Changes in sasmodels/product.py [99658f6:b171acd] in sasmodels

File:

• sasmodels/product.py (modified) (14 diffs)

sasmodels/product.py

@@ -13 +13 @@
 from __future__ import print_function, division
 
-from collections import OrderedDict
-
 from copy import copy
 import numpy as np  # type: ignore
 
-from .modelinfo import ParameterTable, ModelInfo, Parameter, parse_parameter
+from .modelinfo import ParameterTable, ModelInfo
 from .kernel import KernelModel, Kernel
 from .details import make_details, dispersion_mesh
@@ -24 +22 @@
 # pylint: disable=unused-import
 try:
-    from typing import Tuple, Callable, Union
+    from typing import Tuple
 except ImportError:
     pass
@@ -37 +35 @@
 #]
 
-STRUCTURE_MODE_ID = "structure_factor_mode"
-RADIUS_MODE_ID = "radius_effective_mode"
-RADIUS_ID = "radius_effective"
-VOLFRAC_ID = "volfraction"
-def make_extra_pars(p_info):
-    pars = []
-    if p_info.have_Fq:
-        par = parse_parameter(
-                STRUCTURE_MODE_ID,
-                "",
-                0,
-                [["P*S","P*(1+beta*(S-1))"]],
-                "",
-                "Structure factor calculation")
-        pars.append(par)
-    if p_info.effective_radius_type is not None:
-        par = parse_parameter(
-                RADIUS_MODE_ID,
-                "",
-                1,
-                [["unconstrained"] + p_info.effective_radius_type],
-                "",
-                "Effective radius calculation")
-        pars.append(par)
-    return pars
-
+ER_ID = "radius_effective"
+VF_ID = "volfraction"
+
+# TODO: core_shell_sphere model has suppressed the volume ratio calculation
+# revert it after making VR and ER available at run time as constraints.
 def make_product_info(p_info, s_info):
     # type: (ModelInfo, ModelInfo) -> ModelInfo
@@ -73 +50 @@
     # have any magnetic parameters
     if not len(s_info.parameters.kernel_parameters) >= 2:
-        raise TypeError("S needs {} and {} as its first parameters".format(RADIUS_ID, VOLFRAC_ID))
-    if not s_info.parameters.kernel_parameters[0].id == RADIUS_ID:
-        raise TypeError("S needs {} as first parameter".format(RADIUS_ID))
-    if not s_info.parameters.kernel_parameters[1].id == VOLFRAC_ID:
-        raise TypeError("S needs {} as second parameter".format(VOLFRAC_ID))
+        raise TypeError("S needs {} and {} as its first parameters".format(ER_ID, VF_ID))
+    if not s_info.parameters.kernel_parameters[0].id == ER_ID:
+        raise TypeError("S needs {} as first parameter".format(ER_ID))
+    if not s_info.parameters.kernel_parameters[1].id == VF_ID:
+        raise TypeError("S needs {} as second parameter".format(VF_ID))
     if not s_info.parameters.magnetism_index == []:
         raise TypeError("S should not have SLD parameters")
@@ -83 +60 @@
     s_id, s_name, s_pars = s_info.id, s_info.name, s_info.parameters
 
-    # Create list of parameters for the combined model.  If there
+    # Create list of parameters for the combined model.  Skip the first
+    # parameter of S, which we verified above is effective radius.  If there
     # are any names in P that overlap with those in S, modify the name in S
     # to distinguish it.
     p_set = set(p.id for p in p_pars.kernel_parameters)
     s_list = [(_tag_parameter(par) if par.id in p_set else par)
-              for par in s_pars.kernel_parameters]
+              for par in s_pars.kernel_parameters[1:]]
     # Check if still a collision after renaming.  This could happen if for
     # example S has volfrac and P has both volfrac and volfrac_S.
@@ -94 +72 @@
         raise TypeError("name collision: P has P.name and P.name_S while S has S.name")
 
-    # make sure effective radius is not a polydisperse parameter in product
-    s_list[0] = copy(s_list[0])
-    s_list[0].polydisperse = False
-
     translate_name = dict((old.id, new.id) for old, new
-                          in zip(s_pars.kernel_parameters, s_list))
-    combined_pars = p_pars.kernel_parameters + s_list + make_extra_pars(p_info)
+                          in zip(s_pars.kernel_parameters[1:], s_list))
+    combined_pars = p_pars.kernel_parameters + s_list
     parameters = ParameterTable(combined_pars)
     parameters.max_pd = p_pars.max_pd + s_pars.max_pd
     def random():
         combined_pars = p_info.random()
-        s_names = set(par.id for par in s_pars.kernel_parameters)
+        s_names = set(par.id for par in s_pars.kernel_parameters[1:])
         combined_pars.update((translate_name[k], v)
                              for k, v in s_info.random().items()
@@ -126 +100 @@
     #model_info.tests = []
     #model_info.source = []
+    # Iq, Iqxy, form_volume, ER, VR and sesans
     # Remember the component info blocks so we can build the model
     model_info.composition = ('product', [p_info, s_info])
-    model_info.control = p_info.control
     model_info.hidden = p_info.hidden
     if getattr(p_info, 'profile', None) is not None:
@@ -167 +141 @@
     return par
 
-def _intermediates(
-        F1,               # type: np.ndarray
-        F2,               # type: np.ndarray
-        S,                # type: np.ndarray
-        scale,            # type: float
-        effective_radius, # type: float
-        beta_mode,        # type: bool
-        ):
-    # type: (...) -> OrderedDict[str, Union[np.ndarray, float]]
-    """
-    Returns intermediate results for beta approximation-enabled product.
-    The result may be an array or a float.
-    """
-    if beta_mode:
-        # TODO: 1. include calculated Q vector
-        # TODO: 2. consider implications if there are intermediate results in P(Q)
-        parts = OrderedDict((
-            ("P(Q)", scale*F2),
-            ("S(Q)", S),
-            ("beta(Q)", F1**2 / F2),
-            ("S_eff(Q)", 1 + (F1**2 / F2)*(S-1)),
-            ("effective_radius", effective_radius),
-            # ("I(Q)", scale*(F2 + (F1**2)*(S-1)) + bg),
-        ))
-    else:
-        parts = OrderedDict((
-            ("P(Q)", scale*F2),
-            ("S(Q)", S),
-            ("effective_radius", effective_radius),
-        ))
-    return parts
-
 class ProductModel(KernelModel):
     def __init__(self, model_info, P, S):
@@ -208 +150 @@
         #: Structure factor modelling interaction between particles.
         self.S = S
-
         #: Model precision. This is not really relevant, since it is the
         #: individual P and S models that control the effective dtype,
@@ -226 +167 @@
         # in opencl; or both in opencl, but one in single precision and the
         # other in double precision).
-
         p_kernel = self.P.make_kernel(q_vectors)
         s_kernel = self.S.make_kernel(q_vectors)
@@ -251 +191 @@
     def __call__(self, call_details, values, cutoff, magnetic):
         # type: (CallDetails, np.ndarray, float, bool) -> np.ndarray
-
         p_info, s_info = self.info.composition[1]
-        p_npars = p_info.parameters.npars
-        p_length = call_details.length[:p_npars]
-        p_offset = call_details.offset[:p_npars]
-        s_npars = s_info.parameters.npars
-        s_length = call_details.length[p_npars:p_npars+s_npars]
-        s_offset = call_details.offset[p_npars:p_npars+s_npars]
-
-        # Beta mode parameter is the first parameter after P and S parameters
-        have_beta_mode = p_info.have_Fq
-        beta_mode_offset = 2+p_npars+s_npars
-        beta_mode = (values[beta_mode_offset] > 0) if have_beta_mode else False
-        if beta_mode and self.p_kernel.dim== '2d':
-            raise NotImplementedError("beta not yet supported for 2D")
-
-        # R_eff type parameter is the second parameter after P and S parameters
-        # unless the model doesn't support beta mode, in which case it is first
-        have_radius_type = p_info.effective_radius_type is not None
-        radius_type_offset = 2+p_npars+s_npars + (1 if have_beta_mode else 0)
-        radius_type = int(values[radius_type_offset]) if have_radius_type else 0
-
-        # Retrieve the volume fraction, which is the second of the
-        # 'S' parameters in the parameter list, or 2+np in 0-origin,
-        # as well as the scale and background.
-        volfrac = values[3+p_npars]
-        scale, background = values[0], values[1]
 
         # if there are magnetic parameters, they will only be on the
@@ -292 +206 @@
 
         # Construct the calling parameters for P.
+        p_npars = p_info.parameters.npars
+        p_length = call_details.length[:p_npars]
+        p_offset = call_details.offset[:p_npars]
         p_details = make_details(p_info, p_length, p_offset, nweights)
-        p_values = [
-            [1., 0.], # scale=1, background=0,
-            values[2:2+p_npars],
-            magnetism,
-            weights]
+        # Set p scale to the volume fraction in s, which is the first of the
+        # 'S' parameters in the parameter list, or 2+np in 0-origin.
+        volfrac = values[2+p_npars]
+        p_values = [[volfrac, 0.0], values[2:2+p_npars], magnetism, weights]
         spacer = (32 - sum(len(v) for v in p_values)%32)%32
         p_values.append([0.]*spacer)
         p_values = np.hstack(p_values).astype(self.p_kernel.dtype)
 
+        # Call ER and VR for P since these are needed for S.
+        p_er, p_vr = calc_er_vr(p_info, p_details, p_values)
+        s_vr = (volfrac/p_vr if p_vr != 0. else volfrac)
+        #print("volfrac:%g p_er:%g p_vr:%g s_vr:%g"%(volfrac,p_er,p_vr,s_vr))
+
         # Construct the calling parameters for S.
-        if radius_type > 0:
-            # If R_eff comes from form factor, make sure it is monodisperse.
-            # weight is set to 1 later, after the value array is created
-            s_length[0] = 1
-        s_details = make_details(s_info, s_length, s_offset, nweights)
+        # The  effective radius is not in the combined parameter list, so
+        # the number of 'S' parameters is one less than expected.  The
+        # computed effective radius needs to be added into the weights
+        # vector, especially since it is a polydisperse parameter in the
+        # stand-alone structure factor models.  We will added it at the
+        # end so the remaining offsets don't need to change.
+        s_npars = s_info.parameters.npars-1
+        s_length = call_details.length[p_npars:p_npars+s_npars]
+        s_offset = call_details.offset[p_npars:p_npars+s_npars]
+        s_length = np.hstack((1, s_length))
+        s_offset = np.hstack((nweights, s_offset))
+        s_details = make_details(s_info, s_length, s_offset, nweights+1)
+        v, w = weights[:nweights], weights[nweights:]
         s_values = [
-            [1., 0.], # scale=1, background=0,
-            values[2+p_npars:2+p_npars+s_npars],
-            weights,
+            # scale=1, background=0, radius_effective=p_er, volfraction=s_vr
+            [1., 0., p_er, s_vr],
+            # structure factor parameters start after scale, background and
+            # all the form factor parameters.  Skip the volfraction parameter
+            # as well, since it is computed elsewhere, and go to the end of the
+            # parameter list.
+            values[2+p_npars+1:2+p_npars+s_npars],
+            # no magnetism parameters to include for S
+            # add er into the (value, weights) pairs
+            v, [p_er], w, [1.0]
         ]
         spacer = (32 - sum(len(v) for v in s_values)%32)%32
@@ -317 +253 @@
         s_values = np.hstack(s_values).astype(self.s_kernel.dtype)
 
-        # Call the form factor kernel to compute <F> and <F^2>.
-        # If the model doesn't support Fq the returned <F> will be None.
-        F1, F2, effective_radius, shell_volume, volume_ratio = self.p_kernel.Fq(
-            p_details, p_values, cutoff, magnetic, radius_type)
-
-        # Call the structure factor kernel to compute S.
-        # Plug R_eff from the form factor into structure factor parameters
-        # and scale volume fraction by form:shell volume ratio. These changes
-        # needs to be both in the initial value slot as well as the
-        # polydispersity distribution slot in the values array due to
-        # implementation details in kernel_iq.c.
-        #print("R_eff=%d:%g, volfrac=%g, volume ratio=%g"%(radius_type, effective_radius, volfrac, volume_ratio))
-        if radius_type > 0:
-            # set the value to the model R_eff and set the weight to 1
-            s_values[2] = s_values[2+s_npars+s_offset[0]] = effective_radius
-            s_values[2+s_npars+s_offset[0]+nweights] = 1.0
-        s_values[3] = s_values[2+s_npars+s_offset[1]] = volfrac*volume_ratio
-        S = self.s_kernel.Iq(s_details, s_values, cutoff, False)
-
-        # Determine overall scale factor. Hollow shapes are weighted by
-        # shell_volume, so that is needed for volume normalization.  For
-        # solid shapes we can use shell_volume as well since it is equal
-        # to form volume.
-        combined_scale = scale*volfrac/shell_volume
-
-        # Combine form factor and structure factor
-        #print("beta", beta_mode, F1, F2, S)
-        PS = F2 + F1**2*(S-1) if beta_mode else F2*S
-        final_result = combined_scale*PS + background
-
-        # Capture intermediate values so user can see them.  These are
-        # returned as a lazy evaluator since they are only needed in the
-        # GUI, and not for each evaluation during a fit.
-        # TODO: return the results structure with the final results
-        # That way the model calcs are idempotent. Further, we can
-        # generalize intermediates to various other model types if we put it
-        # kernel calling interface.  Could do this as an "optional"
-        # return value in the caller, though in that case we could return
-        # the results directly rather than through a lazy evaluator.
-        self.results = lambda: _intermediates(
-            F1, F2, S, combined_scale, effective_radius, beta_mode)
-
-        return final_result
+        # Call the kernels
+        p_result = self.p_kernel(p_details, p_values, cutoff, magnetic)
+        s_result = self.s_kernel(s_details, s_values, cutoff, False)
+
+        #print("p_npars",p_npars,s_npars,p_er,s_vr,values[2+p_npars+1:2+p_npars+s_npars])
+        #call_details.show(values)
+        #print("values", values)
+        #p_details.show(p_values)
+        #print("=>", p_result)
+        #s_details.show(s_values)
+        #print("=>", s_result)
+
+        # remember the parts for plotting later
+        self.results = [p_result, s_result]
+
+        #import pylab as plt
+        #plt.subplot(211); plt.loglog(self.p_kernel.q_input.q, p_result, '-')
+        #plt.subplot(212); plt.loglog(self.s_kernel.q_input.q, s_result, '-')
+        #plt.figure()
+
+        return values[0]*(p_result*s_result) + values[1]
 
     def release(self):
@@ -365 +279 @@
         self.p_kernel.release()
         self.s_kernel.release()
+
+
+def calc_er_vr(model_info, call_details, values):
+    # type: (ModelInfo, ParameterSet) -> Tuple[float, float]
+
+    if model_info.ER is None and model_info.VR is None:
+        return 1.0, 1.0
+
+    nvalues = model_info.parameters.nvalues
+    value = values[nvalues:nvalues + call_details.num_weights]
+    weight = values[nvalues + call_details.num_weights: nvalues + 2*call_details.num_weights]
+    npars = model_info.parameters.npars
+    # Note: changed from pairs ([v], [w]) to triples (p, [v], [w]), but the
+    # dispersion mesh code doesn't actually care about the nominal parameter
+    # value, p, so set it to None.
+    pairs = [(None, value[offset:offset+length], weight[offset:offset+length])
+             for p, offset, length
+             in zip(model_info.parameters.call_parameters[2:2+npars],
+                    call_details.offset,
+                    call_details.length)
+             if p.type == 'volume']
+    value, weight = dispersion_mesh(model_info, pairs)
+
+    if model_info.ER is not None:
+        individual_radii = model_info.ER(*value)
+        radius_effective = np.sum(weight*individual_radii) / np.sum(weight)
+    else:
+        radius_effective = 1.0
+
+    if model_info.VR is not None:
+        whole, part = model_info.VR(*value)
+        volume_ratio = np.sum(weight*part)/np.sum(weight*whole)
+    else:
+        volume_ratio = 1.0
+
+    return radius_effective, volume_ratio