("==== %s ====="%model_info.name)
135    #values = get_pars(model_info)
136    #print(parlist(model_info, values, is2d=True))
137    return model_info
138
139def _tag_parameter(par):
140    """
141    Tag the parameter name with _S to indicate that the parameter comes from
142    the structure factor parameter set.  This is only necessary if the
143    form factor model includes a parameter of the same name as a parameter
144    in the structure factor.
145    """
146    par = copy(par)
147    # Protect against a vector parameter in S by appending the vector length
148    # to the renamed parameter.  Note: haven't tested this since no existing
149    # structure factor models contain vector parameters.
150    vector_length = par.name[len(par.id):]
151    par.id = par.id + "_S"
152    par.name = par.id + vector_length
153    return par
154
155def _intermediates(P, S):
156    # type: (np.ndarray, np.ndarray) -> OrderedDict[str, np.ndarray]
157    """
158    Returns intermediate results for standard product (P(Q)*S(Q))
159    """
160    return OrderedDict((
161        ("P(Q)", P),
162        ("S(Q)", S),
163    ))
164
165def _intermediates_beta(F1, F2, S, scale, bg):
166    # type: (np.ndarray, np.ndarray, np.ndarray, np.ndarray, np.ndarray) -> OrderedDict[str, np.ndarray]
167    """
168    Returns intermediate results for beta approximation-enabled product
169    """
170    # TODO: 1. include calculated Q vector
171    # TODO: 2. consider implications if there are intermediate results in P(Q)
172    return OrderedDict((
173        ("P(Q)", scale*F2),
174        ("S(Q)", S),
175        ("beta(Q)", F1**2 / F2),
176        ("S_eff(Q)", 1 + (F1**2 / F2)*(S-1)),
177        # ("I(Q)", scale*(F2 + (F1**2)*(S-1)) + bg),
178    ))
179
180class ProductModel(KernelModel):
181    def __init__(self, model_info, P, S):
182        # type: (ModelInfo, KernelModel, KernelModel) -> None
183        #: Combined info plock for the product model
184        self.info = model_info
185        #: Form factor modelling individual particles.
186        self.P = P
187        #: Structure factor modelling interaction between particles.
188        self.S = S
189
190        #: Model precision. This is not really relevant, since it is the
191        #: individual P and S models that control the effective dtype,
192        #: converting the q-vectors to the correct type when the kernels
193        #: for each are created. Ideally this should be set to the more
194        #: precise type to avoid loss of precision, but precision in q is
195        #: not critical (single is good enough for our purposes), so it just
196        #: uses the precision of the form factor.
197        self.dtype = P.dtype  # type: np.dtype
198
199    def make_kernel(self, q_vectors):
200        # type: (List[np.ndarray]) -> Kernel
201        # Note: may be sending the q_vectors to the GPU twice even though they
202        # are only needed once.  It would mess up modularity quite a bit to
203        # handle this optimally, especially since there are many cases where
204        # separate q vectors are needed (e.g., form in python and structure
205        # in opencl; or both in opencl, but one in single precision and the
206        # other in double precision).
207
208        p_kernel = self.P.make_kernel(q_vectors)
209        s_kernel = self.S.make_kernel(q_vectors)
210        return ProductKernel(self.info, p_kernel, s_kernel)
211
212    def release(self):
213        # type: (None) -> None
214        """
215        Free resources associated with the model.
216        """
217        self.P.release()
218        self.S.release()
219
220
221class ProductKernel(Kernel):
222    def __init__(self, model_info, p_kernel, s_kernel):
223        # type: (ModelInfo, Kernel, Kernel) -> None
224        self.info = model_info
225        self.p_kernel = p_kernel
226        self.s_kernel = s_kernel
227        self.dtype = p_kernel.dtype
228        self.results = []  # type: List[np.ndarray]
229
230    def __call__(self, call_details, values, cutoff, magnetic):
231        # type: (CallDetails, np.ndarray, float, bool) -> np.ndarray
232        p_info, s_info = self.info.composition[1]
233
234        # if there are magnetic parameters, they will only be on the
235        # form factor P, not the structure factor S.
236        nmagnetic = len(self.info.parameters.magnetism_index)
237        if nmagnetic:
238            spin_index = self.info.parameters.npars + 2
239            magnetism = values[spin_index: spin_index+3+3*nmagnetic]
240        else:
241            magnetism = []
242        nvalues = self.info.parameters.nvalues
243        nweights = call_details.num_weights
244        weights = values[nvalues:nvalues + 2*nweights]
245
246        # Construct the calling parameters for P.
247        p_npars = p_info.parameters.npars
248        p_length = call_details.length[:p_npars]
249        p_offset = call_details.offset[:p_npars]
250        p_details = make_details(p_info, p_length, p_offset, nweights)
251
252        # Set p scale to the volume fraction in s, which is the first of the
253        # 'S' parameters in the parameter list, or 2+np in 0-origin.
254        volfrac = values[2+p_npars]
255        p_values = [[volfrac, 0.0], values[2:2+p_npars], magnetism, weights]
256        spacer = (32 - sum(len(v) for v in p_values)%32)%32
257        p_values.append([0.]*spacer)
258        p_values = np.hstack(p_values).astype(self.p_kernel.dtype)
259
260        # Call ER and VR for P since these are needed for S.
261        p_er, p_vr = calc_er_vr(p_info, p_details, p_values)
262        s_vr = (volfrac/p_vr if p_vr != 0. else volfrac)
263        #print("volfrac:%g p_er:%g p_vr:%g s_vr:%g"%(volfrac,p_er,p_vr,s_vr))
264
265        # Construct the calling parameters for S.
266        # The  effective radius is not in the combined parameter list, so
267        # the number of 'S' parameters is one less than expected.  The
268        # computed effective radius needs to be added into the weights
269        # vector, especially since it is a polydisperse parameter in the
270        # stand-alone structure factor models.  We will added it at the
271        # end so the remaining offsets don't need to change.
272        s_npars = s_info.parameters.npars-1
273        s_length = call_details.length[p_npars:p_npars+s_npars]
274        s_offset = call_details.offset[p_npars:p_npars+s_npars]
275        s_length = np.hstack((1, s_length))
276        s_offset = np.hstack((nweights, s_offset))
277        s_details = make_details(s_info, s_length, s_offset, nweights+1)
278        v, w = weights[:nweights], weights[nweights:]
279        s_values = [
280            # scale=1, background=0, radius_effective=p_er, volfraction=s_vr
281            [1., 0., p_er, s_vr],
282            # structure factor parameters start after scale, background and
283            # all the form factor parameters.  Skip the volfraction parameter
284            # as well, since it is computed elsewhere, and go to the end of the
285            # parameter list.
286            values[2+p_npars+1:2+p_npars+s_npars],
287            # no magnetism parameters to include for S
288            # add er into the (value, weights) pairs
289            v, [p_er], w, [1.0]
290        ]
291        spacer = (32 - sum(len(v) for v in s_values)%32)%32
292        s_values.append([0.]*spacer)
293        s_values = np.hstack(s_values).astype(self.s_kernel.dtype)
294
295        # beta mode is the first parameter after the structure factor pars
296        beta_index = 2+p_npars+s_npars
297        beta_mode = values[beta_index]
298
299        # Call the kernels
300        s_result = self.s_kernel.Iq(s_details, s_values, cutoff, False)
301        scale, background = values[0], values[1]
302        if beta_mode:
303            F1, F2, volume_avg = self.p_kernel.beta(p_details, p_values, cutoff, magnetic)
304            combined_scale = scale*volfrac/volume_avg
305
306            self.results = lambda: _intermediates_beta(F1, F2, s_result, volfrac/volume_avg, background)
307            final_result = combined_scale*(F2 + (F1**2)*(s_result - 1)) + background
308
309        else:
310            p_result = self.p_kernel.Iq(p_details, p_values, cutoff, magnetic)
311
312            self.results = lambda: _intermediates(p_result, s_result)
313            final_result = scale*(p_result*s_result) + background
314
315        #call_details.show(values)
316        #print("values", values)
317        #p_details.show(p_values)
318        #print("=>", p_result)
319        #s_details.show(s_values)
320        #print("=>", s_result)
321        #import pylab as plt
322        #plt.subplot(211); plt.loglog(self.p_kernel.q_input.q, p_result, '-')
323        #plt.subplot(212); plt.loglog(self.s_kernel.q_input.q, s_result, '-')
324        #plt.figure()
325        return final_result
326
327    def release(self):
328        # type: () -> None
329        self.p_kernel.release()
330        self.s_kernel.release()
331
332
333def calc_er_vr(model_info, call_details, values):
334    # type: (ModelInfo, ParameterSet) -> Tuple[float, float]
335
336    if model_info.ER is None and model_info.VR is None:
337        return 1.0, 1.0
338
339    nvalues = model_info.parameters.nvalues
340    value = values[nvalues:nvalues + call_details.num_weights]
341    weight = values[nvalues + call_details.num_weights: nvalues + 2*call_details.num_weights]
342    npars = model_info.parameters.npars
343    # Note: changed from pairs ([v], [w]) to triples (p, [v], [w]), but the
344    # dispersion mesh code doesn't actually care about the nominal parameter
345    # value, p, so set it to None.
346    pairs = [(None, value[offset:offset+length], weight[offset:offset+length])
347             for p, offset, length
348             in zip(model_info.parameters.call_parameters[2:2+npars],
349                    call_details.offset,
350                    call_details.length)
351             if p.type == 'volume']
352    value, weight = dispersion_mesh(model_info, pairs)
353
354    if model_info.ER is not None:
355        individual_radii = model_info.ER(*value)
356        radius_effective = np.sum(weight*individual_radii) / np.sum(weight)
357    else:
358        radius_effective = 1.0
359
360    if model_info.VR is not None:
361        whole, part = model_info.VR(*value)
362        volume_ratio = np.sum(weight*part)/np.sum(weight*whole)
363    else:
364        volume_ratio = 1.0
365
366    return radius_effective, volume_ratio
Note: See TracBrowser for help on using the repository browser.