source: sasview/DataLoader/qsmearing.py @ ea290ee

ESS_GUIESS_GUI_DocsESS_GUI_batch_fittingESS_GUI_bumps_abstractionESS_GUI_iss1116ESS_GUI_iss879ESS_GUI_iss959ESS_GUI_openclESS_GUI_orderingESS_GUI_sync_sascalccostrafo411magnetic_scattrelease-4.1.1release-4.1.2release-4.2.2release_4.0.1ticket-1009ticket-1094-headlessticket-1242-2d-resolutionticket-1243ticket-1249ticket885unittest-saveload
Last change on this file since ea290ee was c0d9981, checked in by Mathieu Doucet <doucetm@…>, 16 years ago

preparing for release

  • Property mode set to 100644
File size: 10.8 KB
RevLine 
[d00f8ff]1"""
2This software was developed by the University of Tennessee as part of the
3Distributed Data Analysis of Neutron Scattering Experiments (DANSE)
4project funded by the US National Science Foundation.
5
6See the license text in license.txt
7
8copyright 2008, University of Tennessee
9"""
[4fe4394]10
11
[d00f8ff]12import numpy
13import math
14import scipy.special
15
16def smear_selection(data1D):
17    """
18        Creates the right type of smearer according
[4fe4394]19        to the data.
20   
21        The canSAS format has a rule that either
22        slit smearing data OR resolution smearing data
23        is available.
24       
25        For the present purpose, we choose the one that
26        has none-zero data. If both slit and resolution
27        smearing arrays are filled with good data
28        (which should not happen), then we choose the
29        resolution smearing data.
30       
31        @param data1D: Data1D object
[d00f8ff]32    """
[4fe4394]33    # Sanity check. If we are not dealing with a SANS Data1D
34    # object, just return None
35    if data1D.__class__.__name__ != 'Data1D' \
36        or not hasattr(data1D, "dxl") or not hasattr(data1D, "dxw"):
37        return None
38   
39    # Look for resolution smearing data
40    _found_resolution = False
41    if data1D.dx is not None and len(data1D.dx)==len(data1D.x):
42       
43        # Check that we have non-zero data
44        if data1D.dx[0]>0.0:
45            _found_resolution = True
46       
47    # If we found resolution smearing data, return a QSmearer
48    if _found_resolution == True:
49        return QSmearer(data1D)
50
51    # Look for slit smearing data
52    _found_slit = False
53    if data1D.dxl is not None and len(data1D.dxl)==len(data1D.x) \
54        and data1D.dxw is not None and len(data1D.dxw)==len(data1D.x):
55       
56        # Check that we have non-zero data
57        if data1D.dxl[0]>0.0 or data1D.dxw[0]>0.0:
58            _found_slit = True
59       
60        # Sanity check: all data should be the same as a function of Q
61        for item in data1D.dxl:
62            if data1D.dxl[0] != item:
63                _found_resolution = False
64                break
65           
66        for item in data1D.dxw:
67            if data1D.dxw[0] != item:
68                _found_resolution = False
69                break
70           
71    # If we found slit smearing data, return a slit smearer
72    if _found_slit == True:
73        return SlitSmearer(data1D)
74   
75    return None
76           
[d00f8ff]77
78class _BaseSmearer(object):
79   
80    def __init__(self):
81        self.nbins = 0
82        self._weights = None
83       
84    def _compute_matrix(self): return NotImplemented
85
86    def __call__(self, iq):
87        """
88            Return the smeared I(q) value at the given q.
89            The smeared I(q) is computed using a predetermined
90            smearing matrix for a particular binning.
91       
92            @param q: I(q) array
93            @return: smeared I(q)
94        """
95        # Sanity check
96        if len(iq) != self.nbins:
97            raise RuntimeError, "Invalid I(q) vector: inconsistent array length %s <> %s" % (len(iq), self.nbins)
98       
99        if self._weights == None:
100            self._compute_matrix()
101           
102        iq_smeared = numpy.zeros(self.nbins)
103        # Loop over q-values
104        for q_i in range(self.nbins):
105            sum = 0.0
106            counts = 0.0   
107           
108            for i in range(self.nbins):
109                sum += iq[i] * self._weights[q_i][i]
110                counts += self._weights[q_i][i]
111               
112            iq_smeared[q_i] = sum/counts
113               
114        return iq_smeared   
115   
116class _SlitSmearer(_BaseSmearer):
117    """
118        Slit smearing for I(q) array
119    """
120   
121    def __init__(self, nbins=None, width=None, height=None, min=None, max=None):
122        """
123            Initialization
124           
125            @param iq: I(q) array [cm-1]
126            @param width: slit width [A-1]
127            @param height: slit height [A-1]
128            @param min: Q_min [A-1]
129            @param max: Q_max [A-1]
130        """
131        ## Slit width in Q units
132        self.width  = width
133        ## Slit height in Q units
134        self.height = height
135        ## Q_min (Min Q-value for I(q))
136        self.min    = min
137        ## Q_max (Max Q_value for I(q))
138        self.max    = max
139        ## Number of Q bins
140        self.nbins  = nbins
141        ## Number of points used in the smearing computation
142        self.npts   = 1000
143        ## Smearing matrix
144        self._weights = None
145       
146    def _compute_matrix(self):
147        """
148            Compute the smearing matrix for the current I(q) array
149        """
150        weights = numpy.zeros([self.nbins, self.nbins])
151       
152        # Loop over all q-values
153        for i in range(self.nbins):
154            q = self.min + i*(self.max-self.min)/float(self.nbins-1.0)
155           
156            # For each q-value, compute the weight of each other q-bin
157            # in the I(q) array
158            npts_h = self.npts if self.height>0 else 1
159            npts_w = self.npts if self.width>0 else 1
160           
161            # If both height and width are great than zero,
162            # modify the number of points in each direction so
163            # that the total number of points is still what
164            # the user would expect (downgrade resolution)
165            if npts_h>1 and npts_w>1:
166                npts_h = int(math.ceil(math.sqrt(self.npts)))
167                npts_w = npts_h
168               
169            for k in range(npts_h):
170                if npts_h==1:
171                    shift_h = 0
172                else:
173                    shift_h = self.height/(float(npts_h-1.0)) * k
174                for j in range(npts_w):
175                    if npts_w==1:
176                        shift_w = 0
177                    else:
178                        shift_w = self.width/(float(npts_w-1.0)) * j
179                    q_shifted = math.sqrt( ((q-shift_w)*(q-shift_w) + shift_h*shift_h) )
180                    q_i = int(math.floor( (q_shifted-self.min)/((self.max-self.min)/(self.nbins -1.0)) ))
181                   
182                    # Skip the entries outside our I(q) range
183                    #TODO: be careful with edge effect
184                    if q_i<self.nbins:
185                        weights[i][q_i] = weights[i][q_i]+1.0
186                               
187        self._weights = weights
188        return self._weights
189
190class SlitSmearer(_SlitSmearer):
191    """
192        Adaptor for slit smearing class and SANS data
193    """
194    def __init__(self, data1D):
195        """
196            Assumption: equally spaced bins of increasing q-values.
197           
198            @param data1D: data used to set the smearing parameters
199        """
200        # Initialization from parent class
201        super(SlitSmearer, self).__init__()
202       
203        ## Slit width
204        self.width = 0
205        if data1D.dxw is not None and len(data1D.dxw)==len(data1D.x):
206            self.width = data1D.dxw[0]
207            # Sanity check
208            for value in data1D.dxw:
209                if value != self.width:
210                    raise RuntimeError, "Slit smearing parameters must be the same for all data"
211               
212        ## Slit height
213        self.height = 0
214        if data1D.dxl is not None and len(data1D.dxl)==len(data1D.x):
215            self.height = data1D.dxl[0]
216            # Sanity check
217            for value in data1D.dxl:
218                if value != self.height:
219                    raise RuntimeError, "Slit smearing parameters must be the same for all data"
220       
221        ## Number of Q bins
222        self.nbins = len(data1D.x)
223        ## Minimum Q
224        self.min = data1D.x[0]
225        ## Maximum
226        self.max = data1D.x[len(data1D.x)-1]       
227
228
229class _QSmearer(_BaseSmearer):
230    """
231        Perform Gaussian Q smearing
232    """
233       
234    def __init__(self, nbins=None, width=None, min=None, max=None):
235        """
236            Initialization
237           
238            @param nbins: number of Q bins
[c0d9981]239            @param width: array standard deviation in Q [A-1]
[d00f8ff]240            @param min: Q_min [A-1]
241            @param max: Q_max [A-1]
242        """
243        ## Standard deviation in Q [A-1]
244        self.width  = width
245        ## Q_min (Min Q-value for I(q))
246        self.min    = min
247        ## Q_max (Max Q_value for I(q))
248        self.max    = max
249        ## Number of Q bins
250        self.nbins  = nbins
251        ## Smearing matrix
252        self._weights = None
253       
254    def _compute_matrix(self):
255        """
256            Compute the smearing matrix for the current I(q) array
257        """
258        weights = numpy.zeros([self.nbins, self.nbins])
259       
260        # Loop over all q-values
261        step = (self.max-self.min)/float(self.nbins-1.0)
262        for i in range(self.nbins):
263            q = self.min + i*step
264            q_min = q - 0.5*step
265            q_max = q + 0.5*step
266           
267            for j in range(self.nbins):
268                q_j = self.min + j*step
269               
270                # Compute the fraction of the Gaussian contributing
271                # to the q bin between q_min and q_max
[4fe4394]272                value =  scipy.special.erf( (q_max-q_j)/(math.sqrt(2.0)*self.width[j]) ) 
273                value -=scipy.special.erf( (q_min-q_j)/(math.sqrt(2.0)*self.width[j]) ) 
[d00f8ff]274
275                weights[i][j] += value
276                               
277        self._weights = weights
278        return self._weights
279       
280class QSmearer(_QSmearer):
281    """
282        Adaptor for Gaussian Q smearing class and SANS data
283    """
284    def __init__(self, data1D):
285        """
286            Assumption: equally spaced bins of increasing q-values.
287           
288            @param data1D: data used to set the smearing parameters
289        """
290        # Initialization from parent class
291        super(QSmearer, self).__init__()
292       
[c0d9981]293        ## Resolution
[4fe4394]294        self.width = numpy.zeros(len(data1D.x))
[d00f8ff]295        if data1D.dx is not None and len(data1D.dx)==len(data1D.x):
[4fe4394]296            self.width = data1D.dx
[d00f8ff]297       
298        ## Number of Q bins
299        self.nbins = len(data1D.x)
300        ## Minimum Q
301        self.min = data1D.x[0]
302        ## Maximum
303        self.max = data1D.x[len(data1D.x)-1]       
304
305
306if __name__ == '__main__':
307    x = 0.001*numpy.arange(1,11)
308    y = 12.0-numpy.arange(1,11)
309    print x
310    #for i in range(10): print i, 0.001 + i*0.008/9.0
311    #for i in range(100): print i, int(math.floor( (i/ (100/9.0)) ))
312
313   
314    s = _SlitSmearer(nbins=10, width=0.0, height=0.005, min=0.001, max=0.010)
315    #s = _QSmearer(nbins=10, width=0.001, min=0.001, max=0.010)
316    s._compute_matrix()
317
318    sy = s(y)
319    print sy
320   
321    if True:
322        for i in range(10):
323            print x[i], sy[i]
324            #print q, ' : ', s.weight(q), s._compute_iq(q)
325            #print q, ' : ', s(q), s._compute_iq(q)
326            #s._compute_iq(q)
327
328
329
330
Note: See TracBrowser for help on using the repository browser.