1 | """ |
---|
2 | Data manipulations for 2D data sets. |
---|
3 | Using the meta data information, various types of averaging |
---|
4 | are performed in Q-space |
---|
5 | """ |
---|
6 | |
---|
7 | """ |
---|
8 | This software was developed by the University of Tennessee as part of the |
---|
9 | Distributed Data Analysis of Neutron Scattering Experiments (DANSE) |
---|
10 | project funded by the US National Science Foundation. |
---|
11 | |
---|
12 | See the license text in license.txt |
---|
13 | |
---|
14 | copyright 2008, University of Tennessee |
---|
15 | """ |
---|
16 | #TODO: copy the meta data from the 2D object to the resulting 1D object |
---|
17 | |
---|
18 | from data_info import plottable_2D, Data1D |
---|
19 | import math |
---|
20 | import numpy |
---|
21 | |
---|
22 | def get_q(dx, dy, det_dist, wavelength): |
---|
23 | """ |
---|
24 | @param dx: x-distance from beam center [mm] |
---|
25 | @param dy: y-distance from beam center [mm] |
---|
26 | @return: q-value at the given position |
---|
27 | """ |
---|
28 | # Distance from beam center in the plane of detector |
---|
29 | plane_dist = math.sqrt(dx*dx + dy*dy) |
---|
30 | # Half of the scattering angle |
---|
31 | theta = 0.5*math.atan(plane_dist/det_dist) |
---|
32 | return (4.0*math.pi/wavelength)*math.sin(theta) |
---|
33 | |
---|
34 | def get_q_compo(dx, dy, det_dist, wavelength,compo=None): |
---|
35 | #This reduces tiny error at very large q. |
---|
36 | #Implementation of this func is not started yet.<--ToDo |
---|
37 | if dy==0: |
---|
38 | if dx>=0: |
---|
39 | angle_xy=0 |
---|
40 | else: |
---|
41 | angle_xy=math.pi |
---|
42 | else: |
---|
43 | angle_xy=math.atan(dx/dy) |
---|
44 | |
---|
45 | if compo=="x": |
---|
46 | out=get_q(dx, dy, det_dist, wavelength)*cos(angle_xy) |
---|
47 | elif compo=="y": |
---|
48 | out=get_q(dx, dy, det_dist, wavelength)*sin(angle_xy) |
---|
49 | else: |
---|
50 | out=get_q(dx, dy, det_dist, wavelength) |
---|
51 | return out |
---|
52 | |
---|
53 | class _Slab(object): |
---|
54 | """ |
---|
55 | Compute average I(Q) for a region of interest |
---|
56 | """ |
---|
57 | def __init__(self, x_min=0.0, x_max=0.0, y_min=0.0, y_max=0.0, bin_width=0.001): |
---|
58 | # Minimum Qx value [A-1] |
---|
59 | self.x_min = x_min |
---|
60 | # Maximum Qx value [A-1] |
---|
61 | self.x_max = x_max |
---|
62 | # Minimum Qy value [A-1] |
---|
63 | self.y_min = y_min |
---|
64 | # Maximum Qy value [A-1] |
---|
65 | self.y_max = y_max |
---|
66 | # Bin width (step size) [A-1] |
---|
67 | self.bin_width = bin_width |
---|
68 | # If True, I(|Q|) will be return, otherwise, negative q-values are allowed |
---|
69 | self.fold = False |
---|
70 | |
---|
71 | def __call__(self, data2D): return NotImplemented |
---|
72 | |
---|
73 | def _avg(self, data2D, maj): |
---|
74 | """ |
---|
75 | Compute average I(Q_maj) for a region of interest. |
---|
76 | The major axis is defined as the axis of Q_maj. |
---|
77 | The minor axis is the axis that we average over. |
---|
78 | |
---|
79 | @param data2D: Data2D object |
---|
80 | @param maj_min: min value on the major axis |
---|
81 | @return: Data1D object |
---|
82 | """ |
---|
83 | if len(data2D.detector) != 1: |
---|
84 | raise RuntimeError, "_Slab._avg: invalid number of detectors: %g" % len(data2D.detector) |
---|
85 | |
---|
86 | pixel_width_x = data2D.detector[0].pixel_size.x |
---|
87 | pixel_width_y = data2D.detector[0].pixel_size.y |
---|
88 | det_dist = data2D.detector[0].distance |
---|
89 | wavelength = data2D.source.wavelength |
---|
90 | center_x = data2D.detector[0].beam_center.x/pixel_width_x |
---|
91 | center_y = data2D.detector[0].beam_center.y/pixel_width_y |
---|
92 | |
---|
93 | # Build array of Q intervals |
---|
94 | if maj=='x': |
---|
95 | nbins = int(math.ceil((self.x_max-self.x_min)/self.bin_width)) |
---|
96 | qbins = self.bin_width*numpy.arange(nbins)+self.x_min |
---|
97 | elif maj=='y': |
---|
98 | nbins = int(math.ceil((self.y_max-self.y_min)/self.bin_width)) |
---|
99 | qbins = self.bin_width*numpy.arange(nbins)+self.y_min |
---|
100 | else: |
---|
101 | raise RuntimeError, "_Slab._avg: unrecognized axis %s" % str(maj) |
---|
102 | |
---|
103 | x = numpy.zeros(nbins) |
---|
104 | y = numpy.zeros(nbins) |
---|
105 | err_y = numpy.zeros(nbins) |
---|
106 | y_counts = numpy.zeros(nbins) |
---|
107 | |
---|
108 | for i in range(numpy.size(data2D.data,1)): |
---|
109 | # Min and max x-value for the pixel |
---|
110 | minx = pixel_width_x*(i - center_x) |
---|
111 | maxx = pixel_width_x*(i+1.0 - center_x) |
---|
112 | |
---|
113 | qxmin = get_q(minx, 0.0, det_dist, wavelength) |
---|
114 | qxmax = get_q(maxx, 0.0, det_dist, wavelength) |
---|
115 | |
---|
116 | # Get the count fraction in x for that pixel |
---|
117 | frac_min = get_pixel_fraction_square(self.x_min, qxmin, qxmax) |
---|
118 | frac_max = get_pixel_fraction_square(self.x_max, qxmin, qxmax) |
---|
119 | frac_x = frac_max - frac_min |
---|
120 | |
---|
121 | if frac_x == 0: |
---|
122 | continue |
---|
123 | |
---|
124 | if maj=='x': |
---|
125 | dx = pixel_width_x*(i+0.5 - center_x) |
---|
126 | q_value = get_q(dx, 0.0, det_dist, wavelength) |
---|
127 | if self.fold==False and dx<0: |
---|
128 | q_value = -q_value |
---|
129 | i_q = int(math.ceil((q_value-self.x_min)/self.bin_width)) - 1 |
---|
130 | |
---|
131 | if i_q<0 or i_q>=nbins: |
---|
132 | continue |
---|
133 | |
---|
134 | for j in range(numpy.size(data2D.data,0)): |
---|
135 | # Min and max y-value for the pixel |
---|
136 | miny = pixel_width_y*(j - center_y) |
---|
137 | maxy = pixel_width_y*(j+1.0 - center_y) |
---|
138 | |
---|
139 | qymin = get_q(0.0, miny, det_dist, wavelength) |
---|
140 | qymax = get_q(0.0, maxy, det_dist, wavelength) |
---|
141 | |
---|
142 | # Get the count fraction in x for that pixel |
---|
143 | frac_min = get_pixel_fraction_square(self.y_min, qymin, qymax) |
---|
144 | frac_max = get_pixel_fraction_square(self.y_max, qymin, qymax) |
---|
145 | frac_y = frac_max - frac_min |
---|
146 | |
---|
147 | frac = frac_x * frac_y |
---|
148 | |
---|
149 | if frac == 0: |
---|
150 | continue |
---|
151 | |
---|
152 | if maj=='y': |
---|
153 | dy = pixel_width_y*(j+0.5 - center_y) |
---|
154 | q_value = get_q(0.0, dy, det_dist, wavelength) |
---|
155 | if self.fold==False and dy<0: |
---|
156 | q_value = -q_value |
---|
157 | i_q = int(math.ceil((q_value-self.y_min)/self.bin_width)) - 1 |
---|
158 | |
---|
159 | if i_q<0 or i_q>=nbins: |
---|
160 | continue |
---|
161 | |
---|
162 | x[i_q] = q_value |
---|
163 | y[i_q] += frac * data2D.data[j][i] |
---|
164 | if data2D.err_data == None or data2D.err_data[j][i]==0.0: |
---|
165 | err_y[i_q] += frac * frac * math.fabs(data2D.data[j][i]) |
---|
166 | else: |
---|
167 | err_y[i_q] += frac * frac * data2D.err_data[j][i] * data2D.err_data[j][i] |
---|
168 | y_counts[i_q] += frac |
---|
169 | |
---|
170 | # Average the sums |
---|
171 | for i in range(nbins): |
---|
172 | if y_counts[i]>0: |
---|
173 | err_y[i] = math.sqrt(err_y[i])/y_counts[i] |
---|
174 | y[i] = y[i]/y_counts[i] |
---|
175 | |
---|
176 | return Data1D(x=x, y=y, dy=err_y) |
---|
177 | |
---|
178 | class SlabY(_Slab): |
---|
179 | """ |
---|
180 | Compute average I(Qy) for a region of interest |
---|
181 | """ |
---|
182 | def __call__(self, data2D): |
---|
183 | """ |
---|
184 | Compute average I(Qy) for a region of interest |
---|
185 | |
---|
186 | @param data2D: Data2D object |
---|
187 | @return: Data1D object |
---|
188 | """ |
---|
189 | return self._avg(data2D, 'y') |
---|
190 | |
---|
191 | class SlabX(_Slab): |
---|
192 | """ |
---|
193 | Compute average I(Qx) for a region of interest |
---|
194 | """ |
---|
195 | def __call__(self, data2D): |
---|
196 | """ |
---|
197 | Compute average I(Qx) for a region of interest |
---|
198 | |
---|
199 | @param data2D: Data2D object |
---|
200 | @return: Data1D object |
---|
201 | """ |
---|
202 | return self._avg(data2D, 'x') |
---|
203 | |
---|
204 | class Boxsum(object): |
---|
205 | """ |
---|
206 | Perform the sum of counts in a 2D region of interest. |
---|
207 | """ |
---|
208 | def __init__(self, x_min=0.0, x_max=0.0, y_min=0.0, y_max=0.0): |
---|
209 | # Minimum Qx value [A-1] |
---|
210 | self.x_min = x_min |
---|
211 | # Maximum Qx value [A-1] |
---|
212 | self.x_max = x_max |
---|
213 | # Minimum Qy value [A-1] |
---|
214 | self.y_min = y_min |
---|
215 | # Maximum Qy value [A-1] |
---|
216 | self.y_max = y_max |
---|
217 | |
---|
218 | def __call__(self, data2D): |
---|
219 | """ |
---|
220 | Perform the sum in the region of interest |
---|
221 | |
---|
222 | @param data2D: Data2D object |
---|
223 | @return: number of counts, error on number of counts |
---|
224 | """ |
---|
225 | y, err_y, y_counts = self._sum(data2D) |
---|
226 | |
---|
227 | # Average the sums |
---|
228 | counts = 0 if y_counts==0 else y |
---|
229 | error = 0 if y_counts==0 else math.sqrt(err_y) |
---|
230 | |
---|
231 | return counts, error |
---|
232 | |
---|
233 | def _sum(self, data2D): |
---|
234 | """ |
---|
235 | Perform the sum in the region of interest |
---|
236 | @param data2D: Data2D object |
---|
237 | @return: number of counts, error on number of counts, number of entries summed |
---|
238 | """ |
---|
239 | if len(data2D.detector) != 1: |
---|
240 | raise RuntimeError, "Circular averaging: invalid number of detectors: %g" % len(data2D.detector) |
---|
241 | |
---|
242 | pixel_width_x = data2D.detector[0].pixel_size.x |
---|
243 | pixel_width_y = data2D.detector[0].pixel_size.y |
---|
244 | det_dist = data2D.detector[0].distance |
---|
245 | wavelength = data2D.source.wavelength |
---|
246 | center_x = data2D.detector[0].beam_center.x/pixel_width_x |
---|
247 | center_y = data2D.detector[0].beam_center.y/pixel_width_y |
---|
248 | |
---|
249 | y = 0.0 |
---|
250 | err_y = 0.0 |
---|
251 | y_counts = 0.0 |
---|
252 | sign=1 |
---|
253 | for i in range(numpy.size(data2D.data,1)): |
---|
254 | # Min and max x-value for the pixel |
---|
255 | minx = pixel_width_x*(i - center_x) |
---|
256 | maxx = pixel_width_x*(i+1.0 - center_x) |
---|
257 | if minx>=0: |
---|
258 | sign=1 |
---|
259 | else: |
---|
260 | sign=-1 |
---|
261 | qxmin = sign*get_q(minx, 0.0, det_dist, wavelength) |
---|
262 | if maxx>=0: |
---|
263 | sign=1 |
---|
264 | else: |
---|
265 | sign=-1 |
---|
266 | qxmax = sign*get_q(maxx, 0.0, det_dist, wavelength) |
---|
267 | |
---|
268 | # Get the count fraction in x for that pixel |
---|
269 | frac_min = get_pixel_fraction_square(self.x_min, qxmin, qxmax) |
---|
270 | frac_max = get_pixel_fraction_square(self.x_max, qxmin, qxmax) |
---|
271 | frac_x = frac_max - frac_min |
---|
272 | |
---|
273 | for j in range(numpy.size(data2D.data,0)): |
---|
274 | # Min and max y-value for the pixel |
---|
275 | miny = pixel_width_y*(j - center_y) |
---|
276 | maxy = pixel_width_y*(j+1.0 - center_y) |
---|
277 | if miny>=0: |
---|
278 | sign=1 |
---|
279 | else: |
---|
280 | sign=-1 |
---|
281 | |
---|
282 | qymin = sign*get_q(0.0, miny, det_dist, wavelength) |
---|
283 | if maxy>=0: |
---|
284 | sign=1 |
---|
285 | else: |
---|
286 | sign=-1 |
---|
287 | |
---|
288 | qymax = sign*get_q(0.0, maxy, det_dist, wavelength) |
---|
289 | |
---|
290 | # Get the count fraction in x for that pixel |
---|
291 | frac_min = get_pixel_fraction_square(self.y_min, qymin, qymax) |
---|
292 | frac_max = get_pixel_fraction_square(self.y_max, qymin, qymax) |
---|
293 | frac_y = frac_max - frac_min |
---|
294 | |
---|
295 | frac = frac_x * frac_y |
---|
296 | |
---|
297 | y += frac * data2D.data[j][i] |
---|
298 | if data2D.err_data == None or data2D.err_data[j][i]==0.0: |
---|
299 | err_y += frac * frac * math.fabs(data2D.data[j][i]) |
---|
300 | else: |
---|
301 | err_y += frac * frac * data2D.err_data[j][i] * data2D.err_data[j][i] |
---|
302 | y_counts += frac |
---|
303 | |
---|
304 | return y, err_y, y_counts |
---|
305 | |
---|
306 | class Boxavg(Boxsum): |
---|
307 | """ |
---|
308 | Perform the average of counts in a 2D region of interest. |
---|
309 | """ |
---|
310 | def __init__(self, x_min=0.0, x_max=0.0, y_min=0.0, y_max=0.0): |
---|
311 | super(Boxavg, self).__init__(x_min=x_min, x_max=x_max, y_min=y_min, y_max=y_max) |
---|
312 | |
---|
313 | def __call__(self, data2D): |
---|
314 | """ |
---|
315 | Perform the sum in the region of interest |
---|
316 | |
---|
317 | @param data2D: Data2D object |
---|
318 | @return: average counts, error on average counts |
---|
319 | """ |
---|
320 | y, err_y, y_counts = self._sum(data2D) |
---|
321 | |
---|
322 | # Average the sums |
---|
323 | counts = 0 if y_counts==0 else y/y_counts |
---|
324 | error = 0 if y_counts==0 else math.sqrt(err_y)/y_counts |
---|
325 | |
---|
326 | return counts, error |
---|
327 | |
---|
328 | def get_pixel_fraction_square(x, xmin, xmax): |
---|
329 | """ |
---|
330 | Return the fraction of the length |
---|
331 | from xmin to x. |
---|
332 | |
---|
333 | A B |
---|
334 | +-----------+---------+ |
---|
335 | xmin x xmax |
---|
336 | |
---|
337 | @param x: x-value |
---|
338 | @param xmin: minimum x for the length considered |
---|
339 | @param xmax: minimum x for the length considered |
---|
340 | @return: (x-xmin)/(xmax-xmin) when xmin < x < xmax |
---|
341 | |
---|
342 | """ |
---|
343 | if x<=xmin: |
---|
344 | return 0.0 |
---|
345 | if x>xmin and x<xmax: |
---|
346 | return (x-xmin)/(xmax-xmin) |
---|
347 | else: |
---|
348 | return 1.0 |
---|
349 | |
---|
350 | |
---|
351 | class CircularAverage(object): |
---|
352 | """ |
---|
353 | Perform circular averaging on 2D data |
---|
354 | |
---|
355 | The data returned is the distribution of counts |
---|
356 | as a function of Q |
---|
357 | """ |
---|
358 | def __init__(self, r_min=0.0, r_max=0.0, bin_width=0.001): |
---|
359 | # Minimum radius included in the average [A-1] |
---|
360 | self.r_min = r_min |
---|
361 | # Maximum radius included in the average [A-1] |
---|
362 | self.r_max = r_max |
---|
363 | # Bin width (step size) [A-1] |
---|
364 | self.bin_width = bin_width |
---|
365 | |
---|
366 | def __call__(self, data2D): |
---|
367 | """ |
---|
368 | Perform circular averaging on the data |
---|
369 | |
---|
370 | @param data2D: Data2D object |
---|
371 | @return: Data1D object |
---|
372 | """ |
---|
373 | if len(data2D.detector) != 1: |
---|
374 | raise RuntimeError, "Circular averaging: invalid number of detectors: %g" % len(data2D.detector) |
---|
375 | |
---|
376 | pixel_width_x = data2D.detector[0].pixel_size.x |
---|
377 | pixel_width_y = data2D.detector[0].pixel_size.y |
---|
378 | det_dist = data2D.detector[0].distance |
---|
379 | wavelength = data2D.source.wavelength |
---|
380 | center_x = data2D.detector[0].beam_center.x/pixel_width_x |
---|
381 | center_y = data2D.detector[0].beam_center.y/pixel_width_y |
---|
382 | |
---|
383 | # Find out the maximum Q range |
---|
384 | xwidth = numpy.size(data2D.data,1)*pixel_width_x |
---|
385 | dx_max = xwidth - data2D.detector[0].beam_center.x |
---|
386 | if xwidth-dx_max>dx_max: |
---|
387 | dx_max = xwidth-dx_max |
---|
388 | |
---|
389 | ywidth = numpy.size(data2D.data,0)*pixel_width_y |
---|
390 | dy_max = ywidth - data2D.detector[0].beam_center.y |
---|
391 | if ywidth-dy_max>dy_max: |
---|
392 | dy_max = ywidth-dy_max |
---|
393 | |
---|
394 | qmax = get_q(dx_max, dy_max, det_dist, wavelength) |
---|
395 | |
---|
396 | # Build array of Q intervals |
---|
397 | nbins = int(math.ceil((qmax-self.r_min)/self.bin_width)) |
---|
398 | qbins = self.bin_width*numpy.arange(nbins)+self.r_min |
---|
399 | |
---|
400 | x = numpy.zeros(nbins) |
---|
401 | y = numpy.zeros(nbins) |
---|
402 | err_y = numpy.zeros(nbins) |
---|
403 | y_counts = numpy.zeros(nbins) |
---|
404 | |
---|
405 | for i in range(numpy.size(data2D.data,1)): |
---|
406 | dx = pixel_width_x*(i+0.5 - center_x) |
---|
407 | |
---|
408 | # Min and max x-value for the pixel |
---|
409 | minx = pixel_width_x*(i - center_x) |
---|
410 | maxx = pixel_width_x*(i+1.0 - center_x) |
---|
411 | |
---|
412 | for j in range(numpy.size(data2D.data,0)): |
---|
413 | dy = pixel_width_y*(j+0.5 - center_y) |
---|
414 | |
---|
415 | q_value = get_q(dx, dy, det_dist, wavelength) |
---|
416 | |
---|
417 | # Min and max y-value for the pixel |
---|
418 | miny = pixel_width_y*(j - center_y) |
---|
419 | maxy = pixel_width_y*(j+1.0 - center_y) |
---|
420 | |
---|
421 | # Calculate the q-value for each corner |
---|
422 | # q_[x min or max][y min or max] |
---|
423 | q_00 = get_q(minx, miny, det_dist, wavelength) |
---|
424 | q_01 = get_q(minx, maxy, det_dist, wavelength) |
---|
425 | q_10 = get_q(maxx, miny, det_dist, wavelength) |
---|
426 | q_11 = get_q(maxx, maxy, det_dist, wavelength) |
---|
427 | |
---|
428 | # Look for intercept between each side of the pixel |
---|
429 | # and the constant-q ring for qmax |
---|
430 | frac_max = get_pixel_fraction(self.r_max, q_00, q_01, q_10, q_11) |
---|
431 | |
---|
432 | # Look for intercept between each side of the pixel |
---|
433 | # and the constant-q ring for qmin |
---|
434 | frac_min = get_pixel_fraction(self.r_min, q_00, q_01, q_10, q_11) |
---|
435 | |
---|
436 | # We are interested in the region between qmin and qmax |
---|
437 | # therefore the fraction of the surface of the pixel |
---|
438 | # that we will use to calculate the number of counts to |
---|
439 | # include is given by: |
---|
440 | frac = frac_max - frac_min |
---|
441 | |
---|
442 | i_q = int(math.ceil((q_value-self.r_min)/self.bin_width)) - 1 |
---|
443 | |
---|
444 | x[i_q] = q_value |
---|
445 | y[i_q] += frac * data2D.data[j][i] |
---|
446 | if data2D.err_data == None or data2D.err_data[j][i]==0.0: |
---|
447 | err_y[i_q] += frac * frac * math.fabs(data2D.data[j][i]) |
---|
448 | else: |
---|
449 | err_y[i_q] += frac * frac * data2D.err_data[j][i] * data2D.err_data[j][i] |
---|
450 | y_counts[i_q] += frac |
---|
451 | |
---|
452 | # Average the sums |
---|
453 | for i in range(nbins): |
---|
454 | if y_counts[i]>0: |
---|
455 | err_y[i] = math.sqrt(err_y[i])/y_counts[i] |
---|
456 | y[i] = y[i]/y_counts[i] |
---|
457 | |
---|
458 | return Data1D(x=x, y=y, dy=err_y) |
---|
459 | |
---|
460 | |
---|
461 | class Ring(object): |
---|
462 | """ |
---|
463 | Defines a ring on a 2D data set. |
---|
464 | The ring is defined by r_min, r_max, and |
---|
465 | the position of the center of the ring. |
---|
466 | |
---|
467 | The data returned is the distribution of counts |
---|
468 | around the ring as a function of phi. |
---|
469 | |
---|
470 | """ |
---|
471 | |
---|
472 | def __init__(self, r_min=0, r_max=0, center_x=0, center_y=0,nbins=20 ): |
---|
473 | # Minimum radius |
---|
474 | self.r_min = r_min |
---|
475 | # Maximum radius |
---|
476 | self.r_max = r_max |
---|
477 | # Center of the ring in x |
---|
478 | self.center_x = center_x |
---|
479 | # Center of the ring in y |
---|
480 | self.center_y = center_y |
---|
481 | # Number of angular bins |
---|
482 | self.nbins_phi = nbins_phi |
---|
483 | |
---|
484 | def __call__(self, data2D): |
---|
485 | """ |
---|
486 | Apply the ring to the data set. |
---|
487 | Returns the angular distribution for a given q range |
---|
488 | |
---|
489 | @param data2D: Data2D object |
---|
490 | @return: Data1D object |
---|
491 | """ |
---|
492 | if data2D.__class__.__name__ not in ["Data2D", "plottable_2D"]: |
---|
493 | raise RuntimeError, "Ring averaging only take plottable_2D objects" |
---|
494 | |
---|
495 | data = data2D.data |
---|
496 | qmin = self.r_min |
---|
497 | qmax = self.r_max |
---|
498 | |
---|
499 | if len(data2D.detector) != 1: |
---|
500 | raise RuntimeError, "Ring averaging: invalid number of detectors: %g" % len(data2D.detector) |
---|
501 | pixel_width_x = data2D.detector[0].pixel_size.x |
---|
502 | pixel_width_y = data2D.detector[0].pixel_size.y |
---|
503 | det_dist = data2D.detector[0].distance |
---|
504 | wavelength = data2D.source.wavelength |
---|
505 | #center_x = self.center_x/pixel_width_x |
---|
506 | #center_y = self.center_y/pixel_width_y |
---|
507 | center_x = data2D.detector[0].beam_center.x/pixel_width_x |
---|
508 | center_y = data2D.detector[0].beam_center.y/pixel_width_y |
---|
509 | |
---|
510 | |
---|
511 | phi_bins = numpy.zeros(self.nbins_phi) |
---|
512 | phi_counts = numpy.zeros(self.nbins_phi) |
---|
513 | phi_values = numpy.zeros(self.nbins_phi) |
---|
514 | phi_err = numpy.zeros(self.nbins_phi) |
---|
515 | |
---|
516 | for i in range(numpy.size(data,1)): |
---|
517 | dx = pixel_width_x*(i+0.5 - center_x) |
---|
518 | |
---|
519 | # Min and max x-value for the pixel |
---|
520 | minx = pixel_width_x*(i - center_x) |
---|
521 | maxx = pixel_width_x*(i+1.0 - center_x) |
---|
522 | |
---|
523 | for j in range(numpy.size(data,0)): |
---|
524 | dy = pixel_width_y*(j+0.5 - center_y) |
---|
525 | |
---|
526 | q_value = get_q(dx, dy, det_dist, wavelength) |
---|
527 | |
---|
528 | # Min and max y-value for the pixel |
---|
529 | miny = pixel_width_y*(j - center_y) |
---|
530 | maxy = pixel_width_y*(j+1.0 - center_y) |
---|
531 | |
---|
532 | # Calculate the q-value for each corner |
---|
533 | # q_[x min or max][y min or max] |
---|
534 | q_00 = get_q(minx, miny, det_dist, wavelength) |
---|
535 | q_01 = get_q(minx, maxy, det_dist, wavelength) |
---|
536 | q_10 = get_q(maxx, miny, det_dist, wavelength) |
---|
537 | q_11 = get_q(maxx, maxy, det_dist, wavelength) |
---|
538 | |
---|
539 | # Look for intercept between each side of the pixel |
---|
540 | # and the constant-q ring for qmax |
---|
541 | frac_max = get_pixel_fraction(qmax, q_00, q_01, q_10, q_11) |
---|
542 | |
---|
543 | # Look for intercept between each side of the pixel |
---|
544 | # and the constant-q ring for qmin |
---|
545 | frac_min = get_pixel_fraction(qmin, q_00, q_01, q_10, q_11) |
---|
546 | |
---|
547 | # We are interested in the region between qmin and qmax |
---|
548 | # therefore the fraction of the surface of the pixel |
---|
549 | # that we will use to calculate the number of counts to |
---|
550 | # include is given by: |
---|
551 | |
---|
552 | frac = frac_max - frac_min |
---|
553 | |
---|
554 | i_phi = int(math.ceil(self.nbins_phi*(math.atan2(dy, dx)+math.pi)/(2.0*math.pi))) - 1 |
---|
555 | |
---|
556 | phi_bins[i_phi] += frac * data[j][i] |
---|
557 | |
---|
558 | if data2D.err_data == None or data2D.err_data[j][i]==0.0: |
---|
559 | phi_err[i_phi] += frac * frac * math.fabs(data2D.data[j][i]) |
---|
560 | else: |
---|
561 | phi_err[i_phi] += frac * frac * data2D.err_data[j][i] * data2D.err_data[j][i] |
---|
562 | phi_counts[i_phi] += frac |
---|
563 | |
---|
564 | for i in range(self.nbins_phi): |
---|
565 | phi_bins[i] = phi_bins[i] / phi_counts[i] |
---|
566 | phi_err[i] = math.sqrt(phi_err[i]) / phi_counts[i] |
---|
567 | phi_values[i] = 2.0*math.pi/self.nbins_phi*(1.0*i + 0.5)-math.pi # move the pi back to -pi <-->+pi |
---|
568 | |
---|
569 | return Data1D(x=phi_values, y=phi_bins, dy=phi_err) |
---|
570 | |
---|
571 | def get_pixel_fraction(qmax, q_00, q_01, q_10, q_11): |
---|
572 | """ |
---|
573 | Returns the fraction of the pixel defined by |
---|
574 | the four corners (q_00, q_01, q_10, q_11) that |
---|
575 | has q < qmax. |
---|
576 | |
---|
577 | q_01 q_11 |
---|
578 | y=1 +--------------+ |
---|
579 | | | |
---|
580 | | | |
---|
581 | | | |
---|
582 | y=0 +--------------+ |
---|
583 | q_00 q_01 |
---|
584 | |
---|
585 | x=0 x=1 |
---|
586 | |
---|
587 | """ |
---|
588 | |
---|
589 | # y side for x = minx |
---|
590 | x_0 = get_intercept(qmax, q_00, q_01) |
---|
591 | # y side for x = maxx |
---|
592 | x_1 = get_intercept(qmax, q_10, q_11) |
---|
593 | |
---|
594 | # x side for y = miny |
---|
595 | y_0 = get_intercept(qmax, q_00, q_10) |
---|
596 | # x side for y = maxy |
---|
597 | y_1 = get_intercept(qmax, q_01, q_11) |
---|
598 | |
---|
599 | # surface fraction for a 1x1 pixel |
---|
600 | frac_max = 0 |
---|
601 | |
---|
602 | if x_0 and x_1: |
---|
603 | frac_max = (x_0+x_1)/2.0 |
---|
604 | |
---|
605 | elif y_0 and y_1: |
---|
606 | frac_max = (y_0+y_1)/2.0 |
---|
607 | |
---|
608 | elif x_0 and y_0: |
---|
609 | if q_00 < q_10: |
---|
610 | frac_max = x_0*y_0/2.0 |
---|
611 | else: |
---|
612 | frac_max = 1.0-x_0*y_0/2.0 |
---|
613 | |
---|
614 | elif x_0 and y_1: |
---|
615 | if q_00 < q_10: |
---|
616 | frac_max = x_0*y_1/2.0 |
---|
617 | else: |
---|
618 | frac_max = 1.0-x_0*y_1/2.0 |
---|
619 | |
---|
620 | elif x_1 and y_0: |
---|
621 | if q_00 > q_10: |
---|
622 | frac_max = x_1*y_0/2.0 |
---|
623 | else: |
---|
624 | frac_max = 1.0-x_1*y_0/2.0 |
---|
625 | |
---|
626 | elif x_1 and y_1: |
---|
627 | if q_00 < q_10: |
---|
628 | frac_max = 1.0 - (1.0-x_1)*(1.0-y_1)/2.0 |
---|
629 | else: |
---|
630 | frac_max = (1.0-x_1)*(1.0-y_1)/2.0 |
---|
631 | |
---|
632 | # If we make it here, there is no intercept between |
---|
633 | # this pixel and the constant-q ring. We only need |
---|
634 | # to know if we have to include it or exclude it. |
---|
635 | elif (q_00+q_01+q_10+q_11)/4.0 < qmax: |
---|
636 | frac_max = 1.0 |
---|
637 | |
---|
638 | return frac_max |
---|
639 | |
---|
640 | def get_intercept(q, q_0, q_1): |
---|
641 | """ |
---|
642 | Returns the fraction of the side at which the |
---|
643 | q-value intercept the pixel, None otherwise. |
---|
644 | The values returned is the fraction ON THE SIDE |
---|
645 | OF THE LOWEST Q. |
---|
646 | |
---|
647 | |
---|
648 | |
---|
649 | A B |
---|
650 | +-----------+--------+ |
---|
651 | 0 1 <--- pixel size |
---|
652 | |
---|
653 | Q_0 -------- Q ----- Q_1 <--- equivalent Q range |
---|
654 | |
---|
655 | |
---|
656 | if Q_1 > Q_0, A is returned |
---|
657 | if Q_1 < Q_0, B is returned |
---|
658 | |
---|
659 | if Q is outside the range of [Q_0, Q_1], None is returned |
---|
660 | |
---|
661 | """ |
---|
662 | if q_1 > q_0: |
---|
663 | if (q > q_0 and q <= q_1): |
---|
664 | return (q-q_0)/(q_1 - q_0) |
---|
665 | else: |
---|
666 | if (q > q_1 and q <= q_0): |
---|
667 | return (q-q_1)/(q_0 - q_1) |
---|
668 | return None |
---|
669 | |
---|
670 | #This class can be removed. |
---|
671 | class _Sectorold: |
---|
672 | """ |
---|
673 | Defines a sector region on a 2D data set. |
---|
674 | The sector is defined by r_min, r_max, phi_min, phi_max, |
---|
675 | and the position of the center of the ring. |
---|
676 | Phi is defined between 0 and 2pi |
---|
677 | """ |
---|
678 | def __init__(self, r_min, r_max, phi_min, phi_max,nbins=20): |
---|
679 | self.r_min = r_min |
---|
680 | self.r_max = r_max |
---|
681 | self.phi_min = phi_min |
---|
682 | self.phi_max = phi_max |
---|
683 | self.nbins = nbins |
---|
684 | |
---|
685 | def _agv(self, data2D, run='phi'): |
---|
686 | """ |
---|
687 | Perform sector averaging. |
---|
688 | |
---|
689 | @param data2D: Data2D object |
---|
690 | @param run: define the varying parameter ('phi' or 'q') |
---|
691 | @return: Data1D object |
---|
692 | """ |
---|
693 | if data2D.__class__.__name__ not in ["Data2D", "plottable_2D"]: |
---|
694 | raise RuntimeError, "Ring averaging only take plottable_2D objects" |
---|
695 | |
---|
696 | data = data2D.data |
---|
697 | qmax = self.r_max |
---|
698 | qmin = self.r_min |
---|
699 | |
---|
700 | if len(data2D.detector) != 1: |
---|
701 | raise RuntimeError, "Ring averaging: invalid number of detectors: %g" % len(data2D.detector) |
---|
702 | pixel_width_x = data2D.detector[0].pixel_size.x |
---|
703 | pixel_width_y = data2D.detector[0].pixel_size.y |
---|
704 | det_dist = data2D.detector[0].distance |
---|
705 | wavelength = data2D.source.wavelength |
---|
706 | center_x = data2D.detector[0].beam_center.x/pixel_width_x |
---|
707 | center_y = data2D.detector[0].beam_center.y/pixel_width_y |
---|
708 | |
---|
709 | y = numpy.zeros(self.nbins) |
---|
710 | y_counts = numpy.zeros(self.nbins) |
---|
711 | x = numpy.zeros(self.nbins) |
---|
712 | y_err = numpy.zeros(self.nbins) |
---|
713 | |
---|
714 | for i in range(numpy.size(data,1)): |
---|
715 | dx = pixel_width_x*(i+0.5 - center_x) |
---|
716 | |
---|
717 | # Min and max x-value for the pixel |
---|
718 | minx = pixel_width_x*(i - center_x) |
---|
719 | maxx = pixel_width_x*(i+1.0 - center_x) |
---|
720 | |
---|
721 | for j in range(numpy.size(data,0)): |
---|
722 | dy = pixel_width_y*(j+0.5 - center_y) |
---|
723 | |
---|
724 | q_value = get_q(dx, dy, det_dist, wavelength) |
---|
725 | |
---|
726 | # Min and max y-value for the pixel |
---|
727 | miny = pixel_width_y*(j - center_y) |
---|
728 | maxy = pixel_width_y*(j+1.0 - center_y) |
---|
729 | |
---|
730 | # Calculate the q-value for each corner |
---|
731 | # q_[x min or max][y min or max] |
---|
732 | q_00 = get_q(minx, miny, det_dist, wavelength) |
---|
733 | q_01 = get_q(minx, maxy, det_dist, wavelength) |
---|
734 | q_10 = get_q(maxx, miny, det_dist, wavelength) |
---|
735 | q_11 = get_q(maxx, maxy, det_dist, wavelength) |
---|
736 | |
---|
737 | # Look for intercept between each side of the pixel |
---|
738 | # and the constant-q ring for qmax |
---|
739 | frac_max = get_pixel_fraction(qmax, q_00, q_01, q_10, q_11) |
---|
740 | |
---|
741 | # Look for intercept between each side of the pixel |
---|
742 | # and the constant-q ring for qmin |
---|
743 | frac_min = get_pixel_fraction(qmin, q_00, q_01, q_10, q_11) |
---|
744 | |
---|
745 | # We are interested in the region between qmin and qmax |
---|
746 | # therefore the fraction of the surface of the pixel |
---|
747 | # that we will use to calculate the number of counts to |
---|
748 | # include is given by: |
---|
749 | |
---|
750 | frac = frac_max - frac_min |
---|
751 | |
---|
752 | # Compute phi and check whether it's within the limits |
---|
753 | phi_value=math.atan2(dy,dx)+math.pi |
---|
754 | # if phi_value<self.phi_min or phi_value>self.phi_max: |
---|
755 | if phi_value<self.phi_min or phi_value>self.phi_max: |
---|
756 | continue |
---|
757 | |
---|
758 | # Check which type of averaging we need |
---|
759 | if run.lower()=='phi': |
---|
760 | i_bin = int(math.ceil(self.nbins*(phi_value-self.phi_min)/(self.phi_max-self.phi_min))) - 1 |
---|
761 | else: |
---|
762 | # If we don't need this pixel, skip the rest of the work |
---|
763 | #TODO: an improvement here would be to compute the average |
---|
764 | # Q for the pixel from the part that is covered by |
---|
765 | # the ring defined by q_min/q_max rather than the complete |
---|
766 | # pixel |
---|
767 | if q_value<self.r_min or q_value>self.r_max: |
---|
768 | continue |
---|
769 | i_bin = int(math.ceil(self.nbins*(q_value-self.r_min)/(self.r_max-self.r_min))) - 1 |
---|
770 | |
---|
771 | try: |
---|
772 | y[i_bin] += frac * data[j][i] |
---|
773 | except: |
---|
774 | import sys |
---|
775 | print sys.exc_value |
---|
776 | print i_bin, frac |
---|
777 | |
---|
778 | if data2D.err_data == None or data2D.err_data[j][i]==0.0: |
---|
779 | y_err[i_bin] += frac * frac * math.fabs(data2D.data[j][i]) |
---|
780 | else: |
---|
781 | y_err[i_bin] += frac * frac * data2D.err_data[j][i] * data2D.err_data[j][i] |
---|
782 | y_counts[i_bin] += frac |
---|
783 | |
---|
784 | for i in range(self.nbins): |
---|
785 | y[i] = y[i] / y_counts[i] |
---|
786 | y_err[i] = math.sqrt(y_err[i]) / y_counts[i] |
---|
787 | # Check which type of averaging we need |
---|
788 | if run.lower()=='phi': |
---|
789 | x[i] = (self.phi_max-self.phi_min)/self.nbins*(1.0*i + 0.5)+self.phi_min |
---|
790 | else: |
---|
791 | x[i] = (self.r_max-self.r_min)/self.nbins*(1.0*i + 0.5)+self.r_min |
---|
792 | |
---|
793 | return Data1D(x=x, y=y, dy=y_err) |
---|
794 | |
---|
795 | class _Sector: |
---|
796 | """ |
---|
797 | Defines a sector region on a 2D data set. |
---|
798 | The sector is defined by r_min, r_max, phi_min, phi_max, |
---|
799 | and the position of the center of the ring |
---|
800 | where phi_min and phi_max are defined by the right and left lines wrt central line |
---|
801 | and phi_max could be less than phi_min. |
---|
802 | |
---|
803 | Phi is defined between 0 and 2pi |
---|
804 | """ |
---|
805 | def __init__(self, r_min, r_max, phi_min, phi_max,nbins=20): |
---|
806 | self.r_min = r_min |
---|
807 | self.r_max = r_max |
---|
808 | self.phi_min = phi_min |
---|
809 | self.phi_max = phi_max |
---|
810 | self.nbins = nbins |
---|
811 | |
---|
812 | def _agv(self, data2D, run='phi'): |
---|
813 | """ |
---|
814 | Perform sector averaging. |
---|
815 | |
---|
816 | @param data2D: Data2D object |
---|
817 | @param run: define the varying parameter ('phi' or 'q') |
---|
818 | @return: Data1D object |
---|
819 | """ |
---|
820 | if data2D.__class__.__name__ not in ["Data2D", "plottable_2D"]: |
---|
821 | raise RuntimeError, "Ring averaging only take plottable_2D objects" |
---|
822 | |
---|
823 | data = data2D.data |
---|
824 | qmax = self.r_max |
---|
825 | qmin = self.r_min |
---|
826 | |
---|
827 | if len(data2D.detector) != 1: |
---|
828 | raise RuntimeError, "Ring averaging: invalid number of detectors: %g" % len(data2D.detector) |
---|
829 | pixel_width_x = data2D.detector[0].pixel_size.x |
---|
830 | pixel_width_y = data2D.detector[0].pixel_size.y |
---|
831 | det_dist = data2D.detector[0].distance |
---|
832 | wavelength = data2D.source.wavelength |
---|
833 | center_x = data2D.detector[0].beam_center.x/pixel_width_x |
---|
834 | center_y = data2D.detector[0].beam_center.y/pixel_width_y |
---|
835 | |
---|
836 | y = numpy.zeros(self.nbins) |
---|
837 | y_counts = numpy.zeros(self.nbins) |
---|
838 | x = numpy.zeros(self.nbins) |
---|
839 | y_err = numpy.zeros(self.nbins) |
---|
840 | |
---|
841 | # This If finds qmax within ROI defined by sector lines |
---|
842 | temp=0 #to find qmax within ROI or phimax and phimin |
---|
843 | temp0=1000000 |
---|
844 | temp1=0 |
---|
845 | for i in range(numpy.size(data,1)): |
---|
846 | dx = pixel_width_x*(i+0.5 - center_x) |
---|
847 | for j in range(numpy.size(data,0)): |
---|
848 | |
---|
849 | dy = pixel_width_y*(j+0.5 - center_y) |
---|
850 | q_value = get_q(dx, dy, det_dist, wavelength) |
---|
851 | # Compute phi and check whether it's within the limits |
---|
852 | phi_value=math.atan2(dy,dx)+math.pi |
---|
853 | if self.phi_max>2*math.pi: |
---|
854 | self.phi_max=self.phi_max-2*math.pi |
---|
855 | if self.phi_min<0: |
---|
856 | self.phi_max=self.phi_max+2*math.pi |
---|
857 | |
---|
858 | #In case of two ROI (symmetric major and minor regions)(for 'q2') |
---|
859 | if run.lower()=='q2': |
---|
860 | if ((self.phi_max>=0 and self.phi_max<math.pi)and (self.phi_min>=0 and self.phi_min<math.pi)): |
---|
861 | temp_max=self.phi_max+math.pi |
---|
862 | temp_min=self.phi_min+math.pi |
---|
863 | else: |
---|
864 | temp_max=self.phi_max |
---|
865 | temp_min=self.phi_min |
---|
866 | |
---|
867 | if ((temp_max>=math.pi and temp_max<2*math.pi)and (temp_min>=math.pi and temp_min<2*math.pi)): |
---|
868 | if (phi_value<temp_min or phi_value>temp_max): |
---|
869 | if (phi_value<temp_min-math.pi or phi_value>temp_max-math.pi): |
---|
870 | continue |
---|
871 | if (self.phi_max<self.phi_min): |
---|
872 | tmp_max=self.phi_max+math.pi |
---|
873 | tmp_min=self.phi_min-math.pi |
---|
874 | else: |
---|
875 | tmp_max=self.phi_max |
---|
876 | tmp_min=self.phi_min |
---|
877 | if (tmp_min<math.pi and tmp_max>math.pi): |
---|
878 | if((phi_value>tmp_max and phi_value<tmp_min+math.pi)or (phi_value>tmp_max-math.pi and phi_value<tmp_min)): |
---|
879 | continue |
---|
880 | #In case of one ROI (major only)(i.e.,for 'q' and 'phi') |
---|
881 | else: |
---|
882 | if (self.phi_max>=self.phi_min): |
---|
883 | if (phi_value<self.phi_min or phi_value>self.phi_max): |
---|
884 | continue |
---|
885 | else: |
---|
886 | if (phi_value<self.phi_min and phi_value>self.phi_max): |
---|
887 | continue |
---|
888 | if q_value<qmin or q_value>qmax: |
---|
889 | continue |
---|
890 | |
---|
891 | if run.lower()=='phi': |
---|
892 | if temp1<phi_value: |
---|
893 | temp1=phi_value |
---|
894 | if temp0>phi_value: |
---|
895 | temp0=phi_value |
---|
896 | |
---|
897 | elif temp<q_value: |
---|
898 | temp=q_value |
---|
899 | |
---|
900 | if run.lower()=='phi': |
---|
901 | self.phi_max=temp1 |
---|
902 | self.phi_min=temp0 |
---|
903 | else: |
---|
904 | qmax=temp |
---|
905 | |
---|
906 | for i in range(numpy.size(data,1)): |
---|
907 | dx = pixel_width_x*(i+0.5 - center_x) |
---|
908 | |
---|
909 | # Min and max x-value for the pixel |
---|
910 | minx = pixel_width_x*(i - center_x) |
---|
911 | maxx = pixel_width_x*(i+1.0 - center_x) |
---|
912 | |
---|
913 | for j in range(numpy.size(data,0)): |
---|
914 | dy = pixel_width_y*(j+0.5 - center_y) |
---|
915 | |
---|
916 | q_value = get_q(dx, dy, det_dist, wavelength) |
---|
917 | |
---|
918 | # Min and max y-value for the pixel |
---|
919 | miny = pixel_width_y*(j - center_y) |
---|
920 | maxy = pixel_width_y*(j+1.0 - center_y) |
---|
921 | |
---|
922 | # Calculate the q-value for each corner |
---|
923 | # q_[x min or max][y min or max] |
---|
924 | q_00 = get_q(minx, miny, det_dist, wavelength) |
---|
925 | q_01 = get_q(minx, maxy, det_dist, wavelength) |
---|
926 | q_10 = get_q(maxx, miny, det_dist, wavelength) |
---|
927 | q_11 = get_q(maxx, maxy, det_dist, wavelength) |
---|
928 | |
---|
929 | # Compute phi and check whether it's within the limits |
---|
930 | phi_value=math.atan2(dy,dx)+math.pi |
---|
931 | if self.phi_max>2*math.pi: |
---|
932 | self.phi_max=self.phi_max-2*math.pi |
---|
933 | if self.phi_min<0: |
---|
934 | self.phi_max=self.phi_max+2*math.pi |
---|
935 | |
---|
936 | # Look for intercept between each side of the pixel |
---|
937 | # and the constant-q ring for qmax |
---|
938 | frac_max = get_pixel_fraction(qmax, q_00, q_01, q_10, q_11) |
---|
939 | |
---|
940 | # Look for intercept between each side of the pixel |
---|
941 | # and the constant-q ring for qmin |
---|
942 | frac_min = get_pixel_fraction(qmin, q_00, q_01, q_10, q_11) |
---|
943 | |
---|
944 | # We are interested in the region between qmin and qmax |
---|
945 | # therefore the fraction of the surface of the pixel |
---|
946 | # that we will use to calculate the number of counts to |
---|
947 | # include is given by: |
---|
948 | |
---|
949 | frac = frac_max - frac_min |
---|
950 | |
---|
951 | #In case of two ROI (symmetric major and minor regions)(for 'q2') |
---|
952 | if run.lower()=='q2': |
---|
953 | if ((self.phi_max>=0 and self.phi_max<math.pi)and (self.phi_min>=0 and self.phi_min<math.pi)): |
---|
954 | temp_max=self.phi_max+math.pi |
---|
955 | temp_min=self.phi_min+math.pi |
---|
956 | else: |
---|
957 | temp_max=self.phi_max |
---|
958 | temp_min=self.phi_min |
---|
959 | |
---|
960 | if ((temp_max>=math.pi and temp_max<2*math.pi)and (temp_min>=math.pi and temp_min<2*math.pi)): |
---|
961 | if (phi_value<temp_min or phi_value>temp_max): |
---|
962 | if (phi_value<temp_min-math.pi or phi_value>temp_max-math.pi): |
---|
963 | continue |
---|
964 | if (self.phi_max<self.phi_min): |
---|
965 | tmp_max=self.phi_max+math.pi |
---|
966 | tmp_min=self.phi_min-math.pi |
---|
967 | else: |
---|
968 | tmp_max=self.phi_max |
---|
969 | tmp_min=self.phi_min |
---|
970 | if (tmp_min<math.pi and tmp_max>math.pi): |
---|
971 | if((phi_value>tmp_max and phi_value<tmp_min+math.pi)or (phi_value>tmp_max-math.pi and phi_value<tmp_min)): |
---|
972 | continue |
---|
973 | #In case of one ROI (major only)(i.e.,for 'q' and 'phi') |
---|
974 | else: |
---|
975 | if (self.phi_max>=self.phi_min): |
---|
976 | if (phi_value<self.phi_min or phi_value>self.phi_max): |
---|
977 | continue |
---|
978 | |
---|
979 | else: |
---|
980 | if (phi_value<self.phi_min and phi_value>self.phi_max): |
---|
981 | continue |
---|
982 | #print "qmax=",qmax,qmin |
---|
983 | |
---|
984 | if q_value<qmin or q_value>qmax: |
---|
985 | continue |
---|
986 | |
---|
987 | # Check which type of averaging we need |
---|
988 | if run.lower()=='phi': |
---|
989 | i_bin = int(math.ceil(self.nbins*(phi_value-self.phi_min)/(self.phi_max-self.phi_min))) - 1 |
---|
990 | else: |
---|
991 | # If we don't need this pixel, skip the rest of the work |
---|
992 | #TODO: an improvement here would be to compute the average |
---|
993 | # Q for the pixel from the part that is covered by |
---|
994 | # the ring defined by q_min/q_max rather than the complete |
---|
995 | # pixel |
---|
996 | i_bin = int(math.ceil(self.nbins*(q_value-qmin)/(qmax-qmin))) - 1 |
---|
997 | |
---|
998 | try: |
---|
999 | y[i_bin] += frac * data[j][i] |
---|
1000 | except: |
---|
1001 | import sys |
---|
1002 | print sys.exc_value |
---|
1003 | print i_bin, frac |
---|
1004 | |
---|
1005 | if data2D.err_data == None or data2D.err_data[j][i]==0.0: |
---|
1006 | y_err[i_bin] += frac * frac * math.fabs(data2D.data[j][i]) |
---|
1007 | else: |
---|
1008 | y_err[i_bin] += frac * frac * data2D.err_data[j][i] * data2D.err_data[j][i] |
---|
1009 | y_counts[i_bin] += frac |
---|
1010 | |
---|
1011 | for i in range(self.nbins): |
---|
1012 | y[i] = y[i] / y_counts[i] |
---|
1013 | y_err[i] = math.sqrt(y_err[i]) / y_counts[i] |
---|
1014 | # Check which type of averaging we need |
---|
1015 | if run.lower()=='phi': |
---|
1016 | #Calculate x[i] and put back the origin of angle back to the right hand side (from the left). |
---|
1017 | x[i] = ((self.phi_max-self.phi_min)/self.nbins*(1.0*i + 0.5)+self.phi_min-2*math.pi)*180/math.pi |
---|
1018 | if x[i]<0: |
---|
1019 | x[i]=180+x[i] |
---|
1020 | else: |
---|
1021 | x[i] = (qmax-qmin)/self.nbins*(1.0*i + 0.5)+qmin |
---|
1022 | |
---|
1023 | return Data1D(x=x, y=y, dy=y_err) |
---|
1024 | |
---|
1025 | class SectorPhi(_Sector): |
---|
1026 | """ |
---|
1027 | Sector average as a function of phi. |
---|
1028 | I(phi) is return and the data is averaged over Q. |
---|
1029 | |
---|
1030 | A sector is defined by r_min, r_max, phi_min, phi_max. |
---|
1031 | The number of bin in phi also has to be defined. |
---|
1032 | """ |
---|
1033 | def __call__(self, data2D): |
---|
1034 | """ |
---|
1035 | Perform sector average and return I(phi). |
---|
1036 | |
---|
1037 | @param data2D: Data2D object |
---|
1038 | @return: Data1D object |
---|
1039 | """ |
---|
1040 | return self._agv(data2D, 'phi') |
---|
1041 | |
---|
1042 | class SectorQold(_Sector): |
---|
1043 | """ |
---|
1044 | Sector average as a function of Q. |
---|
1045 | I(Q) is return and the data is averaged over phi. |
---|
1046 | |
---|
1047 | A sector is defined by r_min, r_max, phi_min, phi_max. |
---|
1048 | The number of bin in Q also has to be defined. |
---|
1049 | """ |
---|
1050 | def __call__(self, data2D): |
---|
1051 | """ |
---|
1052 | Perform sector average and return I(Q). |
---|
1053 | |
---|
1054 | @param data2D: Data2D object |
---|
1055 | @return: Data1D object |
---|
1056 | """ |
---|
1057 | return self._agv(data2D, 'q') |
---|
1058 | |
---|
1059 | class SectorQ(_Sector): |
---|
1060 | """ |
---|
1061 | Sector average as a function of Q for both symatric wings. |
---|
1062 | I(Q) is return and the data is averaged over phi. |
---|
1063 | |
---|
1064 | A sector is defined by r_min, r_max, phi_min, phi_max. |
---|
1065 | r_min, r_max, phi_min, phi_max >0. |
---|
1066 | The number of bin in Q also has to be defined. |
---|
1067 | """ |
---|
1068 | def __call__(self, data2D): |
---|
1069 | """ |
---|
1070 | Perform sector average and return I(Q). |
---|
1071 | |
---|
1072 | @param data2D: Data2D object |
---|
1073 | @return: Data1D object |
---|
1074 | """ |
---|
1075 | return self._agv(data2D, 'q2') |
---|
1076 | if __name__ == "__main__": |
---|
1077 | |
---|
1078 | from loader import Loader |
---|
1079 | |
---|
1080 | |
---|
1081 | d = Loader().load('test/MAR07232_rest.ASC') |
---|
1082 | #d = Loader().load('test/MP_New.sans') |
---|
1083 | |
---|
1084 | |
---|
1085 | r = SectorQ(r_min=.000001, r_max=.01, phi_min=0.0, phi_max=2*math.pi) |
---|
1086 | o = r(d) |
---|
1087 | |
---|
1088 | s = Ring(r_min=.000001, r_max=.01) |
---|
1089 | p = s(d) |
---|
1090 | |
---|
1091 | for i in range(len(o.x)): |
---|
1092 | print o.x[i], o.y[i], o.dy[i] |
---|
1093 | |
---|
1094 | |
---|
1095 | |
---|