1 | import math |
---|
2 | import logging |
---|
3 | import numpy as np |
---|
4 | |
---|
5 | from sas.qtgui.Plotting.PlotterData import Data1D |
---|
6 | |
---|
7 | PR_FIT_LABEL = r"$P_{fit}(r)$" |
---|
8 | PR_LOADED_LABEL = r"$P_{loaded}(r)$" |
---|
9 | IQ_DATA_LABEL = r"$I_{obs}(q)$" |
---|
10 | IQ_FIT_LABEL = r"$I_{fit}(q)$" |
---|
11 | IQ_SMEARED_LABEL = r"$I_{smeared}(q)$" |
---|
12 | GROUP_ID_IQ_DATA = r"$I_{obs}(q)$" |
---|
13 | GROUP_ID_PR_FIT = r"$P_{fit}(r)$" |
---|
14 | PR_PLOT_PTS = 51 |
---|
15 | |
---|
16 | logger = logging.getLogger(__name__) |
---|
17 | |
---|
18 | |
---|
19 | class InversionLogic(object): |
---|
20 | """ |
---|
21 | All the data-related logic. This class deals exclusively with Data1D/2D |
---|
22 | No QStandardModelIndex here. |
---|
23 | """ |
---|
24 | |
---|
25 | def __init__(self, data=None): |
---|
26 | self._data = data |
---|
27 | self.data_is_loaded = False |
---|
28 | if data is not None: |
---|
29 | self.data_is_loaded = True |
---|
30 | |
---|
31 | @property |
---|
32 | def data(self): |
---|
33 | return self._data |
---|
34 | |
---|
35 | @data.setter |
---|
36 | def data(self, value): |
---|
37 | """ data setter """ |
---|
38 | self._data = value |
---|
39 | self.data_is_loaded = (self._data is not None) |
---|
40 | |
---|
41 | def isLoadedData(self): |
---|
42 | """ accessor """ |
---|
43 | return self.data_is_loaded |
---|
44 | |
---|
45 | def new1DPlot(self, out, pr, q=None): |
---|
46 | """ |
---|
47 | Create a new 1D data instance based on fitting results |
---|
48 | """ |
---|
49 | |
---|
50 | qtemp = pr.x |
---|
51 | if q is not None: |
---|
52 | qtemp = q |
---|
53 | |
---|
54 | # Make a plot |
---|
55 | maxq = max(qtemp) |
---|
56 | |
---|
57 | minq = min(qtemp) |
---|
58 | |
---|
59 | # Check for user min/max |
---|
60 | if pr.q_min is not None and maxq >= pr.q_min >= minq: |
---|
61 | minq = pr.q_min |
---|
62 | if pr.q_max is not None and maxq >= pr.q_max >= minq: |
---|
63 | maxq = pr.q_max |
---|
64 | |
---|
65 | x = np.arange(minq, maxq, maxq / 301.0) |
---|
66 | y = np.zeros(len(x)) |
---|
67 | err = np.zeros(len(x)) |
---|
68 | for i in range(len(x)): |
---|
69 | value = pr.iq(out, x[i]) |
---|
70 | y[i] = value |
---|
71 | try: |
---|
72 | err[i] = math.sqrt(math.fabs(value)) |
---|
73 | except: |
---|
74 | err[i] = 1.0 |
---|
75 | logger.log(("Error getting error", value, x[i])) |
---|
76 | |
---|
77 | new_plot = Data1D(x, y) |
---|
78 | new_plot.name = IQ_FIT_LABEL |
---|
79 | new_plot.xaxis("\\rm{Q}", 'A^{-1}') |
---|
80 | new_plot.yaxis("\\rm{Intensity} ", "cm^{-1}") |
---|
81 | title = "I(q)" |
---|
82 | new_plot.title = title |
---|
83 | |
---|
84 | # If we have a group ID, use it |
---|
85 | if 'plot_group_id' in pr.info: |
---|
86 | new_plot.group_id = pr.info["plot_group_id"] |
---|
87 | new_plot.id = IQ_FIT_LABEL |
---|
88 | |
---|
89 | # If we have used slit smearing, plot the smeared I(q) too |
---|
90 | if pr.slit_width > 0 or pr.slit_height > 0: |
---|
91 | x = np.arange(minq, maxq, maxq / 301.0) |
---|
92 | y = np.zeros(len(x)) |
---|
93 | err = np.zeros(len(x)) |
---|
94 | for i in range(len(x)): |
---|
95 | value = pr.iq_smeared(pr.out, x[i]) |
---|
96 | y[i] = value |
---|
97 | try: |
---|
98 | err[i] = math.sqrt(math.fabs(value)) |
---|
99 | except: |
---|
100 | err[i] = 1.0 |
---|
101 | logger.log(("Error getting error", value, x[i])) |
---|
102 | |
---|
103 | new_plot = Data1D(x, y) |
---|
104 | new_plot.name = IQ_SMEARED_LABEL |
---|
105 | new_plot.xaxis("\\rm{Q}", 'A^{-1}') |
---|
106 | new_plot.yaxis("\\rm{Intensity} ", "cm^{-1}") |
---|
107 | # If we have a group ID, use it |
---|
108 | if 'plot_group_id' in pr.info: |
---|
109 | new_plot.group_id = pr.info["plot_group_id"] |
---|
110 | new_plot.id = IQ_SMEARED_LABEL |
---|
111 | new_plot.title = title |
---|
112 | |
---|
113 | new_plot.symbol = 'Line' |
---|
114 | new_plot.hide_error = True |
---|
115 | |
---|
116 | return new_plot |
---|
117 | |
---|
118 | def newPRPlot(self, out, pr, cov=None): |
---|
119 | """ |
---|
120 | """ |
---|
121 | # Show P(r) |
---|
122 | x = np.arange(0.0, pr.d_max, pr.d_max / PR_PLOT_PTS) |
---|
123 | |
---|
124 | y = np.zeros(len(x)) |
---|
125 | dy = np.zeros(len(x)) |
---|
126 | |
---|
127 | total = 0.0 |
---|
128 | pmax = 0.0 |
---|
129 | cov2 = np.ascontiguousarray(cov) |
---|
130 | |
---|
131 | for i in range(len(x)): |
---|
132 | if cov2 is None: |
---|
133 | value = pr.pr(out, x[i]) |
---|
134 | else: |
---|
135 | (value, dy[i]) = pr.pr_err(out, cov2, x[i]) |
---|
136 | total += value * pr.d_max / len(x) |
---|
137 | |
---|
138 | # keep track of the maximum P(r) value |
---|
139 | if value > pmax: |
---|
140 | pmax = value |
---|
141 | |
---|
142 | y[i] = value |
---|
143 | |
---|
144 | if cov2 is None: |
---|
145 | new_plot = Data1D(x, y) |
---|
146 | else: |
---|
147 | new_plot = Data1D(x, y, dy=dy) |
---|
148 | new_plot.name = PR_FIT_LABEL |
---|
149 | new_plot.xaxis("\\rm{r}", 'A') |
---|
150 | new_plot.yaxis("\\rm{P(r)} ", "cm^{-3}") |
---|
151 | new_plot.title = "P(r) fit" |
---|
152 | new_plot.id = PR_FIT_LABEL |
---|
153 | new_plot.scale = "linear" |
---|
154 | new_plot.group_id = GROUP_ID_PR_FIT |
---|
155 | |
---|
156 | return new_plot |
---|
157 | |
---|
158 | def computeDataRange(self): |
---|
159 | """ |
---|
160 | Wrapper for calculating the data range based on local dataset |
---|
161 | """ |
---|
162 | return self.computeRangeFromData(self.data) |
---|
163 | |
---|
164 | def computeRangeFromData(self, data): |
---|
165 | """ |
---|
166 | Compute the minimum and the maximum range of the data |
---|
167 | return the npts contains in data |
---|
168 | """ |
---|
169 | qmin, qmax = None, None |
---|
170 | if isinstance(data, Data1D): |
---|
171 | try: |
---|
172 | qmin = min(data.x) |
---|
173 | qmax = max(data.x) |
---|
174 | except (ValueError, TypeError): |
---|
175 | msg = "Unable to find min/max/length of \n data named %s" % \ |
---|
176 | self.data.filename |
---|
177 | raise ValueError(msg) |
---|
178 | |
---|
179 | else: |
---|
180 | qmin = 0 |
---|
181 | try: |
---|
182 | x = max(np.fabs(data.xmin), np.fabs(data.xmax)) |
---|
183 | y = max(np.fabs(data.ymin), np.fabs(data.ymax)) |
---|
184 | except (ValueError, TypeError): |
---|
185 | msg = "Unable to find min/max of \n data named %s" % \ |
---|
186 | self.data.filename |
---|
187 | raise ValueError(msg) |
---|
188 | qmax = np.sqrt(x * x + y * y) |
---|
189 | return qmin, qmax |
---|