source: fact/tools/pyscripts/sandbox/dneise/gapd_template_fit/gen_TH1D.py@ 14473

Last change on this file since 14473 was 14253, checked in by neise, 12 years ago
the fit is nice3
  • Property svn:executable set to *
File size: 13.9 KB
Line 
1#!/usr/bin/python -tt
2#
3# Dominik Neise, Werner Lustermann
4# TU Dortmund, ETH Zurich
5#
6import numpy as np
7import sys
8from ROOT import *
9from scipy import interpolate
10
11from plotters import Plotter
12import matplotlib.pyplot as plt
13
14class SignalGenerator(object):
15 """ Signal Generator
16 generates signals for testing several helper classes like:
17 * fir filters
18 * signal extractors
19 """
20
21 def __init__(self, option_str = 'len 100 noise 3', name = 'SignalGenerator'):
22 """ initialize the generator
23 sets default signal to generate
24 """
25 self.__module__ = 'generator'
26 self.option_str = option_str.lower()
27 self.options = make_options_from_str(option_str)
28 self.parse_options()
29 self.name = name
30
31 def parse_options(self):
32 o = self.options #shortcut
33 if 'len' in o:
34 self.npoints = int(o['len'][0])
35 else:
36 self.npoints = 100
37 if 'noise' in o:
38 self.sigma = float(o['noise'][0])
39 else:
40 self.sigma = 1
41 if 'bsl' in o:
42 self.bsl = float(o['bsl'][0])
43 else:
44 self.bsl = -0.5
45
46 if 'step' in o:
47 self.step_height = float(o['step'][0])
48 self.step_start = int(o['step'][1])
49 self.step_stop = int(o['step'][2])
50
51 if 'triangle' in o:
52 self.pulses = []
53 # append 1st pulse to list of pulses
54 self.pulses.append( ( float(o['triangle'][0]) , float(o['triangle'][1]), int(o['triangle'][2]), int(o['triangle'][3]) ) )
55 number_of_pulses_after_1st = (len(o['triangle'])-4)/2
56 for i in range(number_of_pulses_after_1st):
57 self.pulses.append( ( float(o['triangle'][2*i+4]) , float(o['triangle'][2*i+5]), int(o['triangle'][2]), int(o['triangle'][3]) ) )
58
59 if 'spike' in o:
60 self.spikes = []
61 for i in range(len(o['spike'])/2):
62 self.spikes.append( ( int(o['spike'][2*i]), float(o['spike'][2*i+1]) ) )
63
64 def __call__(self, option_str = ''):
65 if option_str:
66 self.option_str = option_str.lower()
67 self.options = make_options_from_str(self.option_str)
68 self.parse_options()
69
70 signal = np.zeros(self.npoints)
71 signal += self.bsl
72 signal += np.random.randn(self.npoints) * self.sigma
73 if 'step' in self.options:
74 signal[self.step_start:self.step_stop] += self.step_height
75 if 'triangle' in self.options:
76 for pulse in self.pulses:
77 pos = pulse[0]
78 height = pulse[1]
79 rise = pulse[2]
80 fall = pulse[3]
81 start = pos - rise
82 stop = pos + fall
83 signal[start:pos] += np.linspace(0., height, rise)
84 signal[pos:stop] += np.linspace(height, 0. , fall)
85 if 'spike' in self.options:
86 for spike in self.spikes:
87 signal[spike[0]] += spike[1]
88 return signal
89
90 def __str__(self):
91 s = self.name + '\n'
92 s += 'possible options and parameters\n'
93 s += ' * len: number of samples (100)\n'
94 s += ' * noise: sigma (1)\n'
95 s += ' * bsl: level (-0.5)\n'
96 s += ' * step: height, start, end\n'
97 s += ' * triangle: pos height risingedge, fallingedge [pos height ...]\n'
98 s += ' * spike: pos height [pos height ...]\n'
99
100 s += 'current options are:\n'
101 for key in self.options.keys():
102 s += key + ':' + str(self.options[key]) + '\n'
103 return s
104
105
106class SignalGeneratorCSV(object):
107
108 def __init__(self, file_name, option_str = 'len 100 noise 3', name = 'SignalGenerator'):
109 time, maxprob, mean, median = np.loadtxt( file_name, delimiter=',', unpack=True)
110 csv_data = median
111
112 self.csv_maxprob = maxprob
113 self.csv_mean = mean
114 self.csv_median = median
115
116 # shift the input data such, that the left baseline is equal to zero
117 csv_data = csv_data - csv_data[:50].mean()
118
119 # I want the data to be longer than they are, so I add 1000 slices left and right
120 self.csv_data = np.zeros(2000 + len(csv_data))
121
122 for i in range(len(csv_data)):
123 self.csv_data[1000+i] = csv_data[i]
124 for i in range(1000):
125 self.csv_data[i] = csv_data[:50].mean()
126 self.csv_data[-1*i] = csv_data[-50:].mean()
127
128
129 x = np.arange(0,len(self.csv_data),1)
130 x2 = np.arange(0,len(self.csv_data),0.1)
131
132 # fermi dirac distribution for weighting the input data.
133 # we do not believe the right tail is really higher than zero, so we make it go to zero.
134 weight_function_right = lambda x : 1. / (1 + np.exp((x - 1180.)/100.))
135 weight_function_left = lambda x : 1. / (1 + np.exp((-1*(x - 1045.))/5.))
136
137 weights = np.array(map(weight_function_right, x))
138 weights *= np.array(map(weight_function_left, x))
139 weights *= 8
140 weights [1200:] += 3
141 weights += 2
142 #y_weighted = weights * y
143
144 #print len(weights) == len(x)
145 #print len(weights) == len(self.csv_data)
146 #print weights > 0
147
148 tck= interpolate.splrep(x, self.csv_data, w=weights)
149 self.spline_int = tck
150
151
152 x2 = np.arange(1000,1500,1)
153 t,c,k = tck
154 y2 = interpolate.splev(x2,(t,c,k),der=0)
155
156
157 #tck = interpolate.splrep(x, self.csv_data ,w=weights, s=50)
158 #xnew = np.arange(0,len(self.csv_data), 0.1)
159 #y = interpolate.splev(x,tck,der=0)
160 #y2 = interpolate.splev(x2,tck,der=0)
161
162
163 #plt.figure()
164 #plt.plot(x,self.csv_data,'.:b',x,weights,'m',x2,y2,'.:r')
165 #plt.show()
166 #raw_input('stop')
167
168 # csv data was downshifted, I shift it up here
169 self.csv_data = csv_data - csv_data.min()
170
171
172 self.__module__ = 'CSV generator'
173 self.option_str = option_str.lower()
174 self.options = make_options_from_str(option_str)
175 self.parse_options()
176 self.name = name
177
178 def parse_options(self):
179 o = self.options #shortcut
180 if 'len' in o:
181 self.npoints = int(o['len'][0])
182 else:
183 self.npoints = 1024
184 if 'noise' in o:
185 self.sigma = float(o['noise'][0])
186 else:
187 self.sigma = 1
188 if 'bsl' in o:
189 self.bsl = float(o['bsl'][0])
190 else:
191 self.bsl = -0.5
192
193 if 'step' in o:
194 self.step_height = float(o['step'][0])
195 self.step_start = int(o['step'][1])
196 self.step_stop = int(o['step'][2])
197
198 if 'triangle' in o:
199 self.pulses = []
200 # append 1st pulse to list of pulses
201 self.pulses.append( ( float(o['triangle'][0]) , float(o['triangle'][1]), int(o['triangle'][2]), int(o['triangle'][3]) ) )
202 number_of_pulses_after_1st = (len(o['triangle'])-4)/2
203 for i in range(number_of_pulses_after_1st):
204 self.pulses.append( ( float(o['triangle'][2*i+4]) , float(o['triangle'][2*i+5]), int(o['triangle'][2]), int(o['triangle'][3]) ) )
205
206 if 'spike' in o:
207 self.spikes = []
208 for i in range(len(o['spike'])/2):
209 self.spikes.append( ( int(o['spike'][2*i]), float(o['spike'][2*i+1]) ) )
210
211 if 'csv' in o:
212 self.csvs = []
213 for i in range(len(o['csv'])/2):
214 time = int( o['csv'][2*i] )
215 amplitude = float( o['csv'][2*i+1] )
216 self.csvs.append( (time, amplitude) )
217 if 'rate' in o:
218 self.rate = float(o['rate'][0])
219
220 if 'signal' in o:
221 self.amplitude = float(o['signal'][0])
222 self.position = int(o['signal'][1])
223
224 def __call__(self, option_str = ''):
225 if option_str:
226 self.option_str = option_str.lower()
227 self.options = make_options_from_str(self.option_str)
228 self.parse_options()
229
230 signal = np.zeros(self.npoints)
231 signal += self.bsl
232
233 # shortcut
234 o = self.options
235
236 if 'step' in o:
237 signal[self.step_start:self.step_stop] += self.step_height
238 if 'triangle' in o:
239 for pulse in o:
240 pos = pulse[0]
241 height = pulse[1]
242 rise = pulse[2]
243 fall = pulse[3]
244 start = pos - rise
245 stop = pos + fall
246 signal[start:pos] += np.linspace(0., height, rise)
247 signal[pos:stop] += np.linspace(height, 0. , fall)
248 if 'spike' in o:
249 for spike in self.spikes:
250 signal[spike[0]] += spike[1]
251
252 if 'csv' in o:
253 for csv in self.csvs:
254 amplitude = csv[1]
255 time = csv[0]
256
257 #csv_data = self.csv_data.copy()
258 #scale
259 #csv_data *= amplitude
260
261 x = np.arange(0,300,1)
262 t,c,k = self.spline_int
263 d = c.copy()
264 d *= amplitude
265 csv_data = interpolate.splev(x,(t,d,k),der=0)
266
267 # add shifted
268 signal[time:time+len(csv_data)] += csv_data
269
270 if 'rate' in o:
271 self._add_template_pulses(signal, self.rate)
272
273 if 'signal' in o:
274 self._add_signal(signal)
275
276 # add noise
277 if 'noise' in o:
278 signal += + np.random.normal(0.0,self.sigma, signal.shape)
279
280 return signal
281
282 def _add_signal(self, signal):
283 a = self.amplitude
284 p = self.position
285
286 #csv_data = self.csv_data
287 #csv = csv_data.copy() * a
288
289 x = np.arange(1000,1500,1)
290 t,c,k = self.spline_int
291 d = c.copy()
292 d *= a
293 csv = interpolate.splev(x,(t,d,k),der=0)
294
295 npoints = self.npoints + 2 * len(csv)
296 internal_signal = np.zeros(npoints)
297
298 internal_signal[p-70+len(csv):p+2*len(csv)-70] += csv
299
300 signal += internal_signal[len(csv):-len(csv)]
301
302 def _add_template_pulses(self, signal, rate):
303
304 #csv_data = self.csv_data
305 x = np.arange(1000,1500,1)
306 t,c,k = self.spline_int
307 d = c.copy()
308 csv_data = interpolate.splev(x,(t,d,k),der=0)
309
310
311 period = 2.*1e9/(float(rate)*1e6) # in slices
312 # in order to simulate pulses which are just at the beginning and
313 # at the end of the signal, we pre- and append some points
314 npoints = self.npoints + 2 * len(csv_data)
315
316 internal_signal = np.zeros(npoints)
317
318 d=np.random.exponential(period)
319 pulse_positions = []
320 while d<npoints-len(csv_data):
321 pulse_positions.append(d)
322 d+=np.random.exponential(period)
323
324 for pos in pulse_positions:
325 pos = int(pos + 0.5)
326 internal_signal[pos:pos+len(csv_data)] += csv_data
327
328 signal += internal_signal[len(csv_data):-len(csv_data)]
329
330 def __str__(self):
331 s = self.name + '\n'
332 s += 'possible options and parameters\n'
333 s += ' * len: number of samples (100)\n'
334 s += ' * noise: sigma (1)\n'
335 s += ' * bsl: level (-0.5)\n'
336 s += ' * step: height, start, end\n'
337 s += ' * triangle: pos height risingedge, fallingedge [pos height ...]\n'
338 s += ' * spike: pos height [pos height ...]\n'
339 s += ' * csv: pos height [pos height ...]\n'
340
341 s += 'current options are:\n'
342 for key in self.options.keys():
343 s += key + ':' + str(self.options[key]) + '\n'
344 return s
345
346
347# Helper function to parse signalname and create a dictionary
348# dictionary layout :
349# key : string
350# value : [list of parameters]
351def make_options_from_str(signalname):
352 options = {}
353 for word in (signalname.lower()).split():
354 if word.isalpha():
355 current_key = word
356 options[current_key] = []
357# if word.isdigit():
358 else:
359 options[current_key].append(word)
360# else:
361# print '-nothing'
362 return options
363
364if __name__ == '__main__':
365
366
367 num_events = 10
368
369 # interesting
370 phe = 1 # i.e. the signal amplitude
371 rate = 10 # MHz
372 noise = 1. # mV sigma
373
374 if len(sys.argv) > 1:
375 num_events = int(sys.argv[1])
376 if len(sys.argv) > 2:
377 phe = float(sys.argv[2])
378 if len(sys.argv) > 3:
379 rate = float(sys.argv[3])
380 if len(sys.argv) > 4:
381 noise = float(sys.argv[4])
382
383
384 #less interesting
385 length = 1000
386 baseline = 0.0
387 signal_pos = 65
388
389 generator_string = 'len ' + str(length) + ' '
390 generator_string += 'bsl ' + str(baseline) + ' '
391 generator_string += 'signal ' + str(phe) + ' ' + str(signal_pos) + ' '
392 generator_string += 'noise ' + str(noise) + ' '
393 generator_string += 'rate ' + str(rate) + ' '
394 gen = SignalGeneratorCSV('PulseTemplate_PointSet_0.csv', generator_string)
395 print gen
396
397 rootfilename = str(phe)+'phe'
398 rootfilename += '_'+str(rate)+'MHz'
399 rootfilename += '_'+str(noise)+'mV'
400 rootfilename += '_'+str(num_events)
401 rootfilename += '.root'
402
403
404
405 file = TFile(rootfilename, 'recreate')
406
407 hist = TH1D('hist','to be done', length, -0.5,length-0.5)
408 hist.SetStats(kFALSE)
409 hist.SetXTitle('time in slices')
410 hist.SetYTitle('amplitude in mV - no baseline adjustment done yet :-( ')
411
412 for event in range(10):
413 signal = gen()
414
415 histogram_title = str(phe)+'phe @' + str(signal_pos)
416 histogram_title += ' and ' + str(rate) +'MHz dark count rate'
417 histogram_title += ', electronics noise with sigma=' + str(noise) +'mV'
418 histogram_title += ' #' + str(event)
419 hist.SetTitle(histogram_title )
420
421 for i,s in enumerate(signal):
422 hist.SetBinContent(i+1,s)
423 hist.Write()
424 hist.Reset()
425 file.Close()
Note: See TracBrowser for help on using the repository browser.