- Timestamp:
- Jan 14, 2010 2:31:24 PM (15 years ago)
- Branches:
- master, ESS_GUI, ESS_GUI_Docs, ESS_GUI_batch_fitting, ESS_GUI_bumps_abstraction, ESS_GUI_iss1116, ESS_GUI_iss879, ESS_GUI_iss959, ESS_GUI_opencl, ESS_GUI_ordering, ESS_GUI_sync_sascalc, costrafo411, magnetic_scatt, release-4.1.1, release-4.1.2, release-4.2.2, release_4.0.1, ticket-1009, ticket-1094-headless, ticket-1242-2d-resolution, ticket-1243, ticket-1249, ticket885, unittest-saveload
- Children:
- 82703a1
- Parents:
- 3082632
- File:
-
- 1 edited
Legend:
- Unmodified
- Added
- Removed
-
Invariant/invariant.py
r3082632 r59a41066 20 20 INTEGRATION_NSTEPS = 1000 21 21 22 def guinier(x, scale=1, radius=60):22 class Transform(object): 23 23 """ 24 Compute a F(x) = scale* e-((radius*x)**2/3). 25 @param x: a vector of q values 26 @param scale: the scale value 27 @param radius: the guinier radius value 28 @return F(x) 29 """ 30 if radius <= 0: 31 raise ValueError("Rg expected positive value, but got %s"%radius) 32 value = numpy.array([math.exp(-((radius * i)**2/3)) for i in x ]) 33 scale = numpy.sqrt(scale*scale) 34 if scale == 0: 35 raise ValueError("scale expected positive value, but got %s"%scale) 36 return scale * value 37 38 def power_law(x, scale=1, power=4): 24 Define interface that need to compute a function or an inverse 25 function given some x, y 39 26 """ 40 F(x) = scale* (x)^(-power) 41 when power= 4. the model is porod 42 else power_law 43 The model has three parameters: 44 @param x: a vector of q values 45 @param power: power of the function 46 @param scale : scale factor value 47 @param F(x) 48 """ 49 if power <=0: 50 raise ValueError("Power_law function expected positive power, but got %s"%power) 51 52 value = numpy.array([ math.pow(i, -power) for i in x ]) 53 scale = numpy.sqrt(scale*scale) 54 if scale == 0: 55 raise ValueError("scale expected positive value, but got %s"%scale) 56 return scale * value 57 58 class FitFunctor: 27 def set_value(self, a, b): 28 """ 29 set private member 30 """ 31 def transform_value(self, value): 32 """ 33 Can use one of the function transform_x2 or transform_logx 34 """ 35 def transform_x2(self, value): 36 """ 37 @param value: float 38 """ 39 return value **2 40 41 def transform_to_logx(self, value): 42 """ 43 @param value: float 44 """ 45 return math.log(value) 46 47 def inverse_transform_value(self, value): 48 """ 49 reverse transform vaalue given a specific function 50 return f-1(value) 51 @param value : float 52 """ 53 def get_extrapolated_data(self, data, q_start=Q_MINIMUM, q_end=Q_MAXIMUM): 54 """ 55 @return extrapolate data create from data 56 """ 57 def transform_data(self, x): 58 """ 59 @param x: vector of float 60 @param y : vector of float 61 return x, y transform given a specific function 62 """ 63 def inverse_transform_data(self, x, y): 64 """ 65 reverse transform x, y given a specific function 66 """ 67 def transform_error(self, y , dy=None): 68 if dy is None: 69 dy = numpy.array([math.sqrt(j) for j in y]) 70 return dy / y 71 72 def get_extrapolated_data(self, data, npts=INTEGRATION_NSTEPS, 73 q_start=Q_MINIMUM, q_end=Q_MAXIMUM, smear_indice=0): 74 """ 75 @return extrapolate data create from data 76 """ 77 #create new Data1D to compute the invariant 78 new_x = numpy.linspace(start=q_start, 79 stop=q_end, 80 num=npts, 81 endpoint=True) 82 new_y = self.transform_data(x=new_x) 83 84 dxl = None 85 dxw = None 86 if data.dxl is not None : 87 dxl = numpy.ones(INTEGRATION_NSTEPS) 88 dxl = dxl * data.dxl[smear_indice] 89 90 if data.dxw is not None : 91 dxw = numpy.ones(INTEGRATION_NSTEPS) 92 dxw = dxw * data.dxw[smear_indice] 93 94 result_data = LoaderData1D(x=new_x, y=new_y) 95 data.clone_without_data(clone=result_data) 96 result_data.dxl = dxl 97 result_data.dxw = dxw 98 99 100 return result_data 101 102 class Guinier(Transform): 103 """ 104 class of type Transform that performs operations related to guinier 105 function 106 """ 107 def __init__(self, scale=1, radius=60): 108 Transform.__init__(self) 109 self.scale = scale 110 self.radius = radius 111 112 def transform_value(self, value): 113 return self.transform_x2(value) 114 115 def set_value(self, a, b): 116 # b is the radius value of the guinier function 117 if b >=0 : 118 raise ValueError("Guinier fit was not converged") 119 else: 120 b = math.sqrt(-3 * b) 121 a = math.exp(a) 122 123 self.scale = a 124 self.radius = b 125 126 def transform_data(self, x): 127 """ 128 given a scale and a radius transform x, y using a guinier 129 function 130 """ 131 return self._guinier(x) 132 133 def inverse_transform_data(self, x, y): 134 """ 135 given a scale and a radius transform x, y using a inverse guinier 136 function 137 """ 138 result_x = numpy.array([i * i for i in x]) 139 result_y = numpy.array([math.log(j) for j in y]) 140 141 return result_x, result_y 142 143 def _guinier(self, x): 144 """ 145 Retrive the guinier function after apply an inverse guinier function 146 to x 147 Compute a F(x) = scale* e-((radius*x)**2/3). 148 @param x: a vector of q values 149 @param scale: the scale value 150 @param radius: the guinier radius value 151 @return F(x) 152 """ 153 # transform the radius of coming from the inverse guinier function to a 154 # a radius of a guinier function 155 if self.radius <= 0: 156 raise ValueError("Rg expected positive value, but got %s"%self.radius) 157 158 value = numpy.array([math.exp(-((self.radius * i)**2/3)) for i in x ]) 159 160 return self.scale * value 161 162 class PowerLaw(Transform): 163 """ 164 class of type transform that perform operation related to power_law 165 function 166 """ 167 def __init__(self, scale=1, power=4): 168 Transform.__init__(self) 169 self.scale = scale 170 self.power = power 171 172 def set_value(self, a, b): 173 b = -1 * b 174 if b <= 0: 175 raise ValueError("Power_law fit expected posive power, but got %s"%power) 176 a = math.exp(a) 177 self.scale = a 178 self.power = b 179 180 def transform_data(self, x, a=None, b=None): 181 """ 182 given a scale and a radius transform x, y using a power_law 183 function 184 """ 185 if a is not None: 186 self.scale = a 187 if b is not None: 188 self.power = b 189 190 return self._power_law(x) 191 192 def inverse_transform_data(self, x, y): 193 """ 194 given a scale and a radius transform x, y using a inverse power_law 195 function 196 """ 197 result_x = numpy.array([i * i for i in x]) 198 result_y = numpy.array([math.log(j) for j in y]) 199 200 return result_x, result_y 201 202 def _power_law(self, x): 203 """ 204 F(x) = scale* (x)^(-power) 205 when power= 4. the model is porod 206 else power_law 207 The model has three parameters: 208 @param x: a vector of q values 209 @param power: power of the function 210 @param scale : scale factor value 211 @param F(x) 212 """ 213 if self.power <= 0: 214 raise ValueError("Power_law function expected positive power, but got %s"%power) 215 if self.scale <= 0: 216 raise ValueError("scale expected positive value, but got %s"%self.scale) 217 218 value = numpy.array([ math.pow(i, -self.power) for i in x ]) 219 return self.scale * value 220 221 class Extrapolator: 59 222 """ 60 223 compute f(x) … … 66 229 """ 67 230 self.data = data 68 231 69 232 # Set qmin as the lowest non-zero value 70 233 self.qmin = Q_MINIMUM 71 234 for q_value in self.data.x: 72 if q_value >0:235 if q_value > 0: 73 236 self.qmin = q_value 74 237 break … … 76 239 77 240 #get the smear object of data 78 self.smearer = smear_selection( self.data)241 self.smearer = smear_selection(self.data) 79 242 # Set the q-range information to allow smearing 80 243 self.set_fit_range() … … 91 254 self._qmin_unsmeared = self.qmin 92 255 self._qmax_unsmeared = self.qmax 256 93 257 if self.smearer is not None: 94 258 self._first_unsmeared_bin, self._last_unsmeared_bin = self.smearer.get_bin_range(self.qmin, self.qmax) … … 107 271 """ 108 272 fx = numpy.zeros(len(self.data.x)) 109 # Uncertainty 273 274 # Uncertainty 110 275 if type(self.data.dy)==numpy.ndarray and len(self.data.dy)==len(self.data.x): 111 276 sigma = self.data.dy 112 277 else: 113 278 sigma = numpy.ones(len(self.data.x)) 114 279 115 280 # Compute theory data f(x) 116 281 fx[self.idx_unsmeared] = self.data.y[self.idx_unsmeared] … … 120 285 121 286 fx[self.idx_unsmeared] = fx[self.idx_unsmeared]/sigma[self.idx_unsmeared] 122 287 123 288 ##power is given only for function = power_law 124 289 if power != None: … … 171 336 # Extrapolation parameters 172 337 self._low_extrapolation_npts = 4 173 self._low_extrapolation_function = guinier338 self._low_extrapolation_function = Guinier() 174 339 self._low_extrapolation_power = None 175 340 176 341 self._high_extrapolation_npts = 4 177 self._high_extrapolation_function = power_law342 self._high_extrapolation_function = PowerLaw() 178 343 self._high_extrapolation_power = None 179 344 … … 192 357 # Copy data that is not copied by the operations 193 358 #TODO: fix this in DataLoader 194 new_data.dxl = data.dxl195 new_data.dxw = data.dxw359 #new_data.dxl = data.dxl 360 #new_data.dxw = data.dxw 196 361 197 362 return new_data … … 212 377 for power_law will be the power value 213 378 """ 214 #fit_x = numpy.array([math.log(x) for x in self._data.x]) 215 if function.__name__ == "guinier": 216 fit_x = numpy.array([x * x for x in self._data.x]) 217 qmin = qmin**2 218 qmax = qmax**2 219 fit_y = numpy.array([math.log(y) for y in self._data.y]) 220 fit_dy = numpy.array([y for y in self._data.y]) 221 fit_dy = numpy.array([dy for dy in self._data.dy])/fit_dy 222 223 elif function.__name__ == "power_law": 224 fit_x = numpy.array([math.log(x) for x in self._data.x]) 225 qmin = math.log(qmin) 226 qmax = math.log(qmax) 227 fit_y = numpy.array([math.log(y) for y in self._data.y]) 228 fit_dy = numpy.array([y for y in self._data.y]) 229 fit_dy = numpy.array([dy for dy in self._data.dy])/fit_dy 230 231 else: 232 raise ValueError("Unknown function used to fit %s"%function.__name__) 233 234 235 #else: 379 transform = function 380 fit_x, fit_y = transform.inverse_transform_data(self._data.x, self._data.y) 381 fit_dy = transform.transform_error(y=self._data.y, dy=self._data.dy) 382 qmin = transform.transform_value(value=qmin) 383 qmax = transform.transform_value(value=qmax) 384 236 385 fit_data = LoaderData1D(x=fit_x, y=fit_y, dy=fit_dy) 237 386 fit_data.dxl = self._data.dxl 238 387 fit_data.dxw = self._data.dxw 239 functor = FitFunctor(data=fit_data)388 functor = Extrapolator(data=fit_data) 240 389 functor.set_fit_range(qmin=qmin, qmax=qmax) 241 390 b, a = functor.fit(power=power) 242 243 244 if function.__name__ == "guinier": 245 # b is the radius value of the guinier function 246 if b>=0: 247 raise ValueError("Guinier fit was not converged") 248 else: 249 b = math.sqrt(-3 * b) 250 251 252 if function.__name__ == "power_law": 253 b = -1 * b 254 if b <= 0: 255 raise ValueError("Power_law fit expected posive power, but got %s"%power) 256 # a is the scale of the guinier function 257 a = math.exp(a) 258 259 return a, b 391 return b, a 260 392 261 393 def _get_qstar(self, data): … … 332 464 or len(data.x)!= len(data.dxl): 333 465 msg = "x, dxl, and y must be have the same length and greater than 1" 466 msg +=" len x=%s, y=%s, dxl=%s"%(len(data.x),len(data.y),len(data.dxl)) 334 467 raise ValueError, msg 335 468 else: … … 438 571 note: if data doesn't contain dy assume dy= math.sqrt(data.y) 439 572 """ 440 if data is None:441 data = self._data573 #if data is None: 574 # data = self._data 442 575 443 576 if not data.is_slit_smeared(): … … 529 662 # Extrapolate the low-Q data 530 663 #TODO: this fit fails. Fix it. 531 a, b= self._fit(function=self._low_extrapolation_function,664 b, a = self._fit(function=self._low_extrapolation_function, 532 665 qmin=qmin, 533 666 qmax=qmax, … … 539 672 q_start = qmin/10 540 673 541 #create new Data1D to compute the invariant 542 new_x = numpy.linspace(start=q_start, 543 stop=qmin, 544 num=INTEGRATION_NSTEPS, 545 endpoint=True) 546 new_y = self._low_extrapolation_function(x=new_x, scale=a, radius=b) 547 dxl = None 548 dxw = None 549 if self._data.dxl is not None: 550 dxl = numpy.ones(INTEGRATION_NSTEPS) 551 dxl = dxl * self._data.dxl[0] 552 if self._data.dxw is not None: 553 dxw = numpy.ones(INTEGRATION_NSTEPS) 554 dxw = dxw * self._data.dxw[0] 555 556 data_min = LoaderData1D(x=new_x, y=new_y) 557 data_min.dxl = dxl 558 data_min.dxw = dxw 559 self._data.clone_without_data( clone= data_min) 560 674 self._low_extrapolation_function.set_value(a= a, b=b) 675 data_min = self._low_extrapolation_function.get_extrapolated_data(data=self._data, 676 npts=INTEGRATION_NSTEPS, 677 q_start=q_start, q_end=qmin, smear_indice=0) 678 561 679 return data_min 562 680 … … 577 695 # Data boundaries for fiiting 578 696 x_len = len(self._data.x) - 1 579 q min= self._data.x[x_len - (self._high_extrapolation_npts - 1)]697 q_start = self._data.x[x_len - (self._high_extrapolation_npts - 1)] 580 698 qmax = self._data.x[x_len] 581 582 try: 583 # fit the data with a model to get the appropriate parameters 584 a, b = self._fit(function=self._high_extrapolation_function, 585 qmin=qmin, 586 qmax=qmax, 587 power=self._high_extrapolation_power) 588 except: 589 return None 590 699 smear_indice = x_len 700 701 # fit the data with a model to get the appropriate parameters 702 b, a = self._fit(function=self._high_extrapolation_function, 703 qmin=q_start, 704 qmax=qmax, 705 power=self._high_extrapolation_power) 706 591 707 #create new Data1D to compute the invariant 592 new_x = numpy.linspace(start=qmax, 593 stop=Q_MAXIMUM, 594 num=INTEGRATION_NSTEPS, 595 endpoint=True) 596 597 new_y = self._high_extrapolation_function(x=new_x, scale=a, power=b) 598 599 dxl = None 600 dxw = None 601 if self._data.dxl is not None: 602 dxl = numpy.ones(INTEGRATION_NSTEPS) 603 dxl = dxl * self._data.dxl[0] 604 if self._data.dxw is not None: 605 dxw = numpy.ones(INTEGRATION_NSTEPS) 606 dxw = dxw * self._data.dxw[0] 607 608 data_max = LoaderData1D(x=new_x, y=new_y) 609 data_max.dxl = dxl 610 data_max.dxw = dxw 611 self._data.clone_without_data(clone=data_max) 612 708 self._high_extrapolation_function.set_value(a=a, b=b) 709 data_max = self._high_extrapolation_function.get_extrapolated_data(data=self._data, 710 npts=INTEGRATION_NSTEPS, 711 q_start=q_start, q_end=qmax, 712 smear_indice=smear_indice) 613 713 return data_max 614 714 … … 638 738 else: 639 739 if function == 'power_law': 640 self._low_extrapolation_function = power_law740 self._low_extrapolation_function = PowerLaw() 641 741 else: 642 self._low_extrapolation_function = guinier742 self._low_extrapolation_function = Guinier() 643 743 self._low_extrapolation_npts = npts 644 744 self._low_extrapolation_power = power
Note: See TracChangeset
for help on using the changeset viewer.