| 1 | """ |
|---|
| 2 | Product model |
|---|
| 3 | ------------- |
|---|
| 4 | |
|---|
| 5 | The product model multiplies the structure factor by the form factor, |
|---|
| 6 | modulated by the effective radius of the form. The resulting model |
|---|
| 7 | has a attributes of both the model description (with parameters, etc.) |
|---|
| 8 | and the module evaluator (with call, release, etc.). |
|---|
| 9 | |
|---|
| 10 | To use it, first load form factor P and structure factor S, then create |
|---|
| 11 | *make_product_info(P, S)*. |
|---|
| 12 | """ |
|---|
| 13 | from __future__ import print_function, division |
|---|
| 14 | |
|---|
| 15 | from collections import OrderedDict |
|---|
| 16 | |
|---|
| 17 | from copy import copy |
|---|
| 18 | import numpy as np # type: ignore |
|---|
| 19 | |
|---|
| 20 | from .modelinfo import ParameterTable, ModelInfo, Parameter, parse_parameter |
|---|
| 21 | from .kernel import KernelModel, Kernel |
|---|
| 22 | from .details import make_details, dispersion_mesh |
|---|
| 23 | |
|---|
| 24 | # pylint: disable=unused-import |
|---|
| 25 | try: |
|---|
| 26 | from typing import Tuple, Callable, Union |
|---|
| 27 | except ImportError: |
|---|
| 28 | pass |
|---|
| 29 | else: |
|---|
| 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 | |
|---|
| 39 | RADIUS_ID = "radius_effective" |
|---|
| 40 | VOLFRAC_ID = "volfraction" |
|---|
| 41 | def make_extra_pars(p_info): |
|---|
| 42 | pars = [] |
|---|
| 43 | if p_info.have_Fq: |
|---|
| 44 | par = parse_parameter( |
|---|
| 45 | "structure_factor_mode", |
|---|
| 46 | "", |
|---|
| 47 | 0, |
|---|
| 48 | [["P*S","P*(1+beta*(S-1))"]], |
|---|
| 49 | "", |
|---|
| 50 | "Structure factor calculation") |
|---|
| 51 | pars.append(par) |
|---|
| 52 | if p_info.effective_radius_type is not None: |
|---|
| 53 | par = parse_parameter( |
|---|
| 54 | "radius_effective_mode", |
|---|
| 55 | "", |
|---|
| 56 | 0, |
|---|
| 57 | [["unconstrained"] + p_info.effective_radius_type], |
|---|
| 58 | "", |
|---|
| 59 | "Effective radius calculation") |
|---|
| 60 | pars.append(par) |
|---|
| 61 | return pars |
|---|
| 62 | |
|---|
| 63 | # TODO: core_shell_sphere model has suppressed the volume ratio calculation |
|---|
| 64 | # revert it after making VR and ER available at run time as constraints. |
|---|
| 65 | def 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 | # Iq, Iqxy, form_volume, ER, VR and sesans |
|---|
| 129 | # Remember the component info blocks so we can build the model |
|---|
| 130 | model_info.composition = ('product', [p_info, s_info]) |
|---|
| 131 | model_info.control = p_info.control |
|---|
| 132 | model_info.hidden = p_info.hidden |
|---|
| 133 | if getattr(p_info, 'profile', None) is not None: |
|---|
| 134 | profile_pars = set(p.id for p in p_info.parameters.kernel_parameters) |
|---|
| 135 | def profile(**kwargs): |
|---|
| 136 | # extract the profile args |
|---|
| 137 | kwargs = dict((k, v) for k, v in kwargs.items() if k in profile_pars) |
|---|
| 138 | return p_info.profile(**kwargs) |
|---|
| 139 | else: |
|---|
| 140 | profile = None |
|---|
| 141 | model_info.profile = profile |
|---|
| 142 | model_info.profile_axes = p_info.profile_axes |
|---|
| 143 | |
|---|
| 144 | # TODO: delegate random to p_info, s_info |
|---|
| 145 | #model_info.random = lambda: {} |
|---|
| 146 | |
|---|
| 147 | ## Show the parameter table |
|---|
| 148 | #from .compare import get_pars, parlist |
|---|
| 149 | #print("==== %s ====="%model_info.name) |
|---|
| 150 | #values = get_pars(model_info) |
|---|
| 151 | #print(parlist(model_info, values, is2d=True)) |
|---|
| 152 | return model_info |
|---|
| 153 | |
|---|
| 154 | def _tag_parameter(par): |
|---|
| 155 | """ |
|---|
| 156 | Tag the parameter name with _S to indicate that the parameter comes from |
|---|
| 157 | the structure factor parameter set. This is only necessary if the |
|---|
| 158 | form factor model includes a parameter of the same name as a parameter |
|---|
| 159 | in the structure factor. |
|---|
| 160 | """ |
|---|
| 161 | par = copy(par) |
|---|
| 162 | # Protect against a vector parameter in S by appending the vector length |
|---|
| 163 | # to the renamed parameter. Note: haven't tested this since no existing |
|---|
| 164 | # structure factor models contain vector parameters. |
|---|
| 165 | vector_length = par.name[len(par.id):] |
|---|
| 166 | par.id = par.id + "_S" |
|---|
| 167 | par.name = par.id + vector_length |
|---|
| 168 | return par |
|---|
| 169 | |
|---|
| 170 | def _intermediates( |
|---|
| 171 | F1, # type: np.ndarray |
|---|
| 172 | F2, # type: np.ndarray |
|---|
| 173 | S, # type: np.ndarray |
|---|
| 174 | scale, # type: float |
|---|
| 175 | effective_radius, # type: float |
|---|
| 176 | beta_mode, # type: bool |
|---|
| 177 | ): |
|---|
| 178 | # type: (...) -> OrderedDict[str, Union[np.ndarray, float]] |
|---|
| 179 | """ |
|---|
| 180 | Returns intermediate results for beta approximation-enabled product. |
|---|
| 181 | The result may be an array or a float. |
|---|
| 182 | """ |
|---|
| 183 | if beta_mode: |
|---|
| 184 | # TODO: 1. include calculated Q vector |
|---|
| 185 | # TODO: 2. consider implications if there are intermediate results in P(Q) |
|---|
| 186 | parts = OrderedDict(( |
|---|
| 187 | ("P(Q)", scale*F2), |
|---|
| 188 | ("S(Q)", S), |
|---|
| 189 | ("beta(Q)", F1**2 / F2), |
|---|
| 190 | ("S_eff(Q)", 1 + (F1**2 / F2)*(S-1)), |
|---|
| 191 | ("effective_radius", effective_radius), |
|---|
| 192 | # ("I(Q)", scale*(F2 + (F1**2)*(S-1)) + bg), |
|---|
| 193 | )) |
|---|
| 194 | else: |
|---|
| 195 | parts = OrderedDict(( |
|---|
| 196 | ("P(Q)", scale*F2), |
|---|
| 197 | ("S(Q)", S), |
|---|
| 198 | ("effective_radius", effective_radius), |
|---|
| 199 | )) |
|---|
| 200 | return parts |
|---|
| 201 | |
|---|
| 202 | class ProductModel(KernelModel): |
|---|
| 203 | def __init__(self, model_info, P, S): |
|---|
| 204 | # type: (ModelInfo, KernelModel, KernelModel) -> None |
|---|
| 205 | #: Combined info plock for the product model |
|---|
| 206 | self.info = model_info |
|---|
| 207 | #: Form factor modelling individual particles. |
|---|
| 208 | self.P = P |
|---|
| 209 | #: Structure factor modelling interaction between particles. |
|---|
| 210 | self.S = S |
|---|
| 211 | |
|---|
| 212 | #: Model precision. This is not really relevant, since it is the |
|---|
| 213 | #: individual P and S models that control the effective dtype, |
|---|
| 214 | #: converting the q-vectors to the correct type when the kernels |
|---|
| 215 | #: for each are created. Ideally this should be set to the more |
|---|
| 216 | #: precise type to avoid loss of precision, but precision in q is |
|---|
| 217 | #: not critical (single is good enough for our purposes), so it just |
|---|
| 218 | #: uses the precision of the form factor. |
|---|
| 219 | self.dtype = P.dtype # type: np.dtype |
|---|
| 220 | |
|---|
| 221 | def make_kernel(self, q_vectors): |
|---|
| 222 | # type: (List[np.ndarray]) -> Kernel |
|---|
| 223 | # Note: may be sending the q_vectors to the GPU twice even though they |
|---|
| 224 | # are only needed once. It would mess up modularity quite a bit to |
|---|
| 225 | # handle this optimally, especially since there are many cases where |
|---|
| 226 | # separate q vectors are needed (e.g., form in python and structure |
|---|
| 227 | # in opencl; or both in opencl, but one in single precision and the |
|---|
| 228 | # other in double precision). |
|---|
| 229 | |
|---|
| 230 | p_kernel = self.P.make_kernel(q_vectors) |
|---|
| 231 | s_kernel = self.S.make_kernel(q_vectors) |
|---|
| 232 | return ProductKernel(self.info, p_kernel, s_kernel) |
|---|
| 233 | |
|---|
| 234 | def release(self): |
|---|
| 235 | # type: (None) -> None |
|---|
| 236 | """ |
|---|
| 237 | Free resources associated with the model. |
|---|
| 238 | """ |
|---|
| 239 | self.P.release() |
|---|
| 240 | self.S.release() |
|---|
| 241 | |
|---|
| 242 | |
|---|
| 243 | class ProductKernel(Kernel): |
|---|
| 244 | def __init__(self, model_info, p_kernel, s_kernel): |
|---|
| 245 | # type: (ModelInfo, Kernel, Kernel) -> None |
|---|
| 246 | self.info = model_info |
|---|
| 247 | self.p_kernel = p_kernel |
|---|
| 248 | self.s_kernel = s_kernel |
|---|
| 249 | self.dtype = p_kernel.dtype |
|---|
| 250 | self.results = [] # type: List[np.ndarray] |
|---|
| 251 | |
|---|
| 252 | def __call__(self, call_details, values, cutoff, magnetic): |
|---|
| 253 | # type: (CallDetails, np.ndarray, float, bool) -> np.ndarray |
|---|
| 254 | |
|---|
| 255 | p_info, s_info = self.info.composition[1] |
|---|
| 256 | p_npars = p_info.parameters.npars |
|---|
| 257 | p_length = call_details.length[:p_npars] |
|---|
| 258 | p_offset = call_details.offset[:p_npars] |
|---|
| 259 | s_npars = s_info.parameters.npars |
|---|
| 260 | s_length = call_details.length[p_npars:p_npars+s_npars] |
|---|
| 261 | s_offset = call_details.offset[p_npars:p_npars+s_npars] |
|---|
| 262 | |
|---|
| 263 | # Beta mode parameter is the first parameter after P and S parameters |
|---|
| 264 | have_beta_mode = p_info.have_Fq |
|---|
| 265 | beta_mode_offset = 2+p_npars+s_npars |
|---|
| 266 | beta_mode = (values[beta_mode_offset] > 0) if have_beta_mode else False |
|---|
| 267 | if beta_mode and self.p_kernel.dim== '2d': |
|---|
| 268 | raise NotImplementedError("beta not yet supported for 2D") |
|---|
| 269 | |
|---|
| 270 | # R_eff type parameter is the second parameter after P and S parameters |
|---|
| 271 | # unless the model doesn't support beta mode, in which case it is first |
|---|
| 272 | have_radius_type = p_info.effective_radius_type is not None |
|---|
| 273 | radius_type_offset = 2+p_npars+s_npars + (1 if have_beta_mode else 0) |
|---|
| 274 | radius_type = int(values[radius_type_offset]) if have_radius_type else 0 |
|---|
| 275 | |
|---|
| 276 | # Retrieve the volume fraction, which is the second of the |
|---|
| 277 | # 'S' parameters in the parameter list, or 2+np in 0-origin, |
|---|
| 278 | # as well as the scale and background. |
|---|
| 279 | volfrac = values[3+p_npars] |
|---|
| 280 | scale, background = values[0], values[1] |
|---|
| 281 | |
|---|
| 282 | # if there are magnetic parameters, they will only be on the |
|---|
| 283 | # form factor P, not the structure factor S. |
|---|
| 284 | nmagnetic = len(self.info.parameters.magnetism_index) |
|---|
| 285 | if nmagnetic: |
|---|
| 286 | spin_index = self.info.parameters.npars + 2 |
|---|
| 287 | magnetism = values[spin_index: spin_index+3+3*nmagnetic] |
|---|
| 288 | else: |
|---|
| 289 | magnetism = [] |
|---|
| 290 | nvalues = self.info.parameters.nvalues |
|---|
| 291 | nweights = call_details.num_weights |
|---|
| 292 | weights = values[nvalues:nvalues + 2*nweights] |
|---|
| 293 | |
|---|
| 294 | # Construct the calling parameters for P. |
|---|
| 295 | p_details = make_details(p_info, p_length, p_offset, nweights) |
|---|
| 296 | p_values = [ |
|---|
| 297 | [1., 0.], # scale=1, background=0, |
|---|
| 298 | values[2:2+p_npars], |
|---|
| 299 | magnetism, |
|---|
| 300 | weights] |
|---|
| 301 | spacer = (32 - sum(len(v) for v in p_values)%32)%32 |
|---|
| 302 | p_values.append([0.]*spacer) |
|---|
| 303 | p_values = np.hstack(p_values).astype(self.p_kernel.dtype) |
|---|
| 304 | |
|---|
| 305 | # Construct the calling parameters for S. |
|---|
| 306 | if radius_type > 0: |
|---|
| 307 | # If R_eff comes from form factor, make sure it is monodisperse. |
|---|
| 308 | # weight is set to 1 later, after the value array is created |
|---|
| 309 | s_length[0] = 1 |
|---|
| 310 | s_details = make_details(s_info, s_length, s_offset, nweights) |
|---|
| 311 | s_values = [ |
|---|
| 312 | [1., 0.], # scale=1, background=0, |
|---|
| 313 | values[2+p_npars:2+p_npars+s_npars], |
|---|
| 314 | weights, |
|---|
| 315 | ] |
|---|
| 316 | spacer = (32 - sum(len(v) for v in s_values)%32)%32 |
|---|
| 317 | s_values.append([0.]*spacer) |
|---|
| 318 | s_values = np.hstack(s_values).astype(self.s_kernel.dtype) |
|---|
| 319 | |
|---|
| 320 | # Call the form factor kernel to compute <F> and <F^2>. |
|---|
| 321 | # If the model doesn't support Fq the returned <F> will be None. |
|---|
| 322 | F1, F2, effective_radius, shell_volume, volume_ratio = self.p_kernel.Fq( |
|---|
| 323 | p_details, p_values, cutoff, magnetic, radius_type) |
|---|
| 324 | |
|---|
| 325 | # Call the structure factor kernel to compute S. |
|---|
| 326 | # Plug R_eff from the form factor into structure factor parameters |
|---|
| 327 | # and scale volume fraction by form:shell volume ratio. These changes |
|---|
| 328 | # needs to be both in the initial value slot as well as the |
|---|
| 329 | # polydispersity distribution slot in the values array due to |
|---|
| 330 | # implementation details in kernel_iq.c. |
|---|
| 331 | #print("R_eff=%d:%g, volfrac=%g, volume ratio=%g"%(radius_type, effective_radius, volfrac, volume_ratio)) |
|---|
| 332 | if radius_type > 0: |
|---|
| 333 | # set the value to the model ER and set the weight to 1 |
|---|
| 334 | s_values[2] = s_values[2+s_npars+s_offset[0]] = effective_radius |
|---|
| 335 | s_values[2+s_npars+s_offset[0]+nweights] = 1.0 |
|---|
| 336 | s_values[3] = s_values[2+s_npars+s_offset[1]] = volfrac*volume_ratio |
|---|
| 337 | S = self.s_kernel.Iq(s_details, s_values, cutoff, False) |
|---|
| 338 | |
|---|
| 339 | # Determine overall scale factor. Hollow shapes are weighted by |
|---|
| 340 | # shell_volume, so that is needed for volume normalization. For |
|---|
| 341 | # solid shapes we can use shell_volume as well since it is equal |
|---|
| 342 | # to form volume. |
|---|
| 343 | combined_scale = scale*volfrac/shell_volume |
|---|
| 344 | |
|---|
| 345 | # Combine form factor and structure factor |
|---|
| 346 | #print("beta", beta_mode, F1, F2, S) |
|---|
| 347 | PS = F2 + F1**2*(S-1) if beta_mode else F2*S |
|---|
| 348 | final_result = combined_scale*PS + background |
|---|
| 349 | |
|---|
| 350 | # Capture intermediate values so user can see them. These are |
|---|
| 351 | # returned as a lazy evaluator since they are only needed in the |
|---|
| 352 | # GUI, and not for each evaluation during a fit. |
|---|
| 353 | # TODO: return the results structure with the final results |
|---|
| 354 | # That way the model calcs are idempotent. Further, we can |
|---|
| 355 | # generalize intermediates to various other model types if we put it |
|---|
| 356 | # kernel calling interface. Could do this as an "optional" |
|---|
| 357 | # return value in the caller, though in that case we could return |
|---|
| 358 | # the results directly rather than through a lazy evaluator. |
|---|
| 359 | self.results = lambda: _intermediates( |
|---|
| 360 | F1, F2, S, combined_scale, effective_radius, beta_mode) |
|---|
| 361 | |
|---|
| 362 | return final_result |
|---|
| 363 | |
|---|
| 364 | def release(self): |
|---|
| 365 | # type: () -> None |
|---|
| 366 | self.p_kernel.release() |
|---|
| 367 | self.s_kernel.release() |
|---|
| 368 | |
|---|
| 369 | |
|---|
| 370 | def calc_er_vr(model_info, call_details, values): |
|---|
| 371 | # type: (ModelInfo, ParameterSet) -> Tuple[float, float] |
|---|
| 372 | |
|---|
| 373 | if model_info.ER is None and model_info.VR is None: |
|---|
| 374 | return 1.0, 1.0 |
|---|
| 375 | |
|---|
| 376 | nvalues = model_info.parameters.nvalues |
|---|
| 377 | value = values[nvalues:nvalues + call_details.num_weights] |
|---|
| 378 | weight = values[nvalues + call_details.num_weights: nvalues + 2*call_details.num_weights] |
|---|
| 379 | npars = model_info.parameters.npars |
|---|
| 380 | # Note: changed from pairs ([v], [w]) to triples (p, [v], [w]), but the |
|---|
| 381 | # dispersion mesh code doesn't actually care about the nominal parameter |
|---|
| 382 | # value, p, so set it to None. |
|---|
| 383 | pairs = [(None, value[offset:offset+length], weight[offset:offset+length]) |
|---|
| 384 | for p, offset, length |
|---|
| 385 | in zip(model_info.parameters.call_parameters[2:2+npars], |
|---|
| 386 | call_details.offset, |
|---|
| 387 | call_details.length) |
|---|
| 388 | if p.type == 'volume'] |
|---|
| 389 | value, weight = dispersion_mesh(model_info, pairs) |
|---|
| 390 | |
|---|
| 391 | if model_info.ER is not None: |
|---|
| 392 | individual_radii = model_info.ER(*value) |
|---|
| 393 | radius_effective = np.sum(weight*individual_radii) / np.sum(weight) |
|---|
| 394 | else: |
|---|
| 395 | radius_effective = 1.0 |
|---|
| 396 | |
|---|
| 397 | if model_info.VR is not None: |
|---|
| 398 | whole, part = model_info.VR(*value) |
|---|
| 399 | volume_ratio = np.sum(weight*part)/np.sum(weight*whole) |
|---|
| 400 | else: |
|---|
| 401 | volume_ratio = 1.0 |
|---|
| 402 | |
|---|
| 403 | return radius_effective, volume_ratio |
|---|
