| 1 | """ |
|---|
| 2 | Python driver for python kernels |
|---|
| 3 | |
|---|
| 4 | Calls the kernel with a vector of $q$ values for a single parameter set. |
|---|
| 5 | Polydispersity is supported by looping over different parameter sets and |
|---|
| 6 | summing the results. The interface to :class:`PyModel` matches those for |
|---|
| 7 | :class:`kernelcl.GpuModel` and :class:`kerneldll.DllModel`. |
|---|
| 8 | """ |
|---|
| 9 | import numpy as np |
|---|
| 10 | from numpy import pi, cos |
|---|
| 11 | |
|---|
| 12 | from .generate import F64 |
|---|
| 13 | |
|---|
| 14 | class PyModel(object): |
|---|
| 15 | """ |
|---|
| 16 | Wrapper for pure python models. |
|---|
| 17 | """ |
|---|
| 18 | def __init__(self, model_info): |
|---|
| 19 | self.info = model_info |
|---|
| 20 | |
|---|
| 21 | def __call__(self, q_vectors): |
|---|
| 22 | q_input = PyInput(q_vectors, dtype=F64) |
|---|
| 23 | kernel = self.info['Iqxy'] if q_input.is_2d else self.info['Iq'] |
|---|
| 24 | return PyKernel(kernel, self.info, q_input) |
|---|
| 25 | |
|---|
| 26 | def release(self): |
|---|
| 27 | """ |
|---|
| 28 | Free resources associated with the model. |
|---|
| 29 | """ |
|---|
| 30 | pass |
|---|
| 31 | |
|---|
| 32 | class PyInput(object): |
|---|
| 33 | """ |
|---|
| 34 | Make q data available to the gpu. |
|---|
| 35 | |
|---|
| 36 | *q_vectors* is a list of q vectors, which will be *[q]* for 1-D data, |
|---|
| 37 | and *[qx, qy]* for 2-D data. Internally, the vectors will be reallocated |
|---|
| 38 | to get the best performance on OpenCL, which may involve shifting and |
|---|
| 39 | stretching the array to better match the memory architecture. Additional |
|---|
| 40 | points will be evaluated with *q=1e-3*. |
|---|
| 41 | |
|---|
| 42 | *dtype* is the data type for the q vectors. The data type should be |
|---|
| 43 | set to match that of the kernel, which is an attribute of |
|---|
| 44 | :class:`GpuProgram`. Note that not all kernels support double |
|---|
| 45 | precision, so even if the program was created for double precision, |
|---|
| 46 | the *GpuProgram.dtype* may be single precision. |
|---|
| 47 | |
|---|
| 48 | Call :meth:`release` when complete. Even if not called directly, the |
|---|
| 49 | buffer will be released when the data object is freed. |
|---|
| 50 | """ |
|---|
| 51 | def __init__(self, q_vectors, dtype): |
|---|
| 52 | self.nq = q_vectors[0].size |
|---|
| 53 | self.dtype = dtype |
|---|
| 54 | self.is_2d = (len(q_vectors) == 2) |
|---|
| 55 | self.q_vectors = [np.ascontiguousarray(q, self.dtype) for q in q_vectors] |
|---|
| 56 | self.q_pointers = [q.ctypes.data for q in self.q_vectors] |
|---|
| 57 | |
|---|
| 58 | def release(self): |
|---|
| 59 | """ |
|---|
| 60 | Free resources associated with the model inputs. |
|---|
| 61 | """ |
|---|
| 62 | self.q_vectors = [] |
|---|
| 63 | |
|---|
| 64 | class PyKernel(object): |
|---|
| 65 | """ |
|---|
| 66 | Callable SAS kernel. |
|---|
| 67 | |
|---|
| 68 | *kernel* is the DllKernel object to call. |
|---|
| 69 | |
|---|
| 70 | *model_info* is the module information |
|---|
| 71 | |
|---|
| 72 | *q_input* is the DllInput q vectors at which the kernel should be |
|---|
| 73 | evaluated. |
|---|
| 74 | |
|---|
| 75 | The resulting call method takes the *pars*, a list of values for |
|---|
| 76 | the fixed parameters to the kernel, and *pd_pars*, a list of (value,weight) |
|---|
| 77 | vectors for the polydisperse parameters. *cutoff* determines the |
|---|
| 78 | integration limits: any points with combined weight less than *cutoff* |
|---|
| 79 | will not be calculated. |
|---|
| 80 | |
|---|
| 81 | Call :meth:`release` when done with the kernel instance. |
|---|
| 82 | """ |
|---|
| 83 | def __init__(self, kernel, model_info, q_input): |
|---|
| 84 | self.info = model_info |
|---|
| 85 | self.q_input = q_input |
|---|
| 86 | self.res = np.empty(q_input.nq, q_input.dtype) |
|---|
| 87 | dim = '2d' if q_input.is_2d else '1d' |
|---|
| 88 | # Loop over q unless user promises that the kernel is vectorized by |
|---|
| 89 | # taggining it with vectorized=True |
|---|
| 90 | if not getattr(kernel, 'vectorized', False): |
|---|
| 91 | if dim == '2d': |
|---|
| 92 | def vector_kernel(qx, qy, *args): |
|---|
| 93 | """ |
|---|
| 94 | Vectorized 2D kernel. |
|---|
| 95 | """ |
|---|
| 96 | return np.array([kernel(qxi, qyi, *args) |
|---|
| 97 | for qxi, qyi in zip(qx, qy)]) |
|---|
| 98 | else: |
|---|
| 99 | def vector_kernel(q, *args): |
|---|
| 100 | """ |
|---|
| 101 | Vectorized 1D kernel. |
|---|
| 102 | """ |
|---|
| 103 | return np.array([kernel(qi, *args) |
|---|
| 104 | for qi in q]) |
|---|
| 105 | self.kernel = vector_kernel |
|---|
| 106 | else: |
|---|
| 107 | self.kernel = kernel |
|---|
| 108 | fixed_pars = model_info['partype']['fixed-' + dim] |
|---|
| 109 | pd_pars = model_info['partype']['pd-' + dim] |
|---|
| 110 | vol_pars = model_info['partype']['volume'] |
|---|
| 111 | |
|---|
| 112 | # First two fixed pars are scale and background |
|---|
| 113 | pars = [p.name for p in model_info['parameters'][2:]] |
|---|
| 114 | offset = len(self.q_input.q_vectors) |
|---|
| 115 | self.args = self.q_input.q_vectors + [None] * len(pars) |
|---|
| 116 | self.fixed_index = np.array([pars.index(p) + offset |
|---|
| 117 | for p in fixed_pars[2:]]) |
|---|
| 118 | self.pd_index = np.array([pars.index(p) + offset |
|---|
| 119 | for p in pd_pars]) |
|---|
| 120 | self.vol_index = np.array([pars.index(p) + offset |
|---|
| 121 | for p in vol_pars]) |
|---|
| 122 | try: self.theta_index = pars.index('theta') + offset |
|---|
| 123 | except ValueError: self.theta_index = -1 |
|---|
| 124 | |
|---|
| 125 | # Caller needs fixed_pars and pd_pars |
|---|
| 126 | self.fixed_pars = fixed_pars |
|---|
| 127 | self.pd_pars = pd_pars |
|---|
| 128 | |
|---|
| 129 | def __call__(self, fixed, pd, cutoff=1e-5): |
|---|
| 130 | #print("fixed",fixed) |
|---|
| 131 | #print("pd", pd) |
|---|
| 132 | args = self.args[:] # grab a copy of the args |
|---|
| 133 | form, form_volume = self.kernel, self.info['form_volume'] |
|---|
| 134 | # First two fixed |
|---|
| 135 | scale, background = fixed[:2] |
|---|
| 136 | for index, value in zip(self.fixed_index, fixed[2:]): |
|---|
| 137 | args[index] = float(value) |
|---|
| 138 | res = _loops(form, form_volume, cutoff, scale, background, args, |
|---|
| 139 | pd, self.pd_index, self.vol_index, self.theta_index) |
|---|
| 140 | |
|---|
| 141 | return res |
|---|
| 142 | |
|---|
| 143 | def release(self): |
|---|
| 144 | """ |
|---|
| 145 | Free resources associated with the kernel. |
|---|
| 146 | """ |
|---|
| 147 | self.q_input = None |
|---|
| 148 | |
|---|
| 149 | def _loops(form, form_volume, cutoff, scale, background, |
|---|
| 150 | args, pd, pd_index, vol_index, theta_index): |
|---|
| 151 | """ |
|---|
| 152 | *form* is the name of the form function, which should be vectorized over |
|---|
| 153 | q, but otherwise have an interface like the opencl kernels, with the |
|---|
| 154 | q parameters first followed by the individual parameters in the order |
|---|
| 155 | given in model.parameters (see :mod:`sasmodels.generate`). |
|---|
| 156 | |
|---|
| 157 | *form_volume* calculates the volume of the shape. *vol_index* gives |
|---|
| 158 | the list of volume parameters |
|---|
| 159 | |
|---|
| 160 | *cutoff* ignores the corners of the dispersion hypercube |
|---|
| 161 | |
|---|
| 162 | *scale*, *background* multiplies the resulting form and adds an offset |
|---|
| 163 | |
|---|
| 164 | *args* is the prepopulated set of arguments to the form function, starting |
|---|
| 165 | with the q vectors, and including slots for all the parameters. The |
|---|
| 166 | values for the parameters will be substituted with values from the |
|---|
| 167 | dispersion functions. |
|---|
| 168 | |
|---|
| 169 | *pd* is the list of dispersion parameters |
|---|
| 170 | |
|---|
| 171 | *pd_index* are the indices of the dispersion parameters in *args* |
|---|
| 172 | |
|---|
| 173 | *vol_index* are the indices of the volume parameters in *args* |
|---|
| 174 | |
|---|
| 175 | *theta_index* is the index of the theta parameter for the sasview |
|---|
| 176 | spherical correction, or -1 if there is no angular dispersion |
|---|
| 177 | """ |
|---|
| 178 | |
|---|
| 179 | ################################################################ |
|---|
| 180 | # # |
|---|
| 181 | # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # |
|---|
| 182 | # !! !! # |
|---|
| 183 | # !! KEEP THIS CODE CONSISTENT WITH KERNEL_TEMPLATE.C !! # |
|---|
| 184 | # !! !! # |
|---|
| 185 | # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # |
|---|
| 186 | # # |
|---|
| 187 | ################################################################ |
|---|
| 188 | |
|---|
| 189 | weight = np.empty(len(pd), 'd') |
|---|
| 190 | if weight.size > 0: |
|---|
| 191 | # weight vector, to be populated by polydispersity loops |
|---|
| 192 | |
|---|
| 193 | # identify which pd parameters are volume parameters |
|---|
| 194 | vol_weight_index = np.array([(index in vol_index) for index in pd_index]) |
|---|
| 195 | |
|---|
| 196 | # Sort parameters in decreasing order of pd length |
|---|
| 197 | Npd = np.array([len(pdi[0]) for pdi in pd], 'i') |
|---|
| 198 | order = np.argsort(Npd)[::-1] |
|---|
| 199 | stride = np.cumprod(Npd[order]) |
|---|
| 200 | pd = [pd[index] for index in order] |
|---|
| 201 | pd_index = pd_index[order] |
|---|
| 202 | vol_weight_index = vol_weight_index[order] |
|---|
| 203 | |
|---|
| 204 | fast_value = pd[0][0] |
|---|
| 205 | fast_weight = pd[0][1] |
|---|
| 206 | else: |
|---|
| 207 | stride = np.array([1]) |
|---|
| 208 | vol_weight_index = slice(None, None) |
|---|
| 209 | # keep lint happy |
|---|
| 210 | fast_value = [None] |
|---|
| 211 | fast_weight = [None] |
|---|
| 212 | |
|---|
| 213 | ret = np.zeros_like(args[0]) |
|---|
| 214 | norm = np.zeros_like(ret) |
|---|
| 215 | vol = np.zeros_like(ret) |
|---|
| 216 | vol_norm = np.zeros_like(ret) |
|---|
| 217 | for k in range(stride[-1]): |
|---|
| 218 | # update polydispersity parameter values |
|---|
| 219 | fast_index = k % stride[0] |
|---|
| 220 | if fast_index == 0: # bottom loop complete ... check all other loops |
|---|
| 221 | if weight.size > 0: |
|---|
| 222 | for i, index, in enumerate(k % stride): |
|---|
| 223 | args[pd_index[i]] = pd[i][0][index] |
|---|
| 224 | weight[i] = pd[i][1][index] |
|---|
| 225 | else: |
|---|
| 226 | args[pd_index[0]] = fast_value[fast_index] |
|---|
| 227 | weight[0] = fast_weight[fast_index] |
|---|
| 228 | |
|---|
| 229 | # Computes the weight, and if it is not sufficient then ignore this |
|---|
| 230 | # parameter set. |
|---|
| 231 | # Note: could precompute w1*...*wn so we only need to multiply by w0 |
|---|
| 232 | w = np.prod(weight) |
|---|
| 233 | if w > cutoff: |
|---|
| 234 | # Note: can precompute spherical correction if theta_index is not |
|---|
| 235 | # the fast index. Correction factor for spherical integration |
|---|
| 236 | #spherical_correction = abs(cos(pi*args[phi_index])) if phi_index>=0 else 1.0 |
|---|
| 237 | spherical_correction = (abs(cos(pi * args[theta_index])) * pi / 2 |
|---|
| 238 | if theta_index >= 0 else 1.0) |
|---|
| 239 | #spherical_correction = 1.0 |
|---|
| 240 | |
|---|
| 241 | # Call the scattering function and adds it to the total. |
|---|
| 242 | I = form(*args) |
|---|
| 243 | if np.isnan(I).any(): continue |
|---|
| 244 | ret += w * I * spherical_correction |
|---|
| 245 | norm += w |
|---|
| 246 | |
|---|
| 247 | # Volume normalization. |
|---|
| 248 | # If there are "volume" polydispersity parameters, then these |
|---|
| 249 | # will be used to call the form_volume function from the user |
|---|
| 250 | # supplied kernel, and accumulate a normalized weight. |
|---|
| 251 | # Note: can precompute volume norm if fast index is not a volume |
|---|
| 252 | if form_volume: |
|---|
| 253 | vol_args = [args[index] for index in vol_index] |
|---|
| 254 | vol_weight = np.prod(weight[vol_weight_index]) |
|---|
| 255 | vol += vol_weight * form_volume(*vol_args) |
|---|
| 256 | vol_norm += vol_weight |
|---|
| 257 | |
|---|
| 258 | positive = (vol * vol_norm != 0.0) |
|---|
| 259 | ret[positive] *= vol_norm[positive] / vol[positive] |
|---|
| 260 | result = scale * ret / norm + background |
|---|
| 261 | return result |
|---|
