source: sasmodels/sasmodels/product.py @ fd7291e

core_shell_microgelsmagnetic_modelticket-1257-vesicle-productticket_1156ticket_1265_superballticket_822_more_unit_tests
Last change on this file since fd7291e was fd7291e, checked in by Paul Kienzle <pkienzle@…>, 8 months ago

Merge branch 'beta_approx' into ticket-1022-sum_multiplicity

  • 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, parse_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, Union
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
39STRUCTURE_MODE_ID = "structure_factor_mode"
40RADIUS_MODE_ID = "radius_effective_mode"
41RADIUS_ID = "radius_effective"
42VOLFRAC_ID = "volfraction"
43def make_extra_pars(p_info):
44    pars = []
45    if p_info.have_Fq:
46        par = parse_parameter(
47                STRUCTURE_MODE_ID,
48                "",
49                0,
50                [["P*S","P*(1+beta*(S-1))"]],
51                "",
52                "Structure factor calculation")
53        pars.append(par)
54    if p_info.effective_radius_type is not None:
55        par = parse_parameter(
56                RADIUS_MODE_ID,
57                "",
58                1,
59                [["unconstrained"] + p_info.effective_radius_type],
60                "",
61                "Effective radius calculation")
62        pars.append(par)
63    return pars
64
65def make_product_info(p_info, s_info):
66    # type: (ModelInfo, ModelInfo) -> ModelInfo
67    """
68    Create info block for product model.
69    """
70    # Make sure effective radius is the first parameter and
71    # make sure volume fraction is the second parameter of the
72    # structure factor calculator.  Structure factors should not
73    # have any magnetic parameters
74    if not len(s_info.parameters.kernel_parameters) >= 2:
75        raise TypeError("S needs {} and {} as its first parameters".format(RADIUS_ID, VOLFRAC_ID))
76    if not s_info.parameters.kernel_parameters[0].id == RADIUS_ID:
77        raise TypeError("S needs {} as first parameter".format(RADIUS_ID))
78    if not s_info.parameters.kernel_parameters[1].id == VOLFRAC_ID:
79        raise TypeError("S needs {} as second parameter".format(VOLFRAC_ID))
80    if not s_info.parameters.magnetism_index == []:
81        raise TypeError("S should not have SLD parameters")
82    p_id, p_name, p_pars = p_info.id, p_info.name, p_info.parameters
83    s_id, s_name, s_pars = s_info.id, s_info.name, s_info.parameters
84
85    # Create list of parameters for the combined model.  If there
86    # are any names in P that overlap with those in S, modify the name in S
87    # to distinguish it.
88    p_set = set(p.id for p in p_pars.kernel_parameters)
89    s_list = [(_tag_parameter(par) if par.id in p_set else par)
90              for par in s_pars.kernel_parameters]
91    # Check if still a collision after renaming.  This could happen if for
92    # example S has volfrac and P has both volfrac and volfrac_S.
93    if any(p.id in p_set for p in s_list):
94        raise TypeError("name collision: P has P.name and P.name_S while S has S.name")
95
96    # make sure effective radius is not a polydisperse parameter in product
97    s_list[0] = copy(s_list[0])
98    s_list[0].polydisperse = False
99
100    translate_name = dict((old.id, new.id) for old, new
101                          in zip(s_pars.kernel_parameters, s_list))
102    combined_pars = p_pars.kernel_parameters + s_list + make_extra_pars(p_info)
103    parameters = ParameterTable(combined_pars)
104    parameters.max_pd = p_pars.max_pd + s_pars.max_pd
105    def random():
106        combined_pars = p_info.random()
107        s_names = set(par.id for par in s_pars.kernel_parameters)
108        combined_pars.update((translate_name[k], v)
109                             for k, v in s_info.random().items()
110                             if k in s_names)
111        return combined_pars
112
113    model_info = ModelInfo()
114    model_info.id = '@'.join((p_id, s_id))
115    model_info.name = '@'.join((p_name, s_name))
116    model_info.filename = None
117    model_info.title = 'Product of %s and %s'%(p_name, s_name)
118    model_info.description = model_info.title
119    model_info.docs = model_info.title
120    model_info.category = "custom"
121    model_info.parameters = parameters
122    model_info.random = random
123    #model_info.single = p_info.single and s_info.single
124    model_info.structure_factor = False
125    model_info.variant_info = None
126    #model_info.tests = []
127    #model_info.source = []
128    # Remember the component info blocks so we can build the model
129    model_info.composition = ('product', [p_info, s_info])
130    model_info.hidden = p_info.hidden
131    if getattr(p_info, 'profile', None) is not None:
132        profile_pars = set(p.id for p in p_info.parameters.kernel_parameters)
133        def profile(**kwargs):
134            # extract the profile args
135            kwargs = dict((k, v) for k, v in kwargs.items() if k in profile_pars)
136            return p_info.profile(**kwargs)
137    else:
138        profile = None
139    model_info.profile = profile
140    model_info.profile_axes = p_info.profile_axes
141
142    # TODO: delegate random to p_info, s_info
143    #model_info.random = lambda: {}
144
145    ## Show the parameter table
146    #from .compare import get_pars, parlist
147    #print("==== %s ====="%model_info.name)
148    #values = get_pars(model_info)
149    #print(parlist(model_info, values, is2d=True))
150    return model_info
151
152def _tag_parameter(par):
153    """
154    Tag the parameter name with _S to indicate that the parameter comes from
155    the structure factor parameter set.  This is only necessary if the
156    form factor model includes a parameter of the same name as a parameter
157    in the structure factor.
158    """
159    par = copy(par)
160    # Protect against a vector parameter in S by appending the vector length
161    # to the renamed parameter.  Note: haven't tested this since no existing
162    # structure factor models contain vector parameters.
163    vector_length = par.name[len(par.id):]
164    par.id = par.id + "_S"
165    par.name = par.id + vector_length
166    return par
167
168def _intermediates(
169        F1,               # type: np.ndarray
170        F2,               # type: np.ndarray
171        S,                # type: np.ndarray
172        scale,            # type: float
173        effective_radius, # type: float
174        beta_mode,        # type: bool
175        ):
176    # type: (...) -> OrderedDict[str, Union[np.ndarray, float]]
177    """
178    Returns intermediate results for beta approximation-enabled product.
179    The result may be an array or a float.
180    """
181    if beta_mode:
182        # TODO: 1. include calculated Q vector
183        # TODO: 2. consider implications if there are intermediate results in P(Q)
184        parts = OrderedDict((
185            ("P(Q)", scale*F2),
186            ("S(Q)", S),
187            ("beta(Q)", F1**2 / F2),
188            ("S_eff(Q)", 1 + (F1**2 / F2)*(S-1)),
189            ("effective_radius", effective_radius),
190            # ("I(Q)", scale*(F2 + (F1**2)*(S-1)) + bg),
191        ))
192    else:
193        parts = OrderedDict((
194            ("P(Q)", scale*F2),
195            ("S(Q)", S),
196            ("effective_radius", effective_radius),
197        ))
198    return parts
199
200class ProductModel(KernelModel):
201    def __init__(self, model_info, P, S):
202        # type: (ModelInfo, KernelModel, KernelModel) -> None
203        #: Combined info plock for the product model
204        self.info = model_info
205        #: Form factor modelling individual particles.
206        self.P = P
207        #: Structure factor modelling interaction between particles.
208        self.S = S
209
210        #: Model precision. This is not really relevant, since it is the
211        #: individual P and S models that control the effective dtype,
212        #: converting the q-vectors to the correct type when the kernels
213        #: for each are created. Ideally this should be set to the more
214        #: precise type to avoid loss of precision, but precision in q is
215        #: not critical (single is good enough for our purposes), so it just
216        #: uses the precision of the form factor.
217        self.dtype = P.dtype  # type: np.dtype
218
219    def make_kernel(self, q_vectors):
220        # type: (List[np.ndarray]) -> Kernel
221        # Note: may be sending the q_vectors to the GPU twice even though they
222        # are only needed once.  It would mess up modularity quite a bit to
223        # handle this optimally, especially since there are many cases where
224        # separate q vectors are needed (e.g., form in python and structure
225        # in opencl; or both in opencl, but one in single precision and the
226        # other in double precision).
227
228        p_kernel = self.P.make_kernel(q_vectors)
229        s_kernel = self.S.make_kernel(q_vectors)
230        return ProductKernel(self.info, p_kernel, s_kernel)
231
232    def release(self):
233        # type: (None) -> None
234        """
235        Free resources associated with the model.
236        """
237        self.P.release()
238        self.S.release()
239
240
241class ProductKernel(Kernel):
242    def __init__(self, model_info, p_kernel, s_kernel):
243        # type: (ModelInfo, Kernel, Kernel) -> None
244        self.info = model_info
245        self.p_kernel = p_kernel
246        self.s_kernel = s_kernel
247        self.dtype = p_kernel.dtype
248        self.results = []  # type: List[np.ndarray]
249
250    def __call__(self, call_details, values, cutoff, magnetic):
251        # type: (CallDetails, np.ndarray, float, bool) -> np.ndarray
252
253        p_info, s_info = self.info.composition[1]
254        p_npars = p_info.parameters.npars
255        p_length = call_details.length[:p_npars]
256        p_offset = call_details.offset[:p_npars]
257        s_npars = s_info.parameters.npars
258        s_length = call_details.length[p_npars:p_npars+s_npars]
259        s_offset = call_details.offset[p_npars:p_npars+s_npars]
260
261        # Beta mode parameter is the first parameter after P and S parameters
262        have_beta_mode = p_info.have_Fq
263        beta_mode_offset = 2+p_npars+s_npars
264        beta_mode = (values[beta_mode_offset] > 0) if have_beta_mode else False
265        if beta_mode and self.p_kernel.dim== '2d':
266            raise NotImplementedError("beta not yet supported for 2D")
267
268        # R_eff type parameter is the second parameter after P and S parameters
269        # unless the model doesn't support beta mode, in which case it is first
270        have_radius_type = p_info.effective_radius_type is not None
271        radius_type_offset = 2+p_npars+s_npars + (1 if have_beta_mode else 0)
272        radius_type = int(values[radius_type_offset]) if have_radius_type else 0
273
274        # Retrieve the volume fraction, which is the second of the
275        # 'S' parameters in the parameter list, or 2+np in 0-origin,
276        # as well as the scale and background.
277        volfrac = values[3+p_npars]
278        scale, background = values[0], values[1]
279
280        # if there are magnetic parameters, they will only be on the
281        # form factor P, not the structure factor S.
282        nmagnetic = len(self.info.parameters.magnetism_index)
283        if nmagnetic:
284            spin_index = self.info.parameters.npars + 2
285            magnetism = values[spin_index: spin_index+3+3*nmagnetic]
286        else:
287            magnetism = []
288        nvalues = self.info.parameters.nvalues
289        nweights = call_details.num_weights
290        weights = values[nvalues:nvalues + 2*nweights]
291
292        # Construct the calling parameters for P.
293        p_details = make_details(p_info, p_length, p_offset, nweights)
294        p_values = [
295            [1., 0.], # scale=1, background=0,
296            values[2:2+p_npars],
297            magnetism,
298            weights]
299        spacer = (32 - sum(len(v) for v in p_values)%32)%32
300        p_values.append([0.]*spacer)
301        p_values = np.hstack(p_values).astype(self.p_kernel.dtype)
302
303        # Construct the calling parameters for S.
304        if radius_type > 0:
305            # If R_eff comes from form factor, make sure it is monodisperse.
306            # weight is set to 1 later, after the value array is created
307            s_length[0] = 1
308        s_details = make_details(s_info, s_length, s_offset, nweights)
309        s_values = [
310            [1., 0.], # scale=1, background=0,
311            values[2+p_npars:2+p_npars+s_npars],
312            weights,
313        ]
314        spacer = (32 - sum(len(v) for v in s_values)%32)%32
315        s_values.append([0.]*spacer)
316        s_values = np.hstack(s_values).astype(self.s_kernel.dtype)
317
318        # Call the form factor kernel to compute <F> and <F^2>.
319        # If the model doesn't support Fq the returned <F> will be None.
320        F1, F2, effective_radius, shell_volume, volume_ratio = self.p_kernel.Fq(
321            p_details, p_values, cutoff, magnetic, radius_type)
322
323        # Call the structure factor kernel to compute S.
324        # Plug R_eff from the form factor into structure factor parameters
325        # and scale volume fraction by form:shell volume ratio. These changes
326        # needs to be both in the initial value slot as well as the
327        # polydispersity distribution slot in the values array due to
328        # implementation details in kernel_iq.c.
329        #print("R_eff=%d:%g, volfrac=%g, volume ratio=%g"%(radius_type, effective_radius, volfrac, volume_ratio))
330        if radius_type > 0:
331            # set the value to the model R_eff and set the weight to 1
332            s_values[2] = s_values[2+s_npars+s_offset[0]] = effective_radius
333            s_values[2+s_npars+s_offset[0]+nweights] = 1.0
334        s_values[3] = s_values[2+s_npars+s_offset[1]] = volfrac*volume_ratio
335        S = self.s_kernel.Iq(s_details, s_values, cutoff, False)
336
337        # Determine overall scale factor. Hollow shapes are weighted by
338        # shell_volume, so that is needed for volume normalization.  For
339        # solid shapes we can use shell_volume as well since it is equal
340        # to form volume.
341        combined_scale = scale*volfrac/shell_volume
342
343        # Combine form factor and structure factor
344        #print("beta", beta_mode, F1, F2, S)
345        PS = F2 + F1**2*(S-1) if beta_mode else F2*S
346        final_result = combined_scale*PS + background
347
348        # Capture intermediate values so user can see them.  These are
349        # returned as a lazy evaluator since they are only needed in the
350        # GUI, and not for each evaluation during a fit.
351        # TODO: return the results structure with the final results
352        # That way the model calcs are idempotent. Further, we can
353        # generalize intermediates to various other model types if we put it
354        # kernel calling interface.  Could do this as an "optional"
355        # return value in the caller, though in that case we could return
356        # the results directly rather than through a lazy evaluator.
357        self.results = lambda: _intermediates(
358            F1, F2, S, combined_scale, effective_radius, beta_mode)
359
360        return final_result
361
362    def release(self):
363        # type: () -> None
364        self.p_kernel.release()
365        self.s_kernel.release()
Note: See TracBrowser for help on using the repository browser.