source: sasmodels/sasmodels/product.py @ c036ddb

core_shell_microgelsmagnetic_modelticket-1257-vesicle-productticket_1156ticket_1265_superballticket_822_more_unit_tests
Last change on this file since c036ddb was c036ddb, checked in by Paul Kienzle <pkienzle@…>, 11 months ago

refactor so Iq is not needed if Fq is defined

  • Property mode set to 100644
File size: 15.3 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 copy import copy
16import numpy as np  # type: ignore
17
18from .modelinfo import ParameterTable, ModelInfo, Parameter
19from .kernel import KernelModel, Kernel
20from .details import make_details, dispersion_mesh
21
22# pylint: disable=unused-import
23try:
24    from typing import Tuple
25except ImportError:
26    pass
27else:
28    from .modelinfo import ParameterSet
29# pylint: enable=unused-import
30
31# TODO: make estimates available to constraints
32#ESTIMATED_PARAMETERS = [
33#    ["est_radius_effective", "A", 0.0, [0, np.inf], "", "Estimated effective radius"],
34#    ["est_volume_ratio", "", 1.0, [0, np.inf], "", "Estimated volume ratio"],
35#]
36
37ER_ID = "radius_effective"
38VF_ID = "volfraction"
39BETA_DEFINITION = ("beta_mode", "", 0, [["P*S"],["P*(1+beta*(S-1))"]], "",
40                   "Structure factor dispersion calculation mode")
41
42# TODO: core_shell_sphere model has suppressed the volume ratio calculation
43# revert it after making VR and ER available at run time as constraints.
44def make_product_info(p_info, s_info):
45    # type: (ModelInfo, ModelInfo) -> ModelInfo
46    """
47    Create info block for product model.
48    """
49    # Make sure effective radius is the first parameter and
50    # make sure volume fraction is the second parameter of the
51    # structure factor calculator.  Structure factors should not
52    # have any magnetic parameters
53    if not len(s_info.parameters.kernel_parameters) >= 2:
54        raise TypeError("S needs {} and {} as its first parameters".format(ER_ID, VF_ID))
55    if not s_info.parameters.kernel_parameters[0].id == ER_ID:
56        raise TypeError("S needs {} as first parameter".format(ER_ID))
57    if not s_info.parameters.kernel_parameters[1].id == VF_ID:
58        raise TypeError("S needs {} as second parameter".format(VF_ID))
59    if not s_info.parameters.magnetism_index == []:
60        raise TypeError("S should not have SLD parameters")
61    p_id, p_name, p_pars = p_info.id, p_info.name, p_info.parameters
62    s_id, s_name, s_pars = s_info.id, s_info.name, s_info.parameters
63
64    # Create list of parameters for the combined model.  Skip the first
65    # parameter of S, which we verified above is effective radius.  If there
66    # are any names in P that overlap with those in S, modify the name in S
67    # to distinguish it.
68    p_set = set(p.id for p in p_pars.kernel_parameters)
69    s_list = [(_tag_parameter(par) if par.id in p_set else par)
70              for par in s_pars.kernel_parameters[1:]]
71    # Check if still a collision after renaming.  This could happen if for
72    # example S has volfrac and P has both volfrac and volfrac_S.
73    if any(p.id in p_set for p in s_list):
74        raise TypeError("name collision: P has P.name and P.name_S while S has S.name")
75
76    translate_name = dict((old.id, new.id) for old, new
77                          in zip(s_pars.kernel_parameters[1:], s_list))
78    beta = [Parameter(*BETA_DEFINITION)] if p_info.have_Fq else []
79    combined_pars = p_pars.kernel_parameters + s_list + beta
80    parameters = ParameterTable(combined_pars)
81    parameters.max_pd = p_pars.max_pd + s_pars.max_pd
82    def random():
83        combined_pars = p_info.random()
84        s_names = set(par.id for par in s_pars.kernel_parameters[1:])
85        combined_pars.update((translate_name[k], v)
86                             for k, v in s_info.random().items()
87                             if k in s_names)
88        return combined_pars
89
90    model_info = ModelInfo()
91    model_info.id = '@'.join((p_id, s_id))
92    model_info.name = '@'.join((p_name, s_name))
93    model_info.filename = None
94    model_info.title = 'Product of %s and %s'%(p_name, s_name)
95    model_info.description = model_info.title
96    model_info.docs = model_info.title
97    model_info.category = "custom"
98    model_info.parameters = parameters
99    model_info.random = random
100    #model_info.single = p_info.single and s_info.single
101    model_info.structure_factor = False
102    model_info.variant_info = None
103    #model_info.tests = []
104    #model_info.source = []
105    # Iq, Iqxy, form_volume, ER, VR and sesans
106    # Remember the component info blocks so we can build the model
107    model_info.composition = ('product', [p_info, s_info])
108    model_info.control = p_info.control
109    model_info.hidden = p_info.hidden
110    if getattr(p_info, 'profile', None) is not None:
111        profile_pars = set(p.id for p in p_info.parameters.kernel_parameters)
112        def profile(**kwargs):
113            # extract the profile args
114            kwargs = dict((k, v) for k, v in kwargs.items() if k in profile_pars)
115            return p_info.profile(**kwargs)
116    else:
117        profile = None
118    model_info.profile = profile
119    model_info.profile_axes = p_info.profile_axes
120
121    # TODO: delegate random to p_info, s_info
122    #model_info.random = lambda: {}
123
124    ## Show the parameter table
125    #from .compare import get_pars, parlist
126    #print("==== %s ====="%model_info.name)
127    #values = get_pars(model_info)
128    #print(parlist(model_info, values, is2d=True))
129    return model_info
130
131def _tag_parameter(par):
132    """
133    Tag the parameter name with _S to indicate that the parameter comes from
134    the structure factor parameter set.  This is only necessary if the
135    form factor model includes a parameter of the same name as a parameter
136    in the structure factor.
137    """
138    par = copy(par)
139    # Protect against a vector parameter in S by appending the vector length
140    # to the renamed parameter.  Note: haven't tested this since no existing
141    # structure factor models contain vector parameters.
142    vector_length = par.name[len(par.id):]
143    par.id = par.id + "_S"
144    par.name = par.id + vector_length
145    return par
146
147class ProductModel(KernelModel):
148    def __init__(self, model_info, P, S):
149        # type: (ModelInfo, KernelModel, KernelModel) -> None
150        #: Combined info plock for the product model
151        self.info = model_info
152        #: Form factor modelling individual particles.
153        self.P = P
154        #: Structure factor modelling interaction between particles.
155        self.S = S
156
157        #: Model precision. This is not really relevant, since it is the
158        #: individual P and S models that control the effective dtype,
159        #: converting the q-vectors to the correct type when the kernels
160        #: for each are created. Ideally this should be set to the more
161        #: precise type to avoid loss of precision, but precision in q is
162        #: not critical (single is good enough for our purposes), so it just
163        #: uses the precision of the form factor.
164        self.dtype = P.dtype  # type: np.dtype
165
166    def make_kernel(self, q_vectors):
167        # type: (List[np.ndarray]) -> Kernel
168        # Note: may be sending the q_vectors to the GPU twice even though they
169        # are only needed once.  It would mess up modularity quite a bit to
170        # handle this optimally, especially since there are many cases where
171        # separate q vectors are needed (e.g., form in python and structure
172        # in opencl; or both in opencl, but one in single precision and the
173        # other in double precision).
174
175        p_kernel = self.P.make_kernel(q_vectors)
176        s_kernel = self.S.make_kernel(q_vectors)
177        return ProductKernel(self.info, p_kernel, s_kernel)
178
179    def release(self):
180        # type: (None) -> None
181        """
182        Free resources associated with the model.
183        """
184        self.P.release()
185        self.S.release()
186
187
188class ProductKernel(Kernel):
189    def __init__(self, model_info, p_kernel, s_kernel):
190        # type: (ModelInfo, Kernel, Kernel) -> None
191        self.info = model_info
192        self.p_kernel = p_kernel
193        self.s_kernel = s_kernel
194        self.dtype = p_kernel.dtype
195        self.results = []  # type: List[np.ndarray]
196
197    def __call__(self, call_details, values, cutoff, magnetic):
198        # type: (CallDetails, np.ndarray, float, bool) -> np.ndarray
199        p_info, s_info = self.info.composition[1]
200
201        # if there are magnetic parameters, they will only be on the
202        # form factor P, not the structure factor S.
203        nmagnetic = len(self.info.parameters.magnetism_index)
204        if nmagnetic:
205            spin_index = self.info.parameters.npars + 2
206            magnetism = values[spin_index: spin_index+3+3*nmagnetic]
207        else:
208            magnetism = []
209        nvalues = self.info.parameters.nvalues
210        nweights = call_details.num_weights
211        weights = values[nvalues:nvalues + 2*nweights]
212
213        # Construct the calling parameters for P.
214        p_npars = p_info.parameters.npars
215        p_length = call_details.length[:p_npars]
216        p_offset = call_details.offset[:p_npars]
217        p_details = make_details(p_info, p_length, p_offset, nweights)
218
219        # Set p scale to the volume fraction in s, which is the first of the
220        # 'S' parameters in the parameter list, or 2+np in 0-origin.
221        volfrac = values[2+p_npars]
222        p_values = [[volfrac, 0.0], values[2:2+p_npars], magnetism, weights]
223        spacer = (32 - sum(len(v) for v in p_values)%32)%32
224        p_values.append([0.]*spacer)
225        p_values = np.hstack(p_values).astype(self.p_kernel.dtype)
226
227        # Call ER and VR for P since these are needed for S.
228        p_er, p_vr = calc_er_vr(p_info, p_details, p_values)
229        s_vr = (volfrac/p_vr if p_vr != 0. else volfrac)
230        #print("volfrac:%g p_er:%g p_vr:%g s_vr:%g"%(volfrac,p_er,p_vr,s_vr))
231
232        # Construct the calling parameters for S.
233        # The  effective radius is not in the combined parameter list, so
234        # the number of 'S' parameters is one less than expected.  The
235        # computed effective radius needs to be added into the weights
236        # vector, especially since it is a polydisperse parameter in the
237        # stand-alone structure factor models.  We will added it at the
238        # end so the remaining offsets don't need to change.
239        s_npars = s_info.parameters.npars-1
240        s_length = call_details.length[p_npars:p_npars+s_npars]
241        s_offset = call_details.offset[p_npars:p_npars+s_npars]
242        s_length = np.hstack((1, s_length))
243        s_offset = np.hstack((nweights, s_offset))
244        s_details = make_details(s_info, s_length, s_offset, nweights+1)
245        v, w = weights[:nweights], weights[nweights:]
246        s_values = [
247            # scale=1, background=0, radius_effective=p_er, volfraction=s_vr
248            [1., 0., p_er, s_vr],
249            # structure factor parameters start after scale, background and
250            # all the form factor parameters.  Skip the volfraction parameter
251            # as well, since it is computed elsewhere, and go to the end of the
252            # parameter list.
253            values[2+p_npars+1:2+p_npars+s_npars],
254            # no magnetism parameters to include for S
255            # add er into the (value, weights) pairs
256            v, [p_er], w, [1.0]
257        ]
258        spacer = (32 - sum(len(v) for v in s_values)%32)%32
259        s_values.append([0.]*spacer)
260        s_values = np.hstack(s_values).astype(self.s_kernel.dtype)
261
262        # beta mode is the first parameter after the structure factor pars
263        beta_index = 2+p_npars+s_npars
264        beta_mode = values[beta_index]
265
266        # Call the kernels
267        s_result = self.s_kernel.Iq(s_details, s_values, cutoff, False)
268        scale, background = values[0], values[1]
269        if beta_mode:
270            F1, F2, volume_avg = self.p_kernel.beta(p_details, p_values, cutoff, magnetic)
271            combined_scale = scale*volfrac/volume_avg
272            # Define lazy results based on intermediate values.
273            # The return value for the calculation should be an ordered
274            # dictionary containing any result the user might want to see
275            # at the end of the calculation, including scalars, strings,
276            # and plottable data.  Don't want to build this structure during
277            # fits, only when displaying the final result (or a one-off
278            # computation which asks for it).
279            # Do not use the current hack of storing the intermediate values
280            # in self.results since that leads to awkward threading issues.
281            # Instead return the function along with the bundle of inputs.
282            # P and Q may themselves have intermediate results they want to
283            # include, such as A and B if P = A + B.  Might use this mechanism
284            # to return the computed effective radius as well.
285            #def lazy_results(Q, S, F1, F2, scale):
286            #    """
287            #    beta = F1**2 / F2  # what about divide by zero errors?
288            #    return {
289            #        'P' : Data1D(Q, scale*F2),
290            #        'beta': Data1D(Q, beta),
291            #        'S' : Data1D(Q, S),
292            #        'Seff': Data1D(Q, 1 + beta*(S-1)),
293            #        'I' : Data1D(Q, scale*(F2 + (F1**2)*(S-1)) + background),
294            #    }
295            #lazy_pars = s_result, F1, F2, combined_scale
296            self.results = [F2, s_result]
297            final_result = combined_scale*(F2 + (F1**2)*(s_result - 1)) + background
298        else:
299            p_result = self.p_kernel.Iq(p_details, p_values, cutoff, magnetic)
300            # remember the parts for plotting later
301            self.results = [p_result, s_result]
302            final_result = scale*(p_result*s_result) + background
303
304        #call_details.show(values)
305        #print("values", values)
306        #p_details.show(p_values)
307        #print("=>", p_result)
308        #s_details.show(s_values)
309        #print("=>", s_result)
310        #import pylab as plt
311        #plt.subplot(211); plt.loglog(self.p_kernel.q_input.q, p_result, '-')
312        #plt.subplot(212); plt.loglog(self.s_kernel.q_input.q, s_result, '-')
313        #plt.figure()
314        return final_result
315
316    def release(self):
317        # type: () -> None
318        self.p_kernel.release()
319        self.s_kernel.release()
320
321
322def calc_er_vr(model_info, call_details, values):
323    # type: (ModelInfo, ParameterSet) -> Tuple[float, float]
324
325    if model_info.ER is None and model_info.VR is None:
326        return 1.0, 1.0
327
328    nvalues = model_info.parameters.nvalues
329    value = values[nvalues:nvalues + call_details.num_weights]
330    weight = values[nvalues + call_details.num_weights: nvalues + 2*call_details.num_weights]
331    npars = model_info.parameters.npars
332    # Note: changed from pairs ([v], [w]) to triples (p, [v], [w]), but the
333    # dispersion mesh code doesn't actually care about the nominal parameter
334    # value, p, so set it to None.
335    pairs = [(None, value[offset:offset+length], weight[offset:offset+length])
336             for p, offset, length
337             in zip(model_info.parameters.call_parameters[2:2+npars],
338                    call_details.offset,
339                    call_details.length)
340             if p.type == 'volume']
341    value, weight = dispersion_mesh(model_info, pairs)
342
343    if model_info.ER is not None:
344        individual_radii = model_info.ER(*value)
345        radius_effective = np.sum(weight*individual_radii) / np.sum(weight)
346    else:
347        radius_effective = 1.0
348
349    if model_info.VR is not None:
350        whole, part = model_info.VR(*value)
351        volume_ratio = np.sum(weight*part)/np.sum(weight*whole)
352    else:
353        volume_ratio = 1.0
354
355    return radius_effective, volume_ratio
Note: See TracBrowser for help on using the repository browser.