| 1 | """ |
|---|
| 2 | Sasmodels core. |
|---|
| 3 | """ |
|---|
| 4 | import sys, os |
|---|
| 5 | import datetime |
|---|
| 6 | |
|---|
| 7 | from sasmodels import sesans |
|---|
| 8 | |
|---|
| 9 | # CRUFT python 2.6 |
|---|
| 10 | if not hasattr(datetime.timedelta, 'total_seconds'): |
|---|
| 11 | def delay(dt): return dt.days*86400 + dt.seconds + 1e-6*dt.microseconds |
|---|
| 12 | else: |
|---|
| 13 | def delay(dt): return dt.total_seconds() |
|---|
| 14 | |
|---|
| 15 | import numpy as np |
|---|
| 16 | |
|---|
| 17 | try: |
|---|
| 18 | from .kernelcl import load_model as _loader |
|---|
| 19 | except RuntimeError,exc: |
|---|
| 20 | import warnings |
|---|
| 21 | warnings.warn(str(exc)) |
|---|
| 22 | warnings.warn("OpenCL not available --- using ctypes instead") |
|---|
| 23 | from .kerneldll import load_model as _loader |
|---|
| 24 | |
|---|
| 25 | def load_model(modelname, dtype='single'): |
|---|
| 26 | """ |
|---|
| 27 | Load model by name. |
|---|
| 28 | """ |
|---|
| 29 | sasmodels = __import__('sasmodels.models.'+modelname) |
|---|
| 30 | module = getattr(sasmodels.models, modelname, None) |
|---|
| 31 | model = _loader(module, dtype=dtype) |
|---|
| 32 | return model |
|---|
| 33 | |
|---|
| 34 | |
|---|
| 35 | def tic(): |
|---|
| 36 | """ |
|---|
| 37 | Timer function. |
|---|
| 38 | |
|---|
| 39 | Use "toc=tic()" to start the clock and "toc()" to measure |
|---|
| 40 | a time interval. |
|---|
| 41 | """ |
|---|
| 42 | then = datetime.datetime.now() |
|---|
| 43 | return lambda: delay(datetime.datetime.now()-then) |
|---|
| 44 | |
|---|
| 45 | |
|---|
| 46 | def load_data(filename): |
|---|
| 47 | """ |
|---|
| 48 | Load data using a sasview loader. |
|---|
| 49 | """ |
|---|
| 50 | from sas.dataloader.loader import Loader |
|---|
| 51 | loader = Loader() |
|---|
| 52 | data = loader.load(filename) |
|---|
| 53 | if data is None: |
|---|
| 54 | raise IOError("Data %r could not be loaded"%filename) |
|---|
| 55 | return data |
|---|
| 56 | |
|---|
| 57 | |
|---|
| 58 | def empty_data1D(q): |
|---|
| 59 | """ |
|---|
| 60 | Create empty 1D data using the given *q* as the x value. |
|---|
| 61 | |
|---|
| 62 | Resolutions dq/q is 5%. |
|---|
| 63 | """ |
|---|
| 64 | |
|---|
| 65 | from sas.dataloader.data_info import Data1D |
|---|
| 66 | |
|---|
| 67 | Iq = 100*np.ones_like(q) |
|---|
| 68 | dIq = np.sqrt(Iq) |
|---|
| 69 | data = Data1D(q, Iq, dx=0.05*q, dy=dIq) |
|---|
| 70 | data.filename = "fake data" |
|---|
| 71 | data.qmin, data.qmax = q.min(), q.max() |
|---|
| 72 | return data |
|---|
| 73 | |
|---|
| 74 | |
|---|
| 75 | def empty_data2D(qx, qy=None): |
|---|
| 76 | """ |
|---|
| 77 | Create empty 2D data using the given mesh. |
|---|
| 78 | |
|---|
| 79 | If *qy* is missing, create a square mesh with *qy=qx*. |
|---|
| 80 | |
|---|
| 81 | Resolution dq/q is 5%. |
|---|
| 82 | """ |
|---|
| 83 | from sas.dataloader.data_info import Data2D, Detector |
|---|
| 84 | |
|---|
| 85 | if qy is None: |
|---|
| 86 | qy = qx |
|---|
| 87 | Qx,Qy = np.meshgrid(qx,qy) |
|---|
| 88 | Qx,Qy = Qx.flatten(), Qy.flatten() |
|---|
| 89 | Iq = 100*np.ones_like(Qx) |
|---|
| 90 | dIq = np.sqrt(Iq) |
|---|
| 91 | mask = np.ones(len(Iq), dtype='bool') |
|---|
| 92 | |
|---|
| 93 | data = Data2D() |
|---|
| 94 | data.filename = "fake data" |
|---|
| 95 | data.qx_data = Qx |
|---|
| 96 | data.qy_data = Qy |
|---|
| 97 | data.data = Iq |
|---|
| 98 | data.err_data = dIq |
|---|
| 99 | data.mask = mask |
|---|
| 100 | |
|---|
| 101 | # 5% dQ/Q resolution |
|---|
| 102 | data.dqx_data = 0.05*Qx |
|---|
| 103 | data.dqy_data = 0.05*Qy |
|---|
| 104 | |
|---|
| 105 | detector = Detector() |
|---|
| 106 | detector.pixel_size.x = 5 # mm |
|---|
| 107 | detector.pixel_size.y = 5 # mm |
|---|
| 108 | detector.distance = 4 # m |
|---|
| 109 | data.detector.append(detector) |
|---|
| 110 | data.xbins = qx |
|---|
| 111 | data.ybins = qy |
|---|
| 112 | data.source.wavelength = 5 # angstroms |
|---|
| 113 | data.source.wavelength_unit = "A" |
|---|
| 114 | data.Q_unit = "1/A" |
|---|
| 115 | data.I_unit = "1/cm" |
|---|
| 116 | data.q_data = np.sqrt(Qx**2 + Qy**2) |
|---|
| 117 | data.xaxis("Q_x", "A^{-1}") |
|---|
| 118 | data.yaxis("Q_y", "A^{-1}") |
|---|
| 119 | data.zaxis("Intensity", r"\text{cm}^{-1}") |
|---|
| 120 | return data |
|---|
| 121 | |
|---|
| 122 | |
|---|
| 123 | def set_beam_stop(data, radius, outer=None): |
|---|
| 124 | """ |
|---|
| 125 | Add a beam stop of the given *radius*. If *outer*, make an annulus. |
|---|
| 126 | """ |
|---|
| 127 | from sas.dataloader.manipulations import Ringcut |
|---|
| 128 | if hasattr(data, 'qx_data'): |
|---|
| 129 | data.mask = Ringcut(0, radius)(data) |
|---|
| 130 | if outer is not None: |
|---|
| 131 | data.mask += Ringcut(outer,np.inf)(data) |
|---|
| 132 | else: |
|---|
| 133 | data.mask = (data.x>=radius) |
|---|
| 134 | if outer is not None: |
|---|
| 135 | data.mask &= (data.x<outer) |
|---|
| 136 | |
|---|
| 137 | |
|---|
| 138 | def set_half(data, half): |
|---|
| 139 | """ |
|---|
| 140 | Select half of the data, either "right" or "left". |
|---|
| 141 | """ |
|---|
| 142 | from sas.dataloader.manipulations import Boxcut |
|---|
| 143 | if half == 'right': |
|---|
| 144 | data.mask += Boxcut(x_min=-np.inf, x_max=0.0, y_min=-np.inf, y_max=np.inf)(data) |
|---|
| 145 | if half == 'left': |
|---|
| 146 | data.mask += Boxcut(x_min=0.0, x_max=np.inf, y_min=-np.inf, y_max=np.inf)(data) |
|---|
| 147 | |
|---|
| 148 | |
|---|
| 149 | def set_top(data, max): |
|---|
| 150 | """ |
|---|
| 151 | Chop the top off the data, above *max*. |
|---|
| 152 | """ |
|---|
| 153 | from sas.dataloader.manipulations import Boxcut |
|---|
| 154 | data.mask += Boxcut(x_min=-np.inf, x_max=np.inf, y_min=-np.inf, y_max=max)(data) |
|---|
| 155 | |
|---|
| 156 | |
|---|
| 157 | def plot_data(data, iq, vmin=None, vmax=None, scale='log'): |
|---|
| 158 | """ |
|---|
| 159 | Plot the target value for the data. This could be the data itself, |
|---|
| 160 | the theory calculation, or the residuals. |
|---|
| 161 | |
|---|
| 162 | *scale* can be 'log' for log scale data, or 'linear'. |
|---|
| 163 | """ |
|---|
| 164 | from numpy.ma import masked_array, masked |
|---|
| 165 | import matplotlib.pyplot as plt |
|---|
| 166 | if hasattr(data, 'qx_data'): |
|---|
| 167 | iq = iq+0 |
|---|
| 168 | valid = np.isfinite(iq) |
|---|
| 169 | if scale == 'log': |
|---|
| 170 | valid[valid] = (iq[valid] > 0) |
|---|
| 171 | iq[valid] = np.log10(iq[valid]) |
|---|
| 172 | iq[~valid|data.mask] = 0 |
|---|
| 173 | #plottable = iq |
|---|
| 174 | plottable = masked_array(iq, ~valid|data.mask) |
|---|
| 175 | xmin, xmax = min(data.qx_data), max(data.qx_data) |
|---|
| 176 | ymin, ymax = min(data.qy_data), max(data.qy_data) |
|---|
| 177 | try: |
|---|
| 178 | if vmin is None: vmin = iq[valid&~data.mask].min() |
|---|
| 179 | if vmax is None: vmax = iq[valid&~data.mask].max() |
|---|
| 180 | except: |
|---|
| 181 | vmin, vmax = 0, 1 |
|---|
| 182 | plt.imshow(plottable.reshape(128,128), |
|---|
| 183 | interpolation='nearest', aspect=1, origin='upper', |
|---|
| 184 | extent=[xmin, xmax, ymin, ymax], vmin=vmin, vmax=vmax) |
|---|
| 185 | else: # 1D data |
|---|
| 186 | if scale == 'linear': |
|---|
| 187 | idx = np.isfinite(iq) |
|---|
| 188 | plt.plot(data.x[idx], iq[idx]) |
|---|
| 189 | else: |
|---|
| 190 | idx = np.isfinite(iq) |
|---|
| 191 | idx[idx] = (iq[idx]>0) |
|---|
| 192 | plt.loglog(data.x[idx], iq[idx]) |
|---|
| 193 | |
|---|
| 194 | |
|---|
| 195 | def _plot_result1D(data, theory, view): |
|---|
| 196 | """ |
|---|
| 197 | Plot the data and residuals for 1D data. |
|---|
| 198 | """ |
|---|
| 199 | import matplotlib.pyplot as plt |
|---|
| 200 | from numpy.ma import masked_array, masked |
|---|
| 201 | #print "not a number",sum(np.isnan(data.y)) |
|---|
| 202 | #data.y[data.y<0.05] = 0.5 |
|---|
| 203 | mdata = masked_array(data.y, data.mask) |
|---|
| 204 | mdata[np.isnan(mdata)] = masked |
|---|
| 205 | if view is 'log': |
|---|
| 206 | mdata[mdata <= 0] = masked |
|---|
| 207 | mtheory = masked_array(theory, mdata.mask) |
|---|
| 208 | mresid = masked_array((theory-data.y)/data.dy, mdata.mask) |
|---|
| 209 | |
|---|
| 210 | plt.subplot(121) |
|---|
| 211 | plt.errorbar(data.x, mdata, yerr=data.dy) |
|---|
| 212 | plt.plot(data.x, mtheory, '-', hold=True) |
|---|
| 213 | plt.yscale(view) |
|---|
| 214 | plt.subplot(122) |
|---|
| 215 | plt.plot(data.x, mresid, 'x') |
|---|
| 216 | #plt.axhline(1, color='black', ls='--',lw=1, hold=True) |
|---|
| 217 | #plt.axhline(0, color='black', lw=1, hold=True) |
|---|
| 218 | #plt.axhline(-1, color='black', ls='--',lw=1, hold=True) |
|---|
| 219 | |
|---|
| 220 | def _plot_sesans(data, theory, view): |
|---|
| 221 | import matplotlib.pyplot as plt |
|---|
| 222 | resid = (theory - data.y)/data.dy |
|---|
| 223 | plt.subplot(121) |
|---|
| 224 | plt.errorbar(data.x, data.y, yerr=data.dy) |
|---|
| 225 | plt.plot(data.x, theory, '-', hold=True) |
|---|
| 226 | plt.xlabel('spin echo length (A)') |
|---|
| 227 | plt.ylabel('polarization') |
|---|
| 228 | plt.subplot(122) |
|---|
| 229 | plt.plot(data.x, resid, 'x') |
|---|
| 230 | plt.xlabel('spin echo length (A)') |
|---|
| 231 | plt.ylabel('residuals') |
|---|
| 232 | |
|---|
| 233 | def _plot_result2D(data, theory, view): |
|---|
| 234 | """ |
|---|
| 235 | Plot the data and residuals for 2D data. |
|---|
| 236 | """ |
|---|
| 237 | import matplotlib.pyplot as plt |
|---|
| 238 | resid = (theory-data.data)/data.err_data |
|---|
| 239 | plt.subplot(131) |
|---|
| 240 | plot_data(data, data.data, scale=view) |
|---|
| 241 | plt.colorbar() |
|---|
| 242 | plt.subplot(132) |
|---|
| 243 | plot_data(data, theory, scale=view) |
|---|
| 244 | plt.colorbar() |
|---|
| 245 | plt.subplot(133) |
|---|
| 246 | plot_data(data, resid, scale='linear') |
|---|
| 247 | plt.colorbar() |
|---|
| 248 | |
|---|
| 249 | class BumpsModel(object): |
|---|
| 250 | """ |
|---|
| 251 | Return a bumps wrapper for a SAS model. |
|---|
| 252 | |
|---|
| 253 | *data* is the data to be fitted. |
|---|
| 254 | |
|---|
| 255 | *model* is the SAS model, e.g., from :func:`gen.opencl_model`. |
|---|
| 256 | |
|---|
| 257 | *cutoff* is the integration cutoff, which avoids computing the |
|---|
| 258 | the SAS model where the polydispersity weight is low. |
|---|
| 259 | |
|---|
| 260 | Model parameters can be initialized with additional keyword |
|---|
| 261 | arguments, or by assigning to model.parameter_name.value. |
|---|
| 262 | |
|---|
| 263 | The resulting bumps model can be used directly in a FitProblem call. |
|---|
| 264 | """ |
|---|
| 265 | def __init__(self, data, model, cutoff=1e-5, **kw): |
|---|
| 266 | from bumps.names import Parameter |
|---|
| 267 | |
|---|
| 268 | # remember inputs so we can inspect from outside |
|---|
| 269 | self.data = data |
|---|
| 270 | self.model = model |
|---|
| 271 | self.cutoff = cutoff |
|---|
| 272 | # TODO if isinstance(data,SESANSData1D) |
|---|
| 273 | if hasattr(data, 'lam'): |
|---|
| 274 | self.data_type = 'sesans' |
|---|
| 275 | elif hasattr(data, 'qx_data'): |
|---|
| 276 | self.data_type = 'Iqxy' |
|---|
| 277 | else: |
|---|
| 278 | self.data_type = 'Iq' |
|---|
| 279 | |
|---|
| 280 | partype = model.info['partype'] |
|---|
| 281 | |
|---|
| 282 | # interpret data |
|---|
| 283 | if self.data_type == 'sesans': |
|---|
| 284 | q = sesans.make_q(data.sample.zacceptance, data.Rmax) |
|---|
| 285 | self.index = slice(None,None) |
|---|
| 286 | self.iq = data.y |
|---|
| 287 | self.diq = data.dy |
|---|
| 288 | self._theory = np.zeros_like(q) |
|---|
| 289 | q_vectors = [q] |
|---|
| 290 | elif self.data_type == 'Iqxy': |
|---|
| 291 | self.index = (data.mask==0) & (~np.isnan(data.data)) |
|---|
| 292 | self.iq = data.data[self.index] |
|---|
| 293 | self.diq = data.err_data[self.index] |
|---|
| 294 | self._theory = np.zeros_like(data.data) |
|---|
| 295 | if not partype['orientation'] and not partype['magnetic']: |
|---|
| 296 | q_vectors = [np.sqrt(data.qx_data**2+data.qy_data**2)] |
|---|
| 297 | else: |
|---|
| 298 | q_vectors = [data.qx_data, data.qy_data] |
|---|
| 299 | elif self.data_type == 'Iq': |
|---|
| 300 | self.index = (data.x>=data.qmin) & (data.x<=data.qmax) & ~np.isnan(data.y) |
|---|
| 301 | self.iq = data.y[self.index] |
|---|
| 302 | self.diq = data.dy[self.index] |
|---|
| 303 | self._theory = np.zeros_like(data.y) |
|---|
| 304 | q_vectors = [data.x] |
|---|
| 305 | else: |
|---|
| 306 | raise ValueError("Unknown data type") # never gets here |
|---|
| 307 | |
|---|
| 308 | # Remember function inputs so we can delay loading the function and |
|---|
| 309 | # so we can save/restore state |
|---|
| 310 | self._fn_inputs = [v[self.index] for v in q_vectors] |
|---|
| 311 | self._fn = None |
|---|
| 312 | |
|---|
| 313 | # define bumps parameters |
|---|
| 314 | pars = [] |
|---|
| 315 | for p in model.info['parameters']: |
|---|
| 316 | name, default, limits, ptype = p[0], p[2], p[3], p[4] |
|---|
| 317 | value = kw.pop(name, default) |
|---|
| 318 | setattr(self, name, Parameter.default(value, name=name, limits=limits)) |
|---|
| 319 | pars.append(name) |
|---|
| 320 | for name in partype['pd-2d']: |
|---|
| 321 | for xpart,xdefault,xlimits in [ |
|---|
| 322 | ('_pd', 0, limits), |
|---|
| 323 | ('_pd_n', 35, (0,1000)), |
|---|
| 324 | ('_pd_nsigma', 3, (0, 10)), |
|---|
| 325 | ('_pd_type', 'gaussian', None), |
|---|
| 326 | ]: |
|---|
| 327 | xname = name+xpart |
|---|
| 328 | xvalue = kw.pop(xname, xdefault) |
|---|
| 329 | if xlimits is not None: |
|---|
| 330 | xvalue = Parameter.default(xvalue, name=xname, limits=xlimits) |
|---|
| 331 | pars.append(xname) |
|---|
| 332 | setattr(self, xname, xvalue) |
|---|
| 333 | self._parameter_names = pars |
|---|
| 334 | if kw: |
|---|
| 335 | raise TypeError("unexpected parameters: %s"%(", ".join(sorted(kw.keys())))) |
|---|
| 336 | self.update() |
|---|
| 337 | |
|---|
| 338 | def update(self): |
|---|
| 339 | self._cache = {} |
|---|
| 340 | |
|---|
| 341 | def numpoints(self): |
|---|
| 342 | return len(self.iq) |
|---|
| 343 | |
|---|
| 344 | def parameters(self): |
|---|
| 345 | return dict((k,getattr(self,k)) for k in self._parameter_names) |
|---|
| 346 | |
|---|
| 347 | def theory(self): |
|---|
| 348 | if 'theory' not in self._cache: |
|---|
| 349 | if self._fn is None: |
|---|
| 350 | input = self.model.make_input(self._fn_inputs) |
|---|
| 351 | self._fn = self.model(input) |
|---|
| 352 | |
|---|
| 353 | fixed_pars = [getattr(self,p).value for p in self._fn.fixed_pars] |
|---|
| 354 | pd_pars = [self._get_weights(p) for p in self._fn.pd_pars] |
|---|
| 355 | #print fixed_pars,pd_pars |
|---|
| 356 | self._theory[self.index] = self._fn(fixed_pars, pd_pars, self.cutoff) |
|---|
| 357 | #self._theory[:] = self._fn.eval(pars, pd_pars) |
|---|
| 358 | if self.data_type == 'sesans': |
|---|
| 359 | P = sesans.hankel(self.data.x, self.data.lam*1e-9, |
|---|
| 360 | self.data.sample.thickness/10, self._fn_inputs[0], |
|---|
| 361 | self._theory) |
|---|
| 362 | self._cache['theory'] = P |
|---|
| 363 | else: |
|---|
| 364 | self._cache['theory'] = self._theory |
|---|
| 365 | return self._cache['theory'] |
|---|
| 366 | |
|---|
| 367 | def residuals(self): |
|---|
| 368 | #if np.any(self.err ==0): print "zeros in err" |
|---|
| 369 | return (self.theory()[self.index]-self.iq)/self.diq |
|---|
| 370 | |
|---|
| 371 | def nllf(self): |
|---|
| 372 | R = self.residuals() |
|---|
| 373 | #if np.any(np.isnan(R)): print "NaN in residuals" |
|---|
| 374 | return 0.5*np.sum(R**2) |
|---|
| 375 | |
|---|
| 376 | def __call__(self): |
|---|
| 377 | return 2*self.nllf()/self.dof |
|---|
| 378 | |
|---|
| 379 | def plot(self, view='log'): |
|---|
| 380 | """ |
|---|
| 381 | Plot the data and residuals. |
|---|
| 382 | """ |
|---|
| 383 | data, theory = self.data, self.theory() |
|---|
| 384 | if self.data_type == 'Iq': |
|---|
| 385 | _plot_result1D(data, theory, view) |
|---|
| 386 | elif self.data_type == 'Iqxy': |
|---|
| 387 | _plot_result2D(data, theory, view) |
|---|
| 388 | elif self.data_type == 'sesans': |
|---|
| 389 | _plot_sesans(data, theory, view) |
|---|
| 390 | else: |
|---|
| 391 | raise ValueError("Unknown data type") |
|---|
| 392 | |
|---|
| 393 | def simulate_data(self, noise=None): |
|---|
| 394 | print "noise", noise |
|---|
| 395 | if noise is None: |
|---|
| 396 | noise = self.diq[self.index] |
|---|
| 397 | else: |
|---|
| 398 | noise = 0.01*noise |
|---|
| 399 | self.diq[self.index] = noise |
|---|
| 400 | y = self.theory() |
|---|
| 401 | y += y*np.random.randn(*y.shape)*noise |
|---|
| 402 | if self.data_type == 'Iq': |
|---|
| 403 | self.data.y[self.index] = y |
|---|
| 404 | elif self.data_type == 'Iqxy': |
|---|
| 405 | self.data.data[self.index] = y |
|---|
| 406 | elif self.data_type == 'sesans': |
|---|
| 407 | self.data.y[self.index] = y |
|---|
| 408 | else: |
|---|
| 409 | raise ValueError("Unknown model") |
|---|
| 410 | |
|---|
| 411 | def save(self, basename): |
|---|
| 412 | pass |
|---|
| 413 | |
|---|
| 414 | def _get_weights(self, par): |
|---|
| 415 | from . import weights |
|---|
| 416 | |
|---|
| 417 | relative = self.model.info['partype']['pd-rel'] |
|---|
| 418 | limits = self.model.info['limits'] |
|---|
| 419 | disperser,value,npts,width,nsigma = [getattr(self, par+ext) |
|---|
| 420 | for ext in ('_pd_type','','_pd_n','_pd','_pd_nsigma')] |
|---|
| 421 | v,w = weights.get_weights( |
|---|
| 422 | disperser, int(npts.value), width.value, nsigma.value, |
|---|
| 423 | value.value, limits[par], par in relative) |
|---|
| 424 | return v,w/w.max() |
|---|
| 425 | |
|---|
| 426 | def __getstate__(self): |
|---|
| 427 | # Can't pickle gpu functions, so instead make them lazy |
|---|
| 428 | state = self.__dict__.copy() |
|---|
| 429 | state['_fn'] = None |
|---|
| 430 | return state |
|---|
| 431 | |
|---|
| 432 | def __setstate__(self, state): |
|---|
| 433 | self.__dict__ = state |
|---|
| 434 | |
|---|
| 435 | |
|---|
| 436 | def demo(): |
|---|
| 437 | data = load_data('DEC07086.DAT') |
|---|
| 438 | set_beam_stop(data, 0.004) |
|---|
| 439 | plot_data(data) |
|---|
| 440 | import matplotlib.pyplot as plt; plt.show() |
|---|
| 441 | |
|---|
| 442 | |
|---|
| 443 | if __name__ == "__main__": |
|---|
| 444 | demo() |
|---|
