| 1 | #!/usr/bin/env python |
|---|
| 2 | # -*- coding: utf-8 -*- |
|---|
| 3 | |
|---|
| 4 | import sys, os |
|---|
| 5 | import datetime |
|---|
| 6 | import warnings |
|---|
| 7 | |
|---|
| 8 | import numpy as np |
|---|
| 9 | |
|---|
| 10 | |
|---|
| 11 | def opencl_model(kernel_module, dtype="single"): |
|---|
| 12 | from sasmodels import gen, gpu |
|---|
| 13 | |
|---|
| 14 | source, info = gen.make(kernel_module) |
|---|
| 15 | ## for debugging, save source to a .cl file, edit it, and reload as model |
|---|
| 16 | #open(modelname+'.cl','w').write(source) |
|---|
| 17 | #source = open(modelname+'.cl','r').read() |
|---|
| 18 | return gpu.GpuModel(source, info, dtype) |
|---|
| 19 | |
|---|
| 20 | |
|---|
| 21 | if sys.platform == 'darwin': |
|---|
| 22 | COMPILE = "gcc-mp-4.7 -shared -fPIC -std=c99 -fopenmp -O2 -Wall %s -o %s -lm -lgomp" |
|---|
| 23 | elif os.name == 'nt': |
|---|
| 24 | COMPILE = "gcc -shared -fPIC -std=c99 -fopenmp -O2 -Wall %s -o %s -lm" |
|---|
| 25 | else: |
|---|
| 26 | COMPILE = "cc -shared -fPIC -std=c99 -fopenmp -O2 -Wall %s -o %s -lm" |
|---|
| 27 | DLL_PATH = "/tmp" |
|---|
| 28 | |
|---|
| 29 | |
|---|
| 30 | def dll_path(info): |
|---|
| 31 | from os.path import join as joinpath, split as splitpath, splitext |
|---|
| 32 | basename = splitext(splitpath(info['filename'])[1])[0] |
|---|
| 33 | return joinpath(DLL_PATH, basename+'.so') |
|---|
| 34 | |
|---|
| 35 | |
|---|
| 36 | def dll_model(kernel_module): |
|---|
| 37 | import os |
|---|
| 38 | import tempfile |
|---|
| 39 | from sasmodels import gen, dll |
|---|
| 40 | |
|---|
| 41 | source, info = gen.make(kernel_module) |
|---|
| 42 | dllpath = dll_path(info) |
|---|
| 43 | if not os.path.exists(dllpath) \ |
|---|
| 44 | or (os.path.getmtime(dllpath) < os.path.getmtime(info['filename'])): |
|---|
| 45 | # Replace with a proper temp file |
|---|
| 46 | srcfile = tempfile.mkstemp(suffix=".c",prefix="sas_"+info['name']) |
|---|
| 47 | open(srcfile, 'w').write(source) |
|---|
| 48 | os.system(COMPILE%(srcfile, dllpath)) |
|---|
| 49 | ## comment the following to keep the generated c file |
|---|
| 50 | #os.unlink(srcfile) |
|---|
| 51 | return dll.DllModel(dllpath, info) |
|---|
| 52 | |
|---|
| 53 | |
|---|
| 54 | TIC = None |
|---|
| 55 | def tic(): |
|---|
| 56 | global TIC |
|---|
| 57 | then = datetime.datetime.now() |
|---|
| 58 | TIC = lambda: (datetime.datetime.now()-then).total_seconds() |
|---|
| 59 | return TIC |
|---|
| 60 | |
|---|
| 61 | |
|---|
| 62 | def toc(): |
|---|
| 63 | return TIC() |
|---|
| 64 | |
|---|
| 65 | |
|---|
| 66 | def load_data(filename): |
|---|
| 67 | from sans.dataloader.loader import Loader |
|---|
| 68 | loader = Loader() |
|---|
| 69 | data = loader.load(filename) |
|---|
| 70 | if data is None: |
|---|
| 71 | raise IOError("Data %r could not be loaded"%filename) |
|---|
| 72 | return data |
|---|
| 73 | |
|---|
| 74 | |
|---|
| 75 | def empty_data2D(qx, qy=None): |
|---|
| 76 | from sans.dataloader.data_info import Data2D, Detector |
|---|
| 77 | |
|---|
| 78 | if qy is None: |
|---|
| 79 | qy = qx |
|---|
| 80 | Qx,Qy = np.meshgrid(qx,qy) |
|---|
| 81 | Qx,Qy = Qx.flatten(), Qy.flatten() |
|---|
| 82 | Iq = 100*np.ones_like(Qx) |
|---|
| 83 | dIq = np.sqrt(Iq) |
|---|
| 84 | mask = np.ones(len(Iq), dtype='bool') |
|---|
| 85 | |
|---|
| 86 | data = Data2D() |
|---|
| 87 | data.filename = "fake data" |
|---|
| 88 | data.qx_data = Qx |
|---|
| 89 | data.qy_data = Qy |
|---|
| 90 | data.data = Iq |
|---|
| 91 | data.err_data = dIq |
|---|
| 92 | data.mask = mask |
|---|
| 93 | |
|---|
| 94 | # 5% dQ/Q resolution |
|---|
| 95 | data.dqx_data = 0.05*Qx |
|---|
| 96 | data.dqy_data = 0.05*Qy |
|---|
| 97 | |
|---|
| 98 | detector = Detector() |
|---|
| 99 | detector.pixel_size.x = 5 # mm |
|---|
| 100 | detector.pixel_size.y = 5 # mm |
|---|
| 101 | detector.distance = 4 # m |
|---|
| 102 | data.detector.append(detector) |
|---|
| 103 | data.xbins = qx |
|---|
| 104 | data.ybins = qy |
|---|
| 105 | data.source.wavelength = 5 # angstroms |
|---|
| 106 | data.source.wavelength_unit = "A" |
|---|
| 107 | data.Q_unit = "1/A" |
|---|
| 108 | data.I_unit = "1/cm" |
|---|
| 109 | data.q_data = np.sqrt(Qx**2 + Qy**2) |
|---|
| 110 | data.xaxis("Q_x", "A^{-1}") |
|---|
| 111 | data.yaxis("Q_y", "A^{-1}") |
|---|
| 112 | data.zaxis("Intensity", r"\text{cm}^{-1}") |
|---|
| 113 | return data |
|---|
| 114 | |
|---|
| 115 | |
|---|
| 116 | def empty_data1D(q): |
|---|
| 117 | from sans.dataloader.data_info import Data1D |
|---|
| 118 | |
|---|
| 119 | Iq = 100*np.ones_like(q) |
|---|
| 120 | dIq = np.sqrt(Iq) |
|---|
| 121 | data = Data1D(q, Iq, dx=0.05*q, dy=dIq) |
|---|
| 122 | data.filename = "fake data" |
|---|
| 123 | data.qmin, data.qmax = q.min(), q.max() |
|---|
| 124 | return data |
|---|
| 125 | |
|---|
| 126 | |
|---|
| 127 | def set_beam_stop(data, radius, outer=None): |
|---|
| 128 | from sans.dataloader.manipulations import Ringcut |
|---|
| 129 | if hasattr(data, 'qx_data'): |
|---|
| 130 | data.mask = Ringcut(0, radius)(data) |
|---|
| 131 | if outer is not None: |
|---|
| 132 | data.mask += Ringcut(outer,np.inf)(data) |
|---|
| 133 | else: |
|---|
| 134 | data.mask = (data.x>=radius) |
|---|
| 135 | if outer is not None: |
|---|
| 136 | data.mask &= (data.x<outer) |
|---|
| 137 | |
|---|
| 138 | |
|---|
| 139 | def set_half(data, half): |
|---|
| 140 | from sans.dataloader.manipulations import Boxcut |
|---|
| 141 | if half == 'right': |
|---|
| 142 | data.mask += Boxcut(x_min=-np.inf, x_max=0.0, y_min=-np.inf, y_max=np.inf)(data) |
|---|
| 143 | if half == 'left': |
|---|
| 144 | data.mask += Boxcut(x_min=0.0, x_max=np.inf, y_min=-np.inf, y_max=np.inf)(data) |
|---|
| 145 | |
|---|
| 146 | |
|---|
| 147 | def set_top(data, max): |
|---|
| 148 | from sans.dataloader.manipulations import Boxcut |
|---|
| 149 | data.mask += Boxcut(x_min=-np.inf, x_max=np.inf, y_min=-np.inf, y_max=max)(data) |
|---|
| 150 | |
|---|
| 151 | |
|---|
| 152 | def plot_data(data, iq, vmin=None, vmax=None, scale='log'): |
|---|
| 153 | from numpy.ma import masked_array, masked |
|---|
| 154 | import matplotlib.pyplot as plt |
|---|
| 155 | if hasattr(data, 'qx_data'): |
|---|
| 156 | img = masked_array(iq, data.mask) |
|---|
| 157 | if scale == 'log': |
|---|
| 158 | img[(img <= 0) | ~np.isfinite(img)] = masked |
|---|
| 159 | img = np.log10(img) |
|---|
| 160 | xmin, xmax = min(data.qx_data), max(data.qx_data) |
|---|
| 161 | ymin, ymax = min(data.qy_data), max(data.qy_data) |
|---|
| 162 | plt.imshow(img.reshape(128,128), |
|---|
| 163 | interpolation='nearest', aspect=1, origin='upper', |
|---|
| 164 | extent=[xmin, xmax, ymin, ymax], vmin=vmin, vmax=vmax) |
|---|
| 165 | else: # 1D data |
|---|
| 166 | if scale == 'linear': |
|---|
| 167 | idx = np.isfinite(iq) |
|---|
| 168 | plt.plot(data.x[idx], iq[idx]) |
|---|
| 169 | else: |
|---|
| 170 | idx = np.isfinite(iq) & (iq>0) |
|---|
| 171 | plt.loglog(data.x[idx], iq[idx]) |
|---|
| 172 | |
|---|
| 173 | |
|---|
| 174 | def plot_result2D(data, theory, view='log'): |
|---|
| 175 | import matplotlib.pyplot as plt |
|---|
| 176 | resid = (theory-data.data)/data.err_data |
|---|
| 177 | plt.subplot(131) |
|---|
| 178 | plot_data(data, data.data, scale=view) |
|---|
| 179 | plt.colorbar() |
|---|
| 180 | plt.subplot(132) |
|---|
| 181 | plot_data(data, theory, scale=view) |
|---|
| 182 | plt.colorbar() |
|---|
| 183 | plt.subplot(133) |
|---|
| 184 | plot_data(data, resid, scale='linear') |
|---|
| 185 | plt.colorbar() |
|---|
| 186 | |
|---|
| 187 | |
|---|
| 188 | def plot_result1D(data, theory, view='log'): |
|---|
| 189 | import matplotlib.pyplot as plt |
|---|
| 190 | from numpy.ma import masked_array, masked |
|---|
| 191 | #print "not a number",sum(np.isnan(data.y)) |
|---|
| 192 | #data.y[data.y<0.05] = 0.5 |
|---|
| 193 | mdata = masked_array(data.y, data.mask) |
|---|
| 194 | mdata[np.isnan(mdata)] = masked |
|---|
| 195 | if view is 'log': |
|---|
| 196 | mdata[mdata <= 0] = masked |
|---|
| 197 | mtheory = masked_array(theory, mdata.mask) |
|---|
| 198 | mresid = masked_array((theory-data.y)/data.dy, mdata.mask) |
|---|
| 199 | |
|---|
| 200 | plt.subplot(121) |
|---|
| 201 | plt.errorbar(data.x, mdata, yerr=data.dy) |
|---|
| 202 | plt.plot(data.x, mtheory, '-', hold=True) |
|---|
| 203 | plt.yscale(view) |
|---|
| 204 | plt.subplot(122) |
|---|
| 205 | plt.plot(data.x, mresid, 'x') |
|---|
| 206 | #plt.axhline(1, color='black', ls='--',lw=1, hold=True) |
|---|
| 207 | #plt.axhline(0, color='black', lw=1, hold=True) |
|---|
| 208 | #plt.axhline(-1, color='black', ls='--',lw=1, hold=True) |
|---|
| 209 | |
|---|
| 210 | |
|---|
| 211 | class BumpsModel(object): |
|---|
| 212 | def __init__(self, data, model, cutoff=1e-5, **kw): |
|---|
| 213 | from bumps.names import Parameter |
|---|
| 214 | from . import gpu |
|---|
| 215 | |
|---|
| 216 | # interpret data |
|---|
| 217 | self.is2D = hasattr(data,'qx_data') |
|---|
| 218 | self.data = data |
|---|
| 219 | if self.is2D: |
|---|
| 220 | self.index = (data.mask==0) & (~np.isnan(data.data)) |
|---|
| 221 | self.iq = data.data[self.index] |
|---|
| 222 | self.diq = data.err_data[self.index] |
|---|
| 223 | self._theory = np.zeros_like(data.data) |
|---|
| 224 | q_vectors = [data.qx_data, data.qy_data] |
|---|
| 225 | else: |
|---|
| 226 | self.index = (data.x>=data.qmin) & (data.x<=data.qmax) & ~np.isnan(data.y) |
|---|
| 227 | self.iq = data.y[self.index] |
|---|
| 228 | self.diq = data.dy[self.index] |
|---|
| 229 | self._theory = np.zeros_like(data.y) |
|---|
| 230 | q_vectors = [data.x] |
|---|
| 231 | #input = model.make_input(q_vectors) |
|---|
| 232 | input = model.make_input([v[self.index] for v in q_vectors]) |
|---|
| 233 | |
|---|
| 234 | # create model |
|---|
| 235 | self.fn = model(input) |
|---|
| 236 | self.cutoff = cutoff |
|---|
| 237 | |
|---|
| 238 | # define bumps parameters |
|---|
| 239 | pars = [] |
|---|
| 240 | extras = [] |
|---|
| 241 | for p in model.info['parameters']: |
|---|
| 242 | name, default, limits, ptype = p[0], p[2], p[3], p[4] |
|---|
| 243 | value = kw.pop(name, default) |
|---|
| 244 | setattr(self, name, Parameter.default(value, name=name, limits=limits)) |
|---|
| 245 | pars.append(name) |
|---|
| 246 | for name in model.info['partype']['pd-2d']: |
|---|
| 247 | for xpart,xdefault,xlimits in [ |
|---|
| 248 | ('_pd', 0, limits), |
|---|
| 249 | ('_pd_n', 35, (0,1000)), |
|---|
| 250 | ('_pd_nsigma', 3, (0, 10)), |
|---|
| 251 | ('_pd_type', 'gaussian', None), |
|---|
| 252 | ]: |
|---|
| 253 | xname = name+xpart |
|---|
| 254 | xvalue = kw.pop(xname, xdefault) |
|---|
| 255 | if xlimits is not None: |
|---|
| 256 | xvalue = Parameter.default(xvalue, name=xname, limits=xlimits) |
|---|
| 257 | pars.append(xname) |
|---|
| 258 | setattr(self, xname, xvalue) |
|---|
| 259 | self._parameter_names = pars |
|---|
| 260 | if kw: |
|---|
| 261 | raise TypeError("unexpected parameters: %s"%(", ".join(sorted(kw.keys())))) |
|---|
| 262 | self.update() |
|---|
| 263 | |
|---|
| 264 | def update(self): |
|---|
| 265 | self._cache = {} |
|---|
| 266 | |
|---|
| 267 | def numpoints(self): |
|---|
| 268 | return len(self.iq) |
|---|
| 269 | |
|---|
| 270 | def parameters(self): |
|---|
| 271 | return dict((k,getattr(self,k)) for k in self._parameter_names) |
|---|
| 272 | |
|---|
| 273 | def theory(self): |
|---|
| 274 | if 'theory' not in self._cache: |
|---|
| 275 | pars = [getattr(self,p).value for p in self.fn.fixed_pars] |
|---|
| 276 | pd_pars = [self._get_weights(p) for p in self.fn.pd_pars] |
|---|
| 277 | #print pars |
|---|
| 278 | self._theory[self.index] = self.fn(pars, pd_pars, self.cutoff) |
|---|
| 279 | #self._theory[:] = self.fn.eval(pars, pd_pars) |
|---|
| 280 | self._cache['theory'] = self._theory |
|---|
| 281 | return self._cache['theory'] |
|---|
| 282 | |
|---|
| 283 | def residuals(self): |
|---|
| 284 | #if np.any(self.err ==0): print "zeros in err" |
|---|
| 285 | return (self.theory()[self.index]-self.iq)/self.diq |
|---|
| 286 | |
|---|
| 287 | def nllf(self): |
|---|
| 288 | R = self.residuals() |
|---|
| 289 | #if np.any(np.isnan(R)): print "NaN in residuals" |
|---|
| 290 | return 0.5*np.sum(R**2) |
|---|
| 291 | |
|---|
| 292 | def __call__(self): |
|---|
| 293 | return 2*self.nllf()/self.dof |
|---|
| 294 | |
|---|
| 295 | def plot(self, view='log'): |
|---|
| 296 | if self.is2D: |
|---|
| 297 | plot_result2D(self.data, self.theory(), view=view) |
|---|
| 298 | else: |
|---|
| 299 | plot_result1D(self.data, self.theory(), view=view) |
|---|
| 300 | |
|---|
| 301 | def save(self, basename): |
|---|
| 302 | pass |
|---|
| 303 | |
|---|
| 304 | def _get_weights(self, par): |
|---|
| 305 | from . import weights |
|---|
| 306 | |
|---|
| 307 | relative = self.fn.info['partype']['pd-rel'] |
|---|
| 308 | limits = self.fn.info['limits'] |
|---|
| 309 | disperser,value,npts,width,nsigma = [getattr(self, par+ext) |
|---|
| 310 | for ext in ('_pd_type','','_pd_n','_pd','_pd_nsigma')] |
|---|
| 311 | v,w = weights.get_weights( |
|---|
| 312 | disperser, int(npts.value), width.value, nsigma.value, |
|---|
| 313 | value.value, limits[par], par in relative) |
|---|
| 314 | return v,w/w.max() |
|---|
| 315 | |
|---|
| 316 | |
|---|
| 317 | def demo(): |
|---|
| 318 | data = load_data('DEC07086.DAT') |
|---|
| 319 | set_beam_stop(data, 0.004) |
|---|
| 320 | plot_data(data) |
|---|
| 321 | import matplotlib.pyplot as plt; plt.show() |
|---|
| 322 | |
|---|
| 323 | |
|---|
| 324 | if __name__ == "__main__": |
|---|
| 325 | demo() |
|---|
