[eafc9fa] | 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 | """ |
---|
[a738209] | 9 | from __future__ import division, print_function |
---|
| 10 | |
---|
[b139ee6] | 11 | import logging |
---|
| 12 | |
---|
[7ae2b7f] | 13 | import numpy as np # type: ignore |
---|
[6e7ba14] | 14 | from numpy import pi |
---|
| 15 | try: |
---|
| 16 | from numpy import cbrt |
---|
| 17 | except ImportError: |
---|
| 18 | def cbrt(x): return x ** (1.0/3.0) |
---|
[b3f6bc3] | 19 | |
---|
[c85db69] | 20 | from .generate import F64 |
---|
[f619de7] | 21 | from .kernel import KernelModel, Kernel |
---|
[b3f6bc3] | 22 | |
---|
[2d81cfe] | 23 | # pylint: disable=unused-import |
---|
[e62a134] | 24 | try: |
---|
[c1a888b] | 25 | from typing import Union, Callable |
---|
[b139ee6] | 26 | except ImportError: |
---|
[e62a134] | 27 | pass |
---|
| 28 | else: |
---|
[2d81cfe] | 29 | from . import details |
---|
[e62a134] | 30 | DType = Union[None, str, np.dtype] |
---|
[2d81cfe] | 31 | # pylint: enable=unused-import |
---|
[e62a134] | 32 | |
---|
[108e70e] | 33 | logger = logging.getLogger(__name__) |
---|
| 34 | |
---|
[3199b17] | 35 | |
---|
[f619de7] | 36 | class PyModel(KernelModel): |
---|
[eafc9fa] | 37 | """ |
---|
| 38 | Wrapper for pure python models. |
---|
| 39 | """ |
---|
[17bbadd] | 40 | def __init__(self, model_info): |
---|
[3199b17] | 41 | # Make sure Iq is available and vectorized. |
---|
[6d6508e] | 42 | _create_default_functions(model_info) |
---|
[17bbadd] | 43 | self.info = model_info |
---|
[883ecf4] | 44 | self.dtype = np.dtype('d') |
---|
[12eec1e] | 45 | logger.info("make python model " + self.info.name) |
---|
[3c56da87] | 46 | |
---|
[1e2a1ba] | 47 | def make_kernel(self, q_vectors): |
---|
[eafc9fa] | 48 | q_input = PyInput(q_vectors, dtype=F64) |
---|
[108e70e] | 49 | return PyKernel(self.info, q_input) |
---|
[3c56da87] | 50 | |
---|
[f734e7d] | 51 | def release(self): |
---|
[eafc9fa] | 52 | """ |
---|
| 53 | Free resources associated with the model. |
---|
| 54 | """ |
---|
[f734e7d] | 55 | pass |
---|
| 56 | |
---|
[3199b17] | 57 | |
---|
[b3f6bc3] | 58 | class PyInput(object): |
---|
| 59 | """ |
---|
| 60 | Make q data available to the gpu. |
---|
| 61 | |
---|
| 62 | *q_vectors* is a list of q vectors, which will be *[q]* for 1-D data, |
---|
| 63 | and *[qx, qy]* for 2-D data. Internally, the vectors will be reallocated |
---|
| 64 | to get the best performance on OpenCL, which may involve shifting and |
---|
| 65 | stretching the array to better match the memory architecture. Additional |
---|
| 66 | points will be evaluated with *q=1e-3*. |
---|
| 67 | |
---|
| 68 | *dtype* is the data type for the q vectors. The data type should be |
---|
| 69 | set to match that of the kernel, which is an attribute of |
---|
| 70 | :class:`GpuProgram`. Note that not all kernels support double |
---|
| 71 | precision, so even if the program was created for double precision, |
---|
| 72 | the *GpuProgram.dtype* may be single precision. |
---|
| 73 | |
---|
| 74 | Call :meth:`release` when complete. Even if not called directly, the |
---|
| 75 | buffer will be released when the data object is freed. |
---|
| 76 | """ |
---|
| 77 | def __init__(self, q_vectors, dtype): |
---|
| 78 | self.nq = q_vectors[0].size |
---|
| 79 | self.dtype = dtype |
---|
[eafc9fa] | 80 | self.is_2d = (len(q_vectors) == 2) |
---|
[48fbd50] | 81 | if self.is_2d: |
---|
| 82 | self.q = np.empty((self.nq, 2), dtype=dtype) |
---|
| 83 | self.q[:, 0] = q_vectors[0] |
---|
| 84 | self.q[:, 1] = q_vectors[1] |
---|
| 85 | else: |
---|
| 86 | self.q = np.empty(self.nq, dtype=dtype) |
---|
| 87 | self.q[:self.nq] = q_vectors[0] |
---|
[b3f6bc3] | 88 | |
---|
| 89 | def release(self): |
---|
[eafc9fa] | 90 | """ |
---|
| 91 | Free resources associated with the model inputs. |
---|
| 92 | """ |
---|
[48fbd50] | 93 | self.q = None |
---|
[b3f6bc3] | 94 | |
---|
[3199b17] | 95 | |
---|
[f619de7] | 96 | class PyKernel(Kernel): |
---|
[b3f6bc3] | 97 | """ |
---|
| 98 | Callable SAS kernel. |
---|
| 99 | |
---|
[108e70e] | 100 | *kernel* is the kernel object to call. |
---|
[b3f6bc3] | 101 | |
---|
[17bbadd] | 102 | *model_info* is the module information |
---|
[b3f6bc3] | 103 | |
---|
[3c56da87] | 104 | *q_input* is the DllInput q vectors at which the kernel should be |
---|
[b3f6bc3] | 105 | evaluated. |
---|
| 106 | |
---|
| 107 | The resulting call method takes the *pars*, a list of values for |
---|
| 108 | the fixed parameters to the kernel, and *pd_pars*, a list of (value,weight) |
---|
| 109 | vectors for the polydisperse parameters. *cutoff* determines the |
---|
| 110 | integration limits: any points with combined weight less than *cutoff* |
---|
| 111 | will not be calculated. |
---|
| 112 | |
---|
| 113 | Call :meth:`release` when done with the kernel instance. |
---|
| 114 | """ |
---|
[108e70e] | 115 | def __init__(self, model_info, q_input): |
---|
[bde38b5] | 116 | # type: (callable, ModelInfo, List[np.ndarray]) -> None |
---|
[6e7ff6d] | 117 | self.dtype = np.dtype('d') |
---|
[17bbadd] | 118 | self.info = model_info |
---|
[3c56da87] | 119 | self.q_input = q_input |
---|
| 120 | self.res = np.empty(q_input.nq, q_input.dtype) |
---|
[6e7ff6d] | 121 | self.dim = '2d' if q_input.is_2d else '1d' |
---|
| 122 | |
---|
[6d6508e] | 123 | partable = model_info.parameters |
---|
[8698a0d] | 124 | #kernel_parameters = (partable.iqxy_parameters if q_input.is_2d |
---|
| 125 | # else partable.iq_parameters) |
---|
| 126 | kernel_parameters = partable.iq_parameters |
---|
[6e7ff6d] | 127 | volume_parameters = partable.form_volume_parameters |
---|
| 128 | |
---|
| 129 | # Create an array to hold the parameter values. There will be a |
---|
| 130 | # single array whose values are updated as the calculator goes |
---|
| 131 | # through the loop. Arguments to the kernel and volume functions |
---|
| 132 | # will use views into this vector, relying on the fact that a |
---|
| 133 | # an array of no dimensions acts like a scalar. |
---|
[a8a7f08] | 134 | parameter_vector = np.empty(len(partable.call_parameters)-2, 'd') |
---|
[6e7ff6d] | 135 | |
---|
[3199b17] | 136 | # Create views into the array to hold the arguments. |
---|
[a8a7f08] | 137 | offset = 0 |
---|
[6e7ff6d] | 138 | kernel_args, volume_args = [], [] |
---|
| 139 | for p in partable.kernel_parameters: |
---|
| 140 | if p.length == 1: |
---|
| 141 | # Scalar values are length 1 vectors with no dimensions. |
---|
| 142 | v = parameter_vector[offset:offset+1].reshape(()) |
---|
[b3f6bc3] | 143 | else: |
---|
[6e7ff6d] | 144 | # Vector values are simple views. |
---|
| 145 | v = parameter_vector[offset:offset+p.length] |
---|
| 146 | offset += p.length |
---|
| 147 | if p in kernel_parameters: |
---|
| 148 | kernel_args.append(v) |
---|
| 149 | if p in volume_parameters: |
---|
[a8a7f08] | 150 | volume_args.append(v) |
---|
[6e7ff6d] | 151 | |
---|
| 152 | # Hold on to the parameter vector so we can use it to call kernel later. |
---|
| 153 | # This may also be required to preserve the views into the vector. |
---|
| 154 | self._parameter_vector = parameter_vector |
---|
| 155 | |
---|
| 156 | # Generate a closure which calls the kernel with the views into the |
---|
| 157 | # parameter array. |
---|
| 158 | if q_input.is_2d: |
---|
[6d6508e] | 159 | form = model_info.Iqxy |
---|
[40a87fa] | 160 | qx, qy = q_input.q[:, 0], q_input.q[:, 1] |
---|
[6e7ff6d] | 161 | self._form = lambda: form(qx, qy, *kernel_args) |
---|
[b3f6bc3] | 162 | else: |
---|
[6d6508e] | 163 | form = model_info.Iq |
---|
[6e7ff6d] | 164 | q = q_input.q |
---|
| 165 | self._form = lambda: form(q, *kernel_args) |
---|
| 166 | |
---|
| 167 | # Generate a closure which calls the form_volume if it exists. |
---|
[6e7ba14] | 168 | self._volume_args = volume_args |
---|
| 169 | volume = model_info.form_volume |
---|
[e44432d] | 170 | shell = model_info.shell_volume |
---|
[6e7ba14] | 171 | radius = model_info.effective_radius |
---|
[e44432d] | 172 | self._volume = ((lambda: (shell(*volume_args), volume(*volume_args))) if shell and volume |
---|
| 173 | else (lambda: [volume(*volume_args)]*2) if volume |
---|
| 174 | else (lambda: (1.0, 1.0))) |
---|
[6e7ba14] | 175 | self._radius = ((lambda mode: radius(mode, *volume_args)) if radius |
---|
| 176 | else (lambda mode: cbrt(0.75/pi*volume(*volume_args))) if volume |
---|
| 177 | else (lambda mode: 1.0)) |
---|
| 178 | |
---|
| 179 | def _call_kernel(self, call_details, values, cutoff, magnetic, effective_radius_type): |
---|
[bde38b5] | 180 | # type: (CallDetails, np.ndarray, np.ndarray, float, bool) -> np.ndarray |
---|
| 181 | if magnetic: |
---|
| 182 | raise NotImplementedError("Magnetism not implemented for pure python models") |
---|
| 183 | #print("Calling python kernel") |
---|
| 184 | #call_details.show(values) |
---|
[6e7ba14] | 185 | radius = ((lambda: 0.0) if effective_radius_type == 0 |
---|
| 186 | else (lambda: self._radius(effective_radius_type))) |
---|
[5399809] | 187 | self.result = _loops( |
---|
| 188 | self._parameter_vector, self._form, self._volume, radius, |
---|
| 189 | self.q_input.nq, call_details, values, cutoff) |
---|
[b3f6bc3] | 190 | |
---|
| 191 | def release(self): |
---|
[bde38b5] | 192 | # type: () -> None |
---|
[eafc9fa] | 193 | """ |
---|
| 194 | Free resources associated with the kernel. |
---|
| 195 | """ |
---|
[f619de7] | 196 | self.q_input.release() |
---|
[3c56da87] | 197 | self.q_input = None |
---|
[b3f6bc3] | 198 | |
---|
[3199b17] | 199 | |
---|
[2d81cfe] | 200 | def _loops(parameters, # type: np.ndarray |
---|
| 201 | form, # type: Callable[[], np.ndarray] |
---|
| 202 | form_volume, # type: Callable[[], float] |
---|
[6e7ba14] | 203 | form_radius, # type: Callable[[], float] |
---|
[2d81cfe] | 204 | nq, # type: int |
---|
| 205 | call_details, # type: details.CallDetails |
---|
| 206 | values, # type: np.ndarray |
---|
[6e7ba14] | 207 | cutoff, # type: float |
---|
[2d81cfe] | 208 | ): |
---|
| 209 | # type: (...) -> None |
---|
[f734e7d] | 210 | ################################################################ |
---|
| 211 | # # |
---|
| 212 | # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # |
---|
| 213 | # !! !! # |
---|
| 214 | # !! KEEP THIS CODE CONSISTENT WITH KERNEL_TEMPLATE.C !! # |
---|
| 215 | # !! !! # |
---|
| 216 | # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # |
---|
| 217 | # # |
---|
| 218 | ################################################################ |
---|
[5399809] | 219 | |
---|
| 220 | # WARNING: Trickery ahead |
---|
| 221 | # The parameters[] vector is embedded in the closures for form(), |
---|
| 222 | # form_volume() and form_radius(). We set the initial vector from |
---|
| 223 | # the values for the model parameters. As we loop through the polydispesity |
---|
| 224 | # mesh, we update the components with the polydispersity values before |
---|
| 225 | # calling the respective functions. |
---|
[40a87fa] | 226 | n_pars = len(parameters) |
---|
| 227 | parameters[:] = values[2:n_pars+2] |
---|
[5399809] | 228 | |
---|
[40a87fa] | 229 | if call_details.num_active == 0: |
---|
[5399809] | 230 | total = form() |
---|
| 231 | weight_norm = 1.0 |
---|
[e44432d] | 232 | weighted_shell, weighted_form = form_volume() |
---|
[5399809] | 233 | weighted_radius = form_radius() |
---|
| 234 | |
---|
| 235 | else: |
---|
| 236 | pd_value = values[2+n_pars:2+n_pars + call_details.num_weights] |
---|
| 237 | pd_weight = values[2+n_pars + call_details.num_weights:] |
---|
| 238 | |
---|
| 239 | weight_norm = 0.0 |
---|
[e44432d] | 240 | weighted_form = 0.0 |
---|
| 241 | weighted_shell = 0.0 |
---|
[5399809] | 242 | weighted_radius = 0.0 |
---|
| 243 | partial_weight = np.NaN |
---|
| 244 | weight = np.NaN |
---|
| 245 | |
---|
| 246 | p0_par = call_details.pd_par[0] |
---|
| 247 | p0_length = call_details.pd_length[0] |
---|
| 248 | p0_index = p0_length |
---|
| 249 | p0_offset = call_details.pd_offset[0] |
---|
| 250 | |
---|
| 251 | pd_par = call_details.pd_par[:call_details.num_active] |
---|
| 252 | pd_offset = call_details.pd_offset[:call_details.num_active] |
---|
| 253 | pd_stride = call_details.pd_stride[:call_details.num_active] |
---|
| 254 | pd_length = call_details.pd_length[:call_details.num_active] |
---|
| 255 | |
---|
| 256 | total = np.zeros(nq, 'd') |
---|
| 257 | for loop_index in range(call_details.num_eval): |
---|
[3199b17] | 258 | # Update polydispersity parameter values. |
---|
[5399809] | 259 | if p0_index == p0_length: |
---|
| 260 | pd_index = (loop_index//pd_stride)%pd_length |
---|
| 261 | parameters[pd_par] = pd_value[pd_offset+pd_index] |
---|
| 262 | partial_weight = np.prod(pd_weight[pd_offset+pd_index][1:]) |
---|
| 263 | p0_index = loop_index%p0_length |
---|
| 264 | |
---|
| 265 | weight = partial_weight * pd_weight[p0_offset + p0_index] |
---|
| 266 | parameters[p0_par] = pd_value[p0_offset + p0_index] |
---|
| 267 | p0_index += 1 |
---|
| 268 | if weight > cutoff: |
---|
[3199b17] | 269 | # Call the scattering function. |
---|
[5399809] | 270 | # Assume that NaNs are only generated if the parameters are bad; |
---|
| 271 | # exclude all q for that NaN. Even better would be to have an |
---|
| 272 | # INVALID expression like the C models, but that is expensive. |
---|
| 273 | Iq = np.asarray(form(), 'd') |
---|
| 274 | if np.isnan(Iq).any(): |
---|
| 275 | continue |
---|
| 276 | |
---|
[3199b17] | 277 | # Update value and norm. |
---|
[5399809] | 278 | total += weight * Iq |
---|
| 279 | weight_norm += weight |
---|
[e44432d] | 280 | shell, form = form_volume() |
---|
| 281 | weighted_form += weight * form |
---|
| 282 | weighted_shell += weight * shell |
---|
[5399809] | 283 | weighted_radius += weight * form_radius() |
---|
[b3f6bc3] | 284 | |
---|
[e44432d] | 285 | result = np.hstack((total, weight_norm, weighted_form, weighted_shell, weighted_radius)) |
---|
[6e7ba14] | 286 | return result |
---|
[6d6508e] | 287 | |
---|
| 288 | |
---|
[4bfd277] | 289 | def _create_default_functions(model_info): |
---|
| 290 | """ |
---|
| 291 | Autogenerate missing functions, such as Iqxy from Iq. |
---|
| 292 | |
---|
| 293 | This only works for Iqxy when Iq is written in python. :func:`make_source` |
---|
| 294 | performs a similar role for Iq written in C. This also vectorizes |
---|
| 295 | any functions that are not already marked as vectorized. |
---|
| 296 | """ |
---|
[3199b17] | 297 | # Note: Must call create_vector_Iq before create_vector_Iqxy. |
---|
[4bfd277] | 298 | _create_vector_Iq(model_info) |
---|
[108e70e] | 299 | _create_vector_Iqxy(model_info) |
---|
[4bfd277] | 300 | |
---|
[6d6508e] | 301 | |
---|
| 302 | def _create_vector_Iq(model_info): |
---|
| 303 | """ |
---|
| 304 | Define Iq as a vector function if it exists. |
---|
| 305 | """ |
---|
| 306 | Iq = model_info.Iq |
---|
| 307 | if callable(Iq) and not getattr(Iq, 'vectorized', False): |
---|
[4bfd277] | 308 | #print("vectorizing Iq") |
---|
[6d6508e] | 309 | def vector_Iq(q, *args): |
---|
| 310 | """ |
---|
| 311 | Vectorized 1D kernel. |
---|
| 312 | """ |
---|
| 313 | return np.array([Iq(qi, *args) for qi in q]) |
---|
| 314 | vector_Iq.vectorized = True |
---|
| 315 | model_info.Iq = vector_Iq |
---|
| 316 | |
---|
[108e70e] | 317 | |
---|
[6d6508e] | 318 | def _create_vector_Iqxy(model_info): |
---|
| 319 | """ |
---|
| 320 | Define Iqxy as a vector function if it exists, or default it from Iq(). |
---|
| 321 | """ |
---|
[108e70e] | 322 | Iqxy = getattr(model_info, 'Iqxy', None) |
---|
[4bfd277] | 323 | if callable(Iqxy): |
---|
| 324 | if not getattr(Iqxy, 'vectorized', False): |
---|
| 325 | #print("vectorizing Iqxy") |
---|
| 326 | def vector_Iqxy(qx, qy, *args): |
---|
| 327 | """ |
---|
| 328 | Vectorized 2D kernel. |
---|
| 329 | """ |
---|
| 330 | return np.array([Iqxy(qxi, qyi, *args) for qxi, qyi in zip(qx, qy)]) |
---|
| 331 | vector_Iqxy.vectorized = True |
---|
| 332 | model_info.Iqxy = vector_Iqxy |
---|
[108e70e] | 333 | else: |
---|
[4bfd277] | 334 | #print("defaulting Iqxy") |
---|
[6d6508e] | 335 | # Iq is vectorized because create_vector_Iq was already called. |
---|
[108e70e] | 336 | Iq = model_info.Iq |
---|
[6d6508e] | 337 | def default_Iqxy(qx, qy, *args): |
---|
| 338 | """ |
---|
| 339 | Default 2D kernel. |
---|
| 340 | """ |
---|
| 341 | return Iq(np.sqrt(qx**2 + qy**2), *args) |
---|
| 342 | default_Iqxy.vectorized = True |
---|
| 343 | model_info.Iqxy = default_Iqxy |
---|