[fd5ac0d] | 1 | """ |
---|
| 2 | Handle Q smearing |
---|
| 3 | """ |
---|
| 4 | ##################################################################### |
---|
| 5 | #This software was developed by the University of Tennessee as part of the |
---|
| 6 | #Distributed Data Analysis of Neutron Scattering Experiments (DANSE) |
---|
| 7 | #project funded by the US National Science Foundation. |
---|
| 8 | #See the license text in license.txt |
---|
| 9 | #copyright 2008, University of Tennessee |
---|
| 10 | ###################################################################### |
---|
| 11 | import numpy |
---|
| 12 | import math |
---|
| 13 | import logging |
---|
| 14 | import sys |
---|
| 15 | import sas.models.sas_extension.smearer as smearer |
---|
| 16 | from sas.models.smearing_2d import Smearer2D |
---|
| 17 | |
---|
| 18 | def smear_selection(data1D, model = None): |
---|
| 19 | """ |
---|
| 20 | Creates the right type of smearer according |
---|
| 21 | to the data. |
---|
| 22 | |
---|
| 23 | The canSAS format has a rule that either |
---|
| 24 | slit smearing data OR resolution smearing data |
---|
| 25 | is available. |
---|
| 26 | |
---|
| 27 | For the present purpose, we choose the one that |
---|
| 28 | has none-zero data. If both slit and resolution |
---|
| 29 | smearing arrays are filled with good data |
---|
| 30 | (which should not happen), then we choose the |
---|
| 31 | resolution smearing data. |
---|
| 32 | |
---|
| 33 | :param data1D: Data1D object |
---|
| 34 | :param model: sas.model instance |
---|
| 35 | """ |
---|
| 36 | # Sanity check. If we are not dealing with a SAS Data1D |
---|
| 37 | # object, just return None |
---|
| 38 | if data1D.__class__.__name__ not in ['Data1D', 'Theory1D']: |
---|
| 39 | if data1D == None: |
---|
| 40 | return None |
---|
| 41 | elif data1D.dqx_data == None or data1D.dqy_data == None: |
---|
| 42 | return None |
---|
| 43 | return Smearer2D(data1D) |
---|
| 44 | |
---|
| 45 | if not hasattr(data1D, "dx") and not hasattr(data1D, "dxl")\ |
---|
| 46 | and not hasattr(data1D, "dxw"): |
---|
| 47 | return None |
---|
| 48 | |
---|
| 49 | # Look for resolution smearing data |
---|
| 50 | _found_resolution = False |
---|
| 51 | if data1D.dx is not None and len(data1D.dx) == len(data1D.x): |
---|
| 52 | |
---|
| 53 | # Check that we have non-zero data |
---|
| 54 | if data1D.dx[0] > 0.0: |
---|
| 55 | _found_resolution = True |
---|
| 56 | #print "_found_resolution",_found_resolution |
---|
| 57 | #print "data1D.dx[0]",data1D.dx[0],data1D.dxl[0] |
---|
| 58 | # If we found resolution smearing data, return a QSmearer |
---|
| 59 | if _found_resolution == True: |
---|
| 60 | return QSmearer(data1D, model) |
---|
[a3f125f0] | 61 | #return pinhole_smear(data1D, model) |
---|
[fd5ac0d] | 62 | |
---|
| 63 | # Look for slit smearing data |
---|
| 64 | _found_slit = False |
---|
| 65 | if data1D.dxl is not None and len(data1D.dxl) == len(data1D.x) \ |
---|
| 66 | and data1D.dxw is not None and len(data1D.dxw) == len(data1D.x): |
---|
| 67 | |
---|
| 68 | # Check that we have non-zero data |
---|
| 69 | if data1D.dxl[0] > 0.0 or data1D.dxw[0] > 0.0: |
---|
| 70 | _found_slit = True |
---|
| 71 | |
---|
| 72 | # Sanity check: all data should be the same as a function of Q |
---|
| 73 | for item in data1D.dxl: |
---|
| 74 | if data1D.dxl[0] != item: |
---|
| 75 | _found_resolution = False |
---|
| 76 | break |
---|
| 77 | |
---|
| 78 | for item in data1D.dxw: |
---|
| 79 | if data1D.dxw[0] != item: |
---|
| 80 | _found_resolution = False |
---|
| 81 | break |
---|
| 82 | # If we found slit smearing data, return a slit smearer |
---|
| 83 | if _found_slit == True: |
---|
[a3f125f0] | 84 | #return SlitSmearer(data1D, model) |
---|
| 85 | return slit_smear(data1D, model) |
---|
[fd5ac0d] | 86 | return None |
---|
| 87 | |
---|
| 88 | |
---|
| 89 | class _BaseSmearer(object): |
---|
| 90 | """ |
---|
| 91 | Base class for smearers |
---|
| 92 | """ |
---|
| 93 | def __init__(self): |
---|
| 94 | self.nbins = 0 |
---|
| 95 | self.nbins_low = 0 |
---|
| 96 | self.nbins_high = 0 |
---|
| 97 | self._weights = None |
---|
| 98 | ## Internal flag to keep track of C++ smearer initialization |
---|
| 99 | self._init_complete = False |
---|
| 100 | self._smearer = None |
---|
| 101 | self.model = None |
---|
| 102 | self.min = None |
---|
| 103 | self.max = None |
---|
| 104 | self.qvalues = [] |
---|
| 105 | |
---|
| 106 | def __deepcopy__(self, memo=None): |
---|
| 107 | """ |
---|
| 108 | Return a valid copy of self. |
---|
| 109 | Avoid copying the _smearer C object and force a matrix recompute |
---|
| 110 | when the copy is used. |
---|
| 111 | """ |
---|
| 112 | result = _BaseSmearer() |
---|
| 113 | result.nbins = self.nbins |
---|
| 114 | return result |
---|
| 115 | |
---|
| 116 | def _compute_matrix(self): |
---|
| 117 | """ |
---|
| 118 | Place holder for matrix computation |
---|
| 119 | """ |
---|
| 120 | return NotImplemented |
---|
| 121 | |
---|
| 122 | def get_unsmeared_range(self, q_min=None, q_max=None): |
---|
| 123 | """ |
---|
| 124 | Place holder for method returning unsmeared range |
---|
| 125 | """ |
---|
| 126 | return NotImplemented |
---|
| 127 | |
---|
| 128 | def get_bin_range(self, q_min=None, q_max=None): |
---|
| 129 | """ |
---|
| 130 | |
---|
| 131 | :param q_min: minimum q-value to smear |
---|
| 132 | :param q_max: maximum q-value to smear |
---|
| 133 | |
---|
| 134 | """ |
---|
| 135 | # If this is the first time we call for smearing, |
---|
| 136 | # initialize the C++ smearer object first |
---|
| 137 | if not self._init_complete: |
---|
| 138 | self._initialize_smearer() |
---|
| 139 | if q_min == None: |
---|
| 140 | q_min = self.min |
---|
| 141 | if q_max == None: |
---|
| 142 | q_max = self.max |
---|
| 143 | |
---|
| 144 | _qmin_unsmeared, _qmax_unsmeared = self.get_unsmeared_range(q_min, |
---|
| 145 | q_max) |
---|
| 146 | _first_bin = None |
---|
| 147 | _last_bin = None |
---|
| 148 | |
---|
| 149 | #step = (self.max - self.min) / (self.nbins - 1.0) |
---|
| 150 | # Find the first and last bin number in all extrapolated and real data |
---|
| 151 | try: |
---|
| 152 | for i in range(self.nbins): |
---|
| 153 | q_i = smearer.get_q(self._smearer, i) |
---|
| 154 | if (q_i >= _qmin_unsmeared) and (q_i <= _qmax_unsmeared): |
---|
| 155 | # Identify first and last bin |
---|
| 156 | if _first_bin is None: |
---|
| 157 | _first_bin = i |
---|
| 158 | else: |
---|
| 159 | _last_bin = i |
---|
| 160 | except: |
---|
| 161 | msg = "_BaseSmearer.get_bin_range: " |
---|
| 162 | msg += " error getting range\n %s" % sys.exc_value |
---|
| 163 | raise RuntimeError, msg |
---|
| 164 | |
---|
| 165 | # Find the first and last bin number only in the real data |
---|
| 166 | _first_bin, _last_bin = self._get_unextrapolated_bin( \ |
---|
| 167 | _first_bin, _last_bin) |
---|
| 168 | |
---|
| 169 | return _first_bin, _last_bin |
---|
| 170 | |
---|
| 171 | def __call__(self, iq_in, first_bin = 0, last_bin = None): |
---|
| 172 | """ |
---|
| 173 | Perform smearing |
---|
| 174 | """ |
---|
| 175 | # If this is the first time we call for smearing, |
---|
| 176 | # initialize the C++ smearer object first |
---|
| 177 | if not self._init_complete: |
---|
| 178 | self._initialize_smearer() |
---|
| 179 | |
---|
| 180 | if last_bin is None or last_bin >= len(iq_in): |
---|
| 181 | last_bin = len(iq_in) - 1 |
---|
| 182 | # Check that the first bin is positive |
---|
| 183 | if first_bin < 0: |
---|
| 184 | first_bin = 0 |
---|
| 185 | |
---|
| 186 | # With a model given, compute I for the extrapolated points and append |
---|
| 187 | # to the iq_in |
---|
| 188 | iq_in_temp = iq_in |
---|
| 189 | if self.model != None: |
---|
| 190 | temp_first, temp_last = self._get_extrapolated_bin( \ |
---|
| 191 | first_bin, last_bin) |
---|
| 192 | if self.nbins_low > 0: |
---|
| 193 | iq_in_low = self.model.evalDistribution( \ |
---|
| 194 | numpy.fabs(self.qvalues[0:self.nbins_low])) |
---|
| 195 | iq_in_high = self.model.evalDistribution( \ |
---|
| 196 | self.qvalues[(len(self.qvalues) - \ |
---|
| 197 | self.nbins_high - 1):]) |
---|
| 198 | # Todo: find out who is sending iq[last_poin] = 0. |
---|
| 199 | if iq_in[len(iq_in) - 1] == 0: |
---|
| 200 | iq_in[len(iq_in) - 1] = iq_in_high[0] |
---|
| 201 | # Append the extrapolated points to the data points |
---|
[a3f125f0] | 202 | if self.nbins_low > 0: |
---|
[fd5ac0d] | 203 | iq_in_temp = numpy.append(iq_in_low, iq_in) |
---|
| 204 | if self.nbins_high > 0: |
---|
| 205 | iq_in_temp = numpy.append(iq_in_temp, iq_in_high[1:]) |
---|
| 206 | else: |
---|
| 207 | temp_first = first_bin |
---|
| 208 | temp_last = last_bin |
---|
| 209 | #iq_in_temp = iq_in |
---|
| 210 | |
---|
| 211 | # Sanity check |
---|
| 212 | if len(iq_in_temp) != self.nbins: |
---|
| 213 | msg = "Invalid I(q) vector: inconsistent array " |
---|
| 214 | msg += " length %d != %s" % (len(iq_in_temp), str(self.nbins)) |
---|
| 215 | raise RuntimeError, msg |
---|
| 216 | |
---|
| 217 | # Storage for smeared I(q) |
---|
| 218 | iq_out = numpy.zeros(self.nbins) |
---|
| 219 | |
---|
| 220 | smear_output = smearer.smear(self._smearer, iq_in_temp, iq_out, |
---|
| 221 | #0, self.nbins - 1) |
---|
| 222 | temp_first, temp_last) |
---|
| 223 | #first_bin, last_bin) |
---|
| 224 | if smear_output < 0: |
---|
| 225 | msg = "_BaseSmearer: could not smear, code = %g" % smear_output |
---|
| 226 | raise RuntimeError, msg |
---|
| 227 | |
---|
| 228 | temp_first = first_bin + self.nbins_low |
---|
| 229 | temp_last = self.nbins - self.nbins_high |
---|
| 230 | out = iq_out[temp_first: temp_last] |
---|
| 231 | |
---|
| 232 | return out |
---|
| 233 | |
---|
| 234 | def _initialize_smearer(self): |
---|
| 235 | """ |
---|
| 236 | Place holder for initializing data smearer |
---|
| 237 | """ |
---|
| 238 | return NotImplemented |
---|
| 239 | |
---|
| 240 | |
---|
| 241 | def _get_unextrapolated_bin(self, first_bin = 0, last_bin = 0): |
---|
| 242 | """ |
---|
| 243 | Get unextrapolated first bin and the last bin |
---|
| 244 | |
---|
| 245 | : param first_bin: extrapolated first_bin |
---|
| 246 | : param last_bin: extrapolated last_bin |
---|
| 247 | |
---|
| 248 | : return fist_bin, last_bin: unextrapolated first and last bin |
---|
| 249 | """ |
---|
| 250 | # For first bin |
---|
| 251 | if first_bin <= self.nbins_low: |
---|
| 252 | first_bin = 0 |
---|
| 253 | else: |
---|
| 254 | first_bin = first_bin - self.nbins_low |
---|
| 255 | # For last bin |
---|
| 256 | if last_bin >= (self.nbins - self.nbins_high): |
---|
| 257 | last_bin = self.nbins - (self.nbins_high + self.nbins_low + 1) |
---|
| 258 | elif last_bin >= self.nbins_low: |
---|
| 259 | last_bin = last_bin - self.nbins_low |
---|
| 260 | else: |
---|
| 261 | last_bin = 0 |
---|
| 262 | return first_bin, last_bin |
---|
| 263 | |
---|
| 264 | def _get_extrapolated_bin(self, first_bin = 0, last_bin = 0): |
---|
| 265 | """ |
---|
| 266 | Get extrapolated first bin and the last bin |
---|
| 267 | |
---|
| 268 | : param first_bin: unextrapolated first_bin |
---|
| 269 | : param last_bin: unextrapolated last_bin |
---|
| 270 | |
---|
| 271 | : return first_bin, last_bin: extrapolated first and last bin |
---|
| 272 | """ |
---|
| 273 | # For the first bin |
---|
| 274 | # In the case that needs low extrapolation data |
---|
| 275 | first_bin = 0 |
---|
| 276 | # For last bin |
---|
| 277 | if last_bin >= self.nbins - (self.nbins_high + self.nbins_low + 1): |
---|
| 278 | # In the case that needs higher q extrapolation data |
---|
| 279 | last_bin = self.nbins - 1 |
---|
| 280 | else: |
---|
| 281 | # In the case that doesn't need higher q extrapolation data |
---|
| 282 | last_bin += self.nbins_low |
---|
| 283 | |
---|
| 284 | return first_bin, last_bin |
---|
| 285 | |
---|
| 286 | class _SlitSmearer(_BaseSmearer): |
---|
| 287 | """ |
---|
| 288 | Slit smearing for I(q) array |
---|
| 289 | """ |
---|
| 290 | |
---|
| 291 | def __init__(self, nbins=None, width=None, height=None, min=None, max=None): |
---|
| 292 | """ |
---|
| 293 | Initialization |
---|
| 294 | |
---|
| 295 | :param iq: I(q) array [cm-1] |
---|
| 296 | :param width: slit width [A-1] |
---|
| 297 | :param height: slit height [A-1] |
---|
| 298 | :param min: Q_min [A-1] |
---|
| 299 | :param max: Q_max [A-1] |
---|
| 300 | |
---|
| 301 | """ |
---|
| 302 | _BaseSmearer.__init__(self) |
---|
| 303 | ## Slit width in Q units |
---|
| 304 | self.width = width |
---|
| 305 | ## Slit height in Q units |
---|
| 306 | self.height = height |
---|
| 307 | ## Q_min (Min Q-value for I(q)) |
---|
| 308 | self.min = min |
---|
| 309 | ## Q_max (Max Q_value for I(q)) |
---|
| 310 | self.max = max |
---|
| 311 | ## Number of Q bins |
---|
| 312 | self.nbins = nbins |
---|
| 313 | ## Number of points used in the smearing computation |
---|
| 314 | self.npts = 3000 |
---|
| 315 | ## Smearing matrix |
---|
| 316 | self._weights = None |
---|
| 317 | self.qvalues = None |
---|
| 318 | |
---|
| 319 | def _initialize_smearer(self): |
---|
| 320 | """ |
---|
| 321 | Initialize the C++ smearer object. |
---|
| 322 | This method HAS to be called before smearing |
---|
| 323 | """ |
---|
| 324 | #self._smearer = smearer.new_slit_smearer(self.width, |
---|
| 325 | # self.height, self.min, self.max, self.nbins) |
---|
| 326 | self._smearer = smearer.new_slit_smearer_with_q(self.width, |
---|
| 327 | self.height, self.qvalues) |
---|
| 328 | self._init_complete = True |
---|
| 329 | |
---|
| 330 | def get_unsmeared_range(self, q_min, q_max): |
---|
| 331 | """ |
---|
| 332 | Determine the range needed in unsmeared-Q to cover |
---|
| 333 | the smeared Q range |
---|
| 334 | """ |
---|
| 335 | # Range used for input to smearing |
---|
| 336 | _qmin_unsmeared = q_min |
---|
| 337 | _qmax_unsmeared = q_max |
---|
| 338 | try: |
---|
| 339 | _qmin_unsmeared = self.min |
---|
| 340 | _qmax_unsmeared = self.max |
---|
| 341 | except: |
---|
| 342 | logging.error("_SlitSmearer.get_bin_range: %s" % sys.exc_value) |
---|
| 343 | return _qmin_unsmeared, _qmax_unsmeared |
---|
| 344 | |
---|
| 345 | class SlitSmearer(_SlitSmearer): |
---|
| 346 | """ |
---|
| 347 | Adaptor for slit smearing class and SAS data |
---|
| 348 | """ |
---|
| 349 | def __init__(self, data1D, model = None): |
---|
| 350 | """ |
---|
| 351 | Assumption: equally spaced bins of increasing q-values. |
---|
| 352 | |
---|
| 353 | :param data1D: data used to set the smearing parameters |
---|
| 354 | """ |
---|
| 355 | # Initialization from parent class |
---|
| 356 | super(SlitSmearer, self).__init__() |
---|
| 357 | |
---|
| 358 | ## Slit width |
---|
| 359 | self.width = 0 |
---|
| 360 | self.nbins_low = 0 |
---|
| 361 | self.nbins_high = 0 |
---|
| 362 | self.model = model |
---|
| 363 | if data1D.dxw is not None and len(data1D.dxw) == len(data1D.x): |
---|
| 364 | self.width = data1D.dxw[0] |
---|
| 365 | # Sanity check |
---|
| 366 | for value in data1D.dxw: |
---|
| 367 | if value != self.width: |
---|
| 368 | msg = "Slit smearing parameters must " |
---|
| 369 | msg += " be the same for all data" |
---|
| 370 | raise RuntimeError, msg |
---|
| 371 | ## Slit height |
---|
| 372 | self.height = 0 |
---|
| 373 | if data1D.dxl is not None and len(data1D.dxl) == len(data1D.x): |
---|
| 374 | self.height = data1D.dxl[0] |
---|
| 375 | # Sanity check |
---|
| 376 | for value in data1D.dxl: |
---|
| 377 | if value != self.height: |
---|
| 378 | msg = "Slit smearing parameters must be" |
---|
| 379 | msg += " the same for all data" |
---|
| 380 | raise RuntimeError, msg |
---|
| 381 | # If a model is given, get the q extrapolation |
---|
| 382 | if self.model == None: |
---|
| 383 | data1d_x = data1D.x |
---|
| 384 | else: |
---|
| 385 | # Take larger sigma |
---|
| 386 | if self.height > self.width: |
---|
| 387 | # The denominator (2.0) covers all the possible w^2 + h^2 range |
---|
| 388 | sigma_in = data1D.dxl / 2.0 |
---|
| 389 | elif self.width > 0: |
---|
| 390 | sigma_in = data1D.dxw / 2.0 |
---|
| 391 | else: |
---|
| 392 | sigma_in = [] |
---|
| 393 | |
---|
| 394 | self.nbins_low, self.nbins_high, _, data1d_x = \ |
---|
| 395 | get_qextrapolate(sigma_in, data1D.x) |
---|
| 396 | |
---|
| 397 | ## Number of Q bins |
---|
| 398 | self.nbins = len(data1d_x) |
---|
| 399 | ## Minimum Q |
---|
| 400 | self.min = min(data1d_x) |
---|
| 401 | ## Maximum |
---|
| 402 | self.max = max(data1d_x) |
---|
| 403 | ## Q-values |
---|
| 404 | self.qvalues = data1d_x |
---|
| 405 | |
---|
| 406 | |
---|
| 407 | class _QSmearer(_BaseSmearer): |
---|
| 408 | """ |
---|
| 409 | Perform Gaussian Q smearing |
---|
| 410 | """ |
---|
| 411 | |
---|
| 412 | def __init__(self, nbins=None, width=None, min=None, max=None): |
---|
| 413 | """ |
---|
| 414 | Initialization |
---|
| 415 | |
---|
| 416 | :param nbins: number of Q bins |
---|
| 417 | :param width: array standard deviation in Q [A-1] |
---|
| 418 | :param min: Q_min [A-1] |
---|
| 419 | :param max: Q_max [A-1] |
---|
| 420 | """ |
---|
| 421 | _BaseSmearer.__init__(self) |
---|
| 422 | ## Standard deviation in Q [A-1] |
---|
| 423 | self.width = width |
---|
| 424 | ## Q_min (Min Q-value for I(q)) |
---|
| 425 | self.min = min |
---|
| 426 | ## Q_max (Max Q_value for I(q)) |
---|
| 427 | self.max = max |
---|
| 428 | ## Number of Q bins |
---|
| 429 | self.nbins = nbins |
---|
| 430 | ## Smearing matrix |
---|
| 431 | self._weights = None |
---|
| 432 | self.qvalues = None |
---|
| 433 | |
---|
| 434 | def _initialize_smearer(self): |
---|
| 435 | """ |
---|
| 436 | Initialize the C++ smearer object. |
---|
| 437 | This method HAS to be called before smearing |
---|
| 438 | """ |
---|
| 439 | #self._smearer = smearer.new_q_smearer(numpy.asarray(self.width), |
---|
| 440 | # self.min, self.max, self.nbins) |
---|
| 441 | self._smearer = smearer.new_q_smearer_with_q(numpy.asarray(self.width), |
---|
| 442 | self.qvalues) |
---|
| 443 | self._init_complete = True |
---|
| 444 | |
---|
| 445 | def get_unsmeared_range(self, q_min, q_max): |
---|
| 446 | """ |
---|
| 447 | Determine the range needed in unsmeared-Q to cover |
---|
| 448 | the smeared Q range |
---|
| 449 | Take 3 sigmas as the offset between smeared and unsmeared space |
---|
| 450 | """ |
---|
| 451 | # Range used for input to smearing |
---|
| 452 | _qmin_unsmeared = q_min |
---|
| 453 | _qmax_unsmeared = q_max |
---|
| 454 | try: |
---|
| 455 | offset = 3.0 * max(self.width) |
---|
| 456 | _qmin_unsmeared = self.min#max([self.min, q_min - offset]) |
---|
| 457 | _qmax_unsmeared = self.max#min([self.max, q_max + offset]) |
---|
| 458 | except: |
---|
| 459 | logging.error("_QSmearer.get_bin_range: %s" % sys.exc_value) |
---|
| 460 | return _qmin_unsmeared, _qmax_unsmeared |
---|
| 461 | |
---|
| 462 | |
---|
| 463 | class QSmearer(_QSmearer): |
---|
| 464 | """ |
---|
| 465 | Adaptor for Gaussian Q smearing class and SAS data |
---|
| 466 | """ |
---|
| 467 | def __init__(self, data1D, model = None): |
---|
| 468 | """ |
---|
| 469 | Assumption: equally spaced bins of increasing q-values. |
---|
| 470 | |
---|
| 471 | :param data1D: data used to set the smearing parameters |
---|
| 472 | """ |
---|
| 473 | # Initialization from parent class |
---|
| 474 | super(QSmearer, self).__init__() |
---|
| 475 | data1d_x = [] |
---|
| 476 | self.nbins_low = 0 |
---|
| 477 | self.nbins_high = 0 |
---|
| 478 | self.model = model |
---|
| 479 | ## Resolution |
---|
| 480 | #self.width = numpy.zeros(len(data1D.x)) |
---|
| 481 | if data1D.dx is not None and len(data1D.dx) == len(data1D.x): |
---|
| 482 | self.width = data1D.dx |
---|
| 483 | |
---|
| 484 | if self.model == None: |
---|
| 485 | data1d_x = data1D.x |
---|
| 486 | else: |
---|
| 487 | self.nbins_low, self.nbins_high, self.width, data1d_x = \ |
---|
| 488 | get_qextrapolate(self.width, data1D.x) |
---|
| 489 | |
---|
| 490 | ## Number of Q bins |
---|
| 491 | self.nbins = len(data1d_x) |
---|
| 492 | ## Minimum Q |
---|
| 493 | self.min = min(data1d_x) |
---|
| 494 | ## Maximum |
---|
| 495 | self.max = max(data1d_x) |
---|
| 496 | ## Q-values |
---|
| 497 | self.qvalues = data1d_x |
---|
| 498 | |
---|
| 499 | |
---|
| 500 | def get_qextrapolate(width, data_x): |
---|
| 501 | """ |
---|
| 502 | Make fake data_x points extrapolated outside of the data_x points |
---|
| 503 | |
---|
| 504 | :param width: array of std of q resolution |
---|
| 505 | :param Data1D.x: Data1D.x array |
---|
| 506 | |
---|
| 507 | :return new_width, data_x_ext: extrapolated width array and x array |
---|
| 508 | |
---|
| 509 | :assumption1: data_x is ordered from lower q to higher q |
---|
| 510 | :assumption2: len(data) = len(width) |
---|
| 511 | :assumption3: the distance between the data points is more compact than the size of width |
---|
| 512 | :Todo1: Make sure that the assumptions are correct for Data1D |
---|
| 513 | :Todo2: This fixes the edge problem in Qsmearer but still needs to make smearer interface |
---|
| 514 | """ |
---|
| 515 | # Length of the width |
---|
| 516 | length = len(width) |
---|
| 517 | width_low = math.fabs(width[0]) |
---|
| 518 | width_high = math.fabs(width[length -1]) |
---|
| 519 | nbins_low = 0.0 |
---|
| 520 | nbins_high = 0.0 |
---|
| 521 | # Compare width(dQ) to the data bin size and take smaller one as the bin |
---|
| 522 | # size of the extrapolation; this will correct some weird behavior |
---|
| 523 | # at the edge: This method was out (commented) |
---|
| 524 | # because it becomes very expansive when |
---|
| 525 | # bin size is very small comparing to the width. |
---|
| 526 | # Now on, we will just give the bin size of the extrapolated points |
---|
| 527 | # based on the width. |
---|
| 528 | # Find bin sizes |
---|
| 529 | #bin_size_low = math.fabs(data_x[1] - data_x[0]) |
---|
| 530 | #bin_size_high = math.fabs(data_x[length - 1] - data_x[length - 2]) |
---|
| 531 | # Let's set the bin size 1/3 of the width(sigma), it is good as long as |
---|
| 532 | # the scattering is monotonous. |
---|
| 533 | #if width_low < (bin_size_low): |
---|
| 534 | bin_size_low = width_low / 10.0 |
---|
| 535 | #if width_high < (bin_size_high): |
---|
| 536 | bin_size_high = width_high / 10.0 |
---|
| 537 | |
---|
| 538 | # Number of q points required below the 1st data point in order to extend |
---|
| 539 | # them 3 times of the width (std) |
---|
| 540 | if width_low > 0.0: |
---|
| 541 | nbins_low = math.ceil(3.0 * width_low / bin_size_low) |
---|
| 542 | # Number of q points required above the last data point |
---|
| 543 | if width_high > 0.0: |
---|
| 544 | nbins_high = math.ceil(3.0 * width_high / bin_size_high) |
---|
| 545 | # Make null q points |
---|
| 546 | extra_low = numpy.zeros(nbins_low) |
---|
| 547 | extra_high = numpy.zeros(nbins_high) |
---|
| 548 | # Give extrapolated values |
---|
| 549 | ind = 0 |
---|
| 550 | qvalue = data_x[0] - bin_size_low |
---|
| 551 | #if qvalue > 0: |
---|
| 552 | while(ind < nbins_low): |
---|
| 553 | extra_low[nbins_low - (ind + 1)] = qvalue |
---|
| 554 | qvalue -= bin_size_low |
---|
| 555 | ind += 1 |
---|
| 556 | #if qvalue <= 0: |
---|
| 557 | # break |
---|
| 558 | # Redefine nbins_low |
---|
| 559 | nbins_low = ind |
---|
| 560 | # Reset ind for another extrapolation |
---|
| 561 | ind = 0 |
---|
| 562 | qvalue = data_x[length -1] + bin_size_high |
---|
| 563 | while(ind < nbins_high): |
---|
| 564 | extra_high[ind] = qvalue |
---|
| 565 | qvalue += bin_size_high |
---|
| 566 | ind += 1 |
---|
| 567 | # Make a new qx array |
---|
| 568 | if nbins_low > 0: |
---|
| 569 | data_x_ext = numpy.append(extra_low, data_x) |
---|
| 570 | else: |
---|
| 571 | data_x_ext = data_x |
---|
| 572 | data_x_ext = numpy.append(data_x_ext, extra_high) |
---|
| 573 | |
---|
| 574 | # Redefine extra_low and high based on corrected nbins |
---|
| 575 | # And note that it is not necessary for extra_width to be a non-zero |
---|
| 576 | if nbins_low > 0: |
---|
| 577 | extra_low = numpy.zeros(nbins_low) |
---|
| 578 | extra_high = numpy.zeros(nbins_high) |
---|
| 579 | # Make new width array |
---|
| 580 | new_width = numpy.append(extra_low, width) |
---|
| 581 | new_width = numpy.append(new_width, extra_high) |
---|
| 582 | |
---|
| 583 | # nbins corrections due to the negative q value |
---|
| 584 | nbins_low = nbins_low - len(data_x_ext[data_x_ext <= 0]) |
---|
| 585 | return nbins_low, nbins_high, \ |
---|
| 586 | new_width[data_x_ext > 0], data_x_ext[data_x_ext > 0] |
---|
[9f7fbd9] | 587 | |
---|
| 588 | |
---|
| 589 | |
---|
[a3f125f0] | 590 | from .resolution import Slit1D, Pinhole1D |
---|
| 591 | class PySmear(object): |
---|
| 592 | """ |
---|
| 593 | Wrapper for pure python sasmodels resolution functions. |
---|
| 594 | """ |
---|
| 595 | def __init__(self, resolution, model): |
---|
[9f7fbd9] | 596 | self.model = model |
---|
[a3f125f0] | 597 | self.resolution = resolution |
---|
| 598 | self.offset = numpy.searchsorted(self.resolution.q_calc, self.resolution.q[0]) |
---|
[9f7fbd9] | 599 | |
---|
[a3f125f0] | 600 | def apply(self, iq_in, first_bin=0, last_bin=None): |
---|
| 601 | """ |
---|
| 602 | Apply the resolution function to the data. |
---|
[9f7fbd9] | 603 | |
---|
[a3f125f0] | 604 | Note that this is called with iq_in matching data.x, but with |
---|
| 605 | iq_in[first_bin:last_bin] set to theory values for these bins, |
---|
| 606 | and the remainder left undefined. The first_bin, last_bin values |
---|
| 607 | should be those returned from get_bin_range. |
---|
| 608 | |
---|
| 609 | The returned value is of the same length as iq_in, with the range |
---|
| 610 | first_bin:last_bin set to the resolution smeared values. |
---|
| 611 | """ |
---|
| 612 | if last_bin is None: last_bin = len(iq_in) |
---|
| 613 | start, end = first_bin + self.offset, last_bin + self.offset |
---|
[9f7fbd9] | 614 | q_calc = self.resolution.q_calc |
---|
| 615 | iq_calc = numpy.empty_like(q_calc) |
---|
[a3f125f0] | 616 | if start > 0: |
---|
| 617 | iq_calc[:start] = self.model.evalDistribution(q_calc[:start]) |
---|
| 618 | if end+1 < len(q_calc): |
---|
| 619 | iq_calc[end+1:] = self.model.evalDistribution(q_calc[end+1:]) |
---|
| 620 | iq_calc[start:end+1] = iq_in[first_bin:last_bin+1] |
---|
| 621 | smeared = self.resolution.apply(iq_calc) |
---|
| 622 | return smeared |
---|
| 623 | __call__ = apply |
---|
[9f7fbd9] | 624 | |
---|
| 625 | def get_bin_range(self, q_min=None, q_max=None): |
---|
| 626 | """ |
---|
[a3f125f0] | 627 | For a given q_min, q_max, find the corresponding indices in the data. |
---|
[9f7fbd9] | 628 | |
---|
[a3f125f0] | 629 | Returns first, last. |
---|
[9f7fbd9] | 630 | |
---|
[a3f125f0] | 631 | Note that these are indexes into q from the data, not the q_calc |
---|
| 632 | needed by the resolution function. Note also that these are the |
---|
| 633 | indices, not the range limits. That is, the complete range will be |
---|
| 634 | q[first:last+1]. |
---|
[9f7fbd9] | 635 | """ |
---|
[a3f125f0] | 636 | q = self.resolution.q |
---|
| 637 | first = numpy.searchsorted(q, q_min) |
---|
| 638 | last = numpy.searchsorted(q, q_max) |
---|
| 639 | return first, min(last,len(q)-1) |
---|
| 640 | |
---|
| 641 | def slit_smear(data, model=None): |
---|
| 642 | q = data.x |
---|
| 643 | width = data.dxw if data.dxw is not None else 0 |
---|
| 644 | height = data.dxl if data.dxl is not None else 0 |
---|
| 645 | # TODO: width and height seem to be reversed |
---|
| 646 | return PySmear(Slit1D(q, height, width), model) |
---|
| 647 | |
---|
| 648 | def pinhole_smear(data, model=None): |
---|
| 649 | q = data.x |
---|
| 650 | width = data.dx if data.dx is not None else 0 |
---|
| 651 | return PySmear(Pinhole1D(q, width), model) |
---|