1 | import re
|
---|
2 | from asap.scantable import scantable
|
---|
3 | from asap.parameters import rcParams
|
---|
4 | from asap.logging import asaplog, asaplog_post_dec
|
---|
5 | from asap.selector import selector
|
---|
6 | from asap.asapplotter import new_asaplot
|
---|
7 | from matplotlib import rc as rcp
|
---|
8 |
|
---|
9 | @asaplog_post_dec
|
---|
10 | def average_time(*args, **kwargs):
|
---|
11 | """
|
---|
12 | Return the (time) average of a scan or list of scans. [in channels only]
|
---|
13 | The cursor of the output scan is set to 0
|
---|
14 | Parameters:
|
---|
15 | one scan or comma separated scans or a list of scans
|
---|
16 | mask: an optional mask (only used for 'var' and 'tsys' weighting)
|
---|
17 | scanav: True averages each scan separately.
|
---|
18 | False (default) averages all scans together,
|
---|
19 | weight: Weighting scheme.
|
---|
20 | 'none' (mean no weight)
|
---|
21 | 'var' (1/var(spec) weighted)
|
---|
22 | 'tsys' (1/Tsys**2 weighted)
|
---|
23 | 'tint' (integration time weighted)
|
---|
24 | 'tintsys' (Tint/Tsys**2)
|
---|
25 | 'median' ( median averaging)
|
---|
26 | align: align the spectra in velocity before averaging. It takes
|
---|
27 | the time of the first spectrum in the first scantable
|
---|
28 | as reference time.
|
---|
29 | compel: True forces to average overwrapped IFs.
|
---|
30 | Example:
|
---|
31 | # return a time averaged scan from scana and scanb
|
---|
32 | # without using a mask
|
---|
33 | scanav = average_time(scana,scanb)
|
---|
34 | # or equivalent
|
---|
35 | # scanav = average_time([scana, scanb])
|
---|
36 | # return the (time) averaged scan, i.e. the average of
|
---|
37 | # all correlator cycles
|
---|
38 | scanav = average_time(scan, scanav=True)
|
---|
39 | """
|
---|
40 | scanav = False
|
---|
41 | if kwargs.has_key('scanav'):
|
---|
42 | scanav = kwargs.get('scanav')
|
---|
43 | weight = 'tint'
|
---|
44 | if kwargs.has_key('weight'):
|
---|
45 | weight = kwargs.get('weight')
|
---|
46 | mask = ()
|
---|
47 | if kwargs.has_key('mask'):
|
---|
48 | mask = kwargs.get('mask')
|
---|
49 | align = False
|
---|
50 | if kwargs.has_key('align'):
|
---|
51 | align = kwargs.get('align')
|
---|
52 | compel = False
|
---|
53 | if kwargs.has_key('compel'):
|
---|
54 | compel = kwargs.get('compel')
|
---|
55 | varlist = vars()
|
---|
56 | if isinstance(args[0],list):
|
---|
57 | lst = args[0]
|
---|
58 | elif isinstance(args[0],tuple):
|
---|
59 | lst = list(args[0])
|
---|
60 | else:
|
---|
61 | lst = list(args)
|
---|
62 |
|
---|
63 | del varlist["kwargs"]
|
---|
64 | varlist["args"] = "%d scantables" % len(lst)
|
---|
65 | # need special formatting here for history...
|
---|
66 |
|
---|
67 | from asap._asap import stmath
|
---|
68 | stm = stmath()
|
---|
69 | for s in lst:
|
---|
70 | if not isinstance(s,scantable):
|
---|
71 | msg = "Please give a list of scantables"
|
---|
72 | raise TypeError(msg)
|
---|
73 | if scanav: scanav = "SCAN"
|
---|
74 | else: scanav = "NONE"
|
---|
75 | alignedlst = []
|
---|
76 | if align:
|
---|
77 | refepoch = lst[0].get_time(0)
|
---|
78 | for scan in lst:
|
---|
79 | alignedlst.append(scan.freq_align(refepoch,insitu=False))
|
---|
80 | else:
|
---|
81 | alignedlst = lst
|
---|
82 | if weight.upper() == 'MEDIAN':
|
---|
83 | # median doesn't support list of scantables - merge first
|
---|
84 | merged = None
|
---|
85 | if len(alignedlst) > 1:
|
---|
86 | merged = merge(alignedlst)
|
---|
87 | else:
|
---|
88 | merged = alignedlst[0]
|
---|
89 | s = scantable(stm._averagechannel(merged, 'MEDIAN', scanav))
|
---|
90 | del merged
|
---|
91 | else:
|
---|
92 | #s = scantable(stm._average(alignedlst, mask, weight.upper(), scanav))
|
---|
93 | s = scantable(stm._new_average(alignedlst, compel, mask,
|
---|
94 | weight.upper(), scanav))
|
---|
95 | s._add_history("average_time",varlist)
|
---|
96 |
|
---|
97 | return s
|
---|
98 |
|
---|
99 | @asaplog_post_dec
|
---|
100 | def quotient(source, reference, preserve=True):
|
---|
101 | """
|
---|
102 | Return the quotient of a 'source' (signal) scan and a 'reference' scan.
|
---|
103 | The reference can have just one scan, even if the signal has many. Otherwise
|
---|
104 | they must have the same number of scans.
|
---|
105 | The cursor of the output scan is set to 0
|
---|
106 | Parameters:
|
---|
107 | source: the 'on' scan
|
---|
108 | reference: the 'off' scan
|
---|
109 | preserve: you can preserve (default) the continuum or
|
---|
110 | remove it. The equations used are
|
---|
111 | preserve: Output = Toff * (on/off) - Toff
|
---|
112 | remove: Output = Toff * (on/off) - Ton
|
---|
113 | """
|
---|
114 | varlist = vars()
|
---|
115 | from asap._asap import stmath
|
---|
116 | stm = stmath()
|
---|
117 | stm._setinsitu(False)
|
---|
118 | s = scantable(stm._quotient(source, reference, preserve))
|
---|
119 | s._add_history("quotient",varlist)
|
---|
120 | return s
|
---|
121 |
|
---|
122 | @asaplog_post_dec
|
---|
123 | def dototalpower(calon, caloff, tcalval=0.0):
|
---|
124 | """
|
---|
125 | Do calibration for CAL on,off signals.
|
---|
126 | Adopted from GBTIDL dototalpower
|
---|
127 | Parameters:
|
---|
128 | calon: the 'cal on' subintegration
|
---|
129 | caloff: the 'cal off' subintegration
|
---|
130 | tcalval: user supplied Tcal value
|
---|
131 | """
|
---|
132 | varlist = vars()
|
---|
133 | from asap._asap import stmath
|
---|
134 | stm = stmath()
|
---|
135 | stm._setinsitu(False)
|
---|
136 | s = scantable(stm._dototalpower(calon, caloff, tcalval))
|
---|
137 | s._add_history("dototalpower",varlist)
|
---|
138 | return s
|
---|
139 |
|
---|
140 | @asaplog_post_dec
|
---|
141 | def dosigref(sig, ref, smooth, tsysval=0.0, tauval=0.0):
|
---|
142 | """
|
---|
143 | Calculate a quotient (sig-ref/ref * Tsys)
|
---|
144 | Adopted from GBTIDL dosigref
|
---|
145 | Parameters:
|
---|
146 | sig: on source data
|
---|
147 | ref: reference data
|
---|
148 | smooth: width of box car smoothing for reference
|
---|
149 | tsysval: user specified Tsys (scalar only)
|
---|
150 | tauval: user specified Tau (required if tsysval is set)
|
---|
151 | """
|
---|
152 | varlist = vars()
|
---|
153 | from asap._asap import stmath
|
---|
154 | stm = stmath()
|
---|
155 | stm._setinsitu(False)
|
---|
156 | s = scantable(stm._dosigref(sig, ref, smooth, tsysval, tauval))
|
---|
157 | s._add_history("dosigref",varlist)
|
---|
158 | return s
|
---|
159 |
|
---|
160 | @asaplog_post_dec
|
---|
161 | def calps(scantab, scannos, smooth=1, tsysval=0.0, tauval=0.0, tcalval=0.0, verify=False):
|
---|
162 | """
|
---|
163 | Calibrate GBT position switched data
|
---|
164 | Adopted from GBTIDL getps
|
---|
165 | Currently calps identify the scans as position switched data if source
|
---|
166 | type enum is pson or psoff. The data must contains 'CAL' signal
|
---|
167 | on/off in each integration. To identify 'CAL' on state, the source type
|
---|
168 | enum of poncal and poffcal need to be present.
|
---|
169 |
|
---|
170 | Parameters:
|
---|
171 | scantab: scantable
|
---|
172 | scannos: list of scan numbers
|
---|
173 | smooth: optional box smoothing order for the reference
|
---|
174 | (default is 1 = no smoothing)
|
---|
175 | tsysval: optional user specified Tsys (default is 0.0,
|
---|
176 | use Tsys in the data)
|
---|
177 | tauval: optional user specified Tau
|
---|
178 | tcalval: optional user specified Tcal (default is 0.0,
|
---|
179 | use Tcal value in the data)
|
---|
180 | verify: Verify calibration if true
|
---|
181 | """
|
---|
182 | varlist = vars()
|
---|
183 | # check for the appropriate data
|
---|
184 | ## s = scantab.get_scan('*_ps*')
|
---|
185 | ## if s is None:
|
---|
186 | ## msg = "The input data appear to contain no position-switch mode data."
|
---|
187 | ## raise TypeError(msg)
|
---|
188 | s = scantab.copy()
|
---|
189 | from asap._asap import srctype
|
---|
190 | sel = selector()
|
---|
191 | sel.set_types( srctype.pson )
|
---|
192 | try:
|
---|
193 | scantab.set_selection( sel )
|
---|
194 | except Exception, e:
|
---|
195 | msg = "The input data appear to contain no position-switch mode data."
|
---|
196 | raise TypeError(msg)
|
---|
197 | s.set_selection()
|
---|
198 | sel.reset()
|
---|
199 | ssub = s.get_scan(scannos)
|
---|
200 | if ssub is None:
|
---|
201 | msg = "No data was found with given scan numbers!"
|
---|
202 | raise TypeError(msg)
|
---|
203 | #ssubon = ssub.get_scan('*calon')
|
---|
204 | #ssuboff = ssub.get_scan('*[^calon]')
|
---|
205 | sel.set_types( [srctype.poncal,srctype.poffcal] )
|
---|
206 | ssub.set_selection( sel )
|
---|
207 | ssubon = ssub.copy()
|
---|
208 | ssub.set_selection()
|
---|
209 | sel.reset()
|
---|
210 | sel.set_types( [srctype.pson,srctype.psoff] )
|
---|
211 | ssub.set_selection( sel )
|
---|
212 | ssuboff = ssub.copy()
|
---|
213 | ssub.set_selection()
|
---|
214 | sel.reset()
|
---|
215 | if ssubon.nrow() != ssuboff.nrow():
|
---|
216 | msg = "mismatch in numbers of CAL on/off scans. Cannot calibrate. Check the scan numbers."
|
---|
217 | raise TypeError(msg)
|
---|
218 | cals = dototalpower(ssubon, ssuboff, tcalval)
|
---|
219 | #sig = cals.get_scan('*ps')
|
---|
220 | #ref = cals.get_scan('*psr')
|
---|
221 | sel.set_types( srctype.pson )
|
---|
222 | cals.set_selection( sel )
|
---|
223 | sig = cals.copy()
|
---|
224 | cals.set_selection()
|
---|
225 | sel.reset()
|
---|
226 | sel.set_types( srctype.psoff )
|
---|
227 | cals.set_selection( sel )
|
---|
228 | ref = cals.copy()
|
---|
229 | cals.set_selection()
|
---|
230 | sel.reset()
|
---|
231 | if sig.nscan() != ref.nscan():
|
---|
232 | msg = "mismatch in numbers of on/off scans. Cannot calibrate. Check the scan numbers."
|
---|
233 | raise TypeError(msg)
|
---|
234 |
|
---|
235 | #for user supplied Tsys
|
---|
236 | if tsysval>0.0:
|
---|
237 | if tauval<=0.0:
|
---|
238 | msg = "Need to supply a valid tau to use the supplied Tsys"
|
---|
239 | raise TypeError(msg)
|
---|
240 | else:
|
---|
241 | sig.recalc_azel()
|
---|
242 | ref.recalc_azel()
|
---|
243 | #msg = "Use of user specified Tsys is not fully implemented yet."
|
---|
244 | #raise TypeError(msg)
|
---|
245 | # use get_elevation to get elevation and
|
---|
246 | # calculate a scaling factor using the formula
|
---|
247 | # -> tsys use to dosigref
|
---|
248 |
|
---|
249 | #ress = dosigref(sig, ref, smooth, tsysval)
|
---|
250 | ress = dosigref(sig, ref, smooth, tsysval, tauval)
|
---|
251 | ###
|
---|
252 | if verify:
|
---|
253 | # get data
|
---|
254 | import numpy
|
---|
255 | precal={}
|
---|
256 | postcal=[]
|
---|
257 | keys=['ps','ps_calon','psr','psr_calon']
|
---|
258 | types=[srctype.pson,srctype.poncal,srctype.psoff,srctype.poffcal]
|
---|
259 | ifnos=list(ssub.getifnos())
|
---|
260 | polnos=list(ssub.getpolnos())
|
---|
261 | sel=selector()
|
---|
262 | for i in range(2):
|
---|
263 | #ss=ssuboff.get_scan('*'+keys[2*i])
|
---|
264 | ll=[]
|
---|
265 | for j in range(len(ifnos)):
|
---|
266 | for k in range(len(polnos)):
|
---|
267 | sel.set_ifs(ifnos[j])
|
---|
268 | sel.set_polarizations(polnos[k])
|
---|
269 | sel.set_types(types[2*i])
|
---|
270 | try:
|
---|
271 | #ss.set_selection(sel)
|
---|
272 | ssuboff.set_selection(sel)
|
---|
273 | except:
|
---|
274 | continue
|
---|
275 | #ll.append(numpy.array(ss._getspectrum(0)))
|
---|
276 | ll.append(numpy.array(ssuboff._getspectrum(0)))
|
---|
277 | sel.reset()
|
---|
278 | ssuboff.set_selection()
|
---|
279 | precal[keys[2*i]]=ll
|
---|
280 | #del ss
|
---|
281 | #ss=ssubon.get_scan('*'+keys[2*i+1])
|
---|
282 | ll=[]
|
---|
283 | for j in range(len(ifnos)):
|
---|
284 | for k in range(len(polnos)):
|
---|
285 | sel.set_ifs(ifnos[j])
|
---|
286 | sel.set_polarizations(polnos[k])
|
---|
287 | sel.set_types(types[2*i+1])
|
---|
288 | try:
|
---|
289 | #ss.set_selection(sel)
|
---|
290 | ssubon.set_selection(sel)
|
---|
291 | except:
|
---|
292 | continue
|
---|
293 | #ll.append(numpy.array(ss._getspectrum(0)))
|
---|
294 | ll.append(numpy.array(ssubon._getspectrum(0)))
|
---|
295 | sel.reset()
|
---|
296 | ssubon.set_selection()
|
---|
297 | precal[keys[2*i+1]]=ll
|
---|
298 | #del ss
|
---|
299 | for j in range(len(ifnos)):
|
---|
300 | for k in range(len(polnos)):
|
---|
301 | sel.set_ifs(ifnos[j])
|
---|
302 | sel.set_polarizations(polnos[k])
|
---|
303 | try:
|
---|
304 | ress.set_selection(sel)
|
---|
305 | except:
|
---|
306 | continue
|
---|
307 | postcal.append(numpy.array(ress._getspectrum(0)))
|
---|
308 | sel.reset()
|
---|
309 | ress.set_selection()
|
---|
310 | del sel
|
---|
311 | # plot
|
---|
312 | asaplog.post()
|
---|
313 | asaplog.push('Plot only first spectrum for each [if,pol] pairs to verify calibration.')
|
---|
314 | asaplog.post('WARN')
|
---|
315 | p=new_asaplot()
|
---|
316 | rcp('lines', linewidth=1)
|
---|
317 | #nr=min(6,len(ifnos)*len(polnos))
|
---|
318 | nr=len(ifnos)*len(polnos)
|
---|
319 | titles=[]
|
---|
320 | btics=[]
|
---|
321 | if nr<4:
|
---|
322 | p.set_panels(rows=nr,cols=2,nplots=2*nr,ganged=False)
|
---|
323 | for i in range(2*nr):
|
---|
324 | b=False
|
---|
325 | if i >= 2*nr-2:
|
---|
326 | b=True
|
---|
327 | btics.append(b)
|
---|
328 | elif nr==4:
|
---|
329 | p.set_panels(rows=2,cols=4,nplots=8,ganged=False)
|
---|
330 | for i in range(2*nr):
|
---|
331 | b=False
|
---|
332 | if i >= 2*nr-4:
|
---|
333 | b=True
|
---|
334 | btics.append(b)
|
---|
335 | elif nr<7:
|
---|
336 | p.set_panels(rows=3,cols=4,nplots=2*nr,ganged=False)
|
---|
337 | for i in range(2*nr):
|
---|
338 | if i >= 2*nr-4:
|
---|
339 | b=True
|
---|
340 | btics.append(b)
|
---|
341 | else:
|
---|
342 | asaplog.post()
|
---|
343 | asaplog.push('Only first 6 [if,pol] pairs are plotted.')
|
---|
344 | asaplog.post('WARN')
|
---|
345 | nr=6
|
---|
346 | for i in range(2*nr):
|
---|
347 | b=False
|
---|
348 | if i >= 2*nr-4:
|
---|
349 | b=True
|
---|
350 | btics.append(b)
|
---|
351 | p.set_panels(rows=3,cols=4,nplots=2*nr,ganged=False)
|
---|
352 | for i in range(nr):
|
---|
353 | p.subplot(2*i)
|
---|
354 | p.color=0
|
---|
355 | title='raw data IF%s POL%s' % (ifnos[int(i/len(polnos))],polnos[i%len(polnos)])
|
---|
356 | titles.append(title)
|
---|
357 | #p.set_axes('title',title,fontsize=40)
|
---|
358 | ymin=1.0e100
|
---|
359 | ymax=-1.0e100
|
---|
360 | nchan=s.nchan(ifnos[int(i/len(polnos))])
|
---|
361 | edge=int(nchan*0.01)
|
---|
362 | for j in range(4):
|
---|
363 | spmin=min(precal[keys[j]][i][edge:nchan-edge])
|
---|
364 | spmax=max(precal[keys[j]][i][edge:nchan-edge])
|
---|
365 | ymin=min(ymin,spmin)
|
---|
366 | ymax=max(ymax,spmax)
|
---|
367 | for j in range(4):
|
---|
368 | if i==0:
|
---|
369 | p.set_line(label=keys[j])
|
---|
370 | else:
|
---|
371 | p.legend()
|
---|
372 | p.plot(precal[keys[j]][i])
|
---|
373 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
374 | if not btics[2*i]:
|
---|
375 | p.axes.set_xticks([])
|
---|
376 | p.subplot(2*i+1)
|
---|
377 | p.color=0
|
---|
378 | title='cal data IF%s POL%s' % (ifnos[int(i/len(polnos))],polnos[i%len(polnos)])
|
---|
379 | titles.append(title)
|
---|
380 | #p.set_axes('title',title)
|
---|
381 | p.legend()
|
---|
382 | ymin=postcal[i][edge:nchan-edge].min()
|
---|
383 | ymax=postcal[i][edge:nchan-edge].max()
|
---|
384 | p.plot(postcal[i])
|
---|
385 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
386 | if not btics[2*i+1]:
|
---|
387 | p.axes.set_xticks([])
|
---|
388 | for i in range(2*nr):
|
---|
389 | p.subplot(i)
|
---|
390 | p.set_axes('title',titles[i],fontsize='medium')
|
---|
391 | x=raw_input('Accept calibration ([y]/n): ' )
|
---|
392 | if x.upper() == 'N':
|
---|
393 | p.quit()
|
---|
394 | del p
|
---|
395 | return scantab
|
---|
396 | p.quit()
|
---|
397 | del p
|
---|
398 | ###
|
---|
399 | ress._add_history("calps", varlist)
|
---|
400 | return ress
|
---|
401 |
|
---|
402 | @asaplog_post_dec
|
---|
403 | def calnod(scantab, scannos=[], smooth=1, tsysval=0.0, tauval=0.0, tcalval=0.0, verify=False):
|
---|
404 | """
|
---|
405 | Do full (but a pair of scans at time) processing of GBT Nod data
|
---|
406 | calibration.
|
---|
407 | Adopted from GBTIDL's getnod
|
---|
408 | Parameters:
|
---|
409 | scantab: scantable
|
---|
410 | scannos: a pair of scan numbers, or the first scan number of the pair
|
---|
411 | smooth: box car smoothing order
|
---|
412 | tsysval: optional user specified Tsys value
|
---|
413 | tauval: optional user specified tau value (not implemented yet)
|
---|
414 | tcalval: optional user specified Tcal value
|
---|
415 | verify: Verify calibration if true
|
---|
416 | """
|
---|
417 | varlist = vars()
|
---|
418 | from asap._asap import stmath
|
---|
419 | from asap._asap import srctype
|
---|
420 | stm = stmath()
|
---|
421 | stm._setinsitu(False)
|
---|
422 |
|
---|
423 | # check for the appropriate data
|
---|
424 | ## s = scantab.get_scan('*_nod*')
|
---|
425 | ## if s is None:
|
---|
426 | ## msg = "The input data appear to contain no Nod observing mode data."
|
---|
427 | ## raise TypeError(msg)
|
---|
428 | s = scantab.copy()
|
---|
429 | sel = selector()
|
---|
430 | sel.set_types( srctype.nod )
|
---|
431 | try:
|
---|
432 | s.set_selection( sel )
|
---|
433 | except Exception, e:
|
---|
434 | msg = "The input data appear to contain no Nod observing mode data."
|
---|
435 | raise TypeError(msg)
|
---|
436 | sel.reset()
|
---|
437 | del sel
|
---|
438 | del s
|
---|
439 |
|
---|
440 | # need check correspondance of each beam with sig-ref ...
|
---|
441 | # check for timestamps, scan numbers, subscan id (not available in
|
---|
442 | # ASAP data format...). Assume 1st scan of the pair (beam 0 - sig
|
---|
443 | # and beam 1 - ref...)
|
---|
444 | # First scan number of paired scans or list of pairs of
|
---|
445 | # scan numbers (has to have even number of pairs.)
|
---|
446 |
|
---|
447 | #data splitting
|
---|
448 | scan1no = scan2no = 0
|
---|
449 |
|
---|
450 | if len(scannos)==1:
|
---|
451 | scan1no = scannos[0]
|
---|
452 | scan2no = scannos[0]+1
|
---|
453 | pairScans = [scan1no, scan2no]
|
---|
454 | else:
|
---|
455 | #if len(scannos)>2:
|
---|
456 | # msg = "calnod can only process a pair of nod scans at time."
|
---|
457 | # raise TypeError(msg)
|
---|
458 | #
|
---|
459 | #if len(scannos)==2:
|
---|
460 | # scan1no = scannos[0]
|
---|
461 | # scan2no = scannos[1]
|
---|
462 | pairScans = list(scannos)
|
---|
463 |
|
---|
464 | if tsysval>0.0:
|
---|
465 | if tauval<=0.0:
|
---|
466 | msg = "Need to supply a valid tau to use the supplied Tsys"
|
---|
467 | raise TypeError(msg)
|
---|
468 | else:
|
---|
469 | scantab.recalc_azel()
|
---|
470 | resspec = scantable(stm._donod(scantab, pairScans, smooth, tsysval,tauval,tcalval))
|
---|
471 | ###
|
---|
472 | if verify:
|
---|
473 | # get data
|
---|
474 | import numpy
|
---|
475 | precal={}
|
---|
476 | postcal=[]
|
---|
477 | keys=['','_calon']
|
---|
478 | types=[srctype.nod,srctype.nodcal]
|
---|
479 | ifnos=list(scantab.getifnos())
|
---|
480 | polnos=list(scantab.getpolnos())
|
---|
481 | sel=selector()
|
---|
482 | ss = scantab.copy()
|
---|
483 | for i in range(2):
|
---|
484 | #ss=scantab.get_scan('*'+keys[i])
|
---|
485 | ll=[]
|
---|
486 | ll2=[]
|
---|
487 | for j in range(len(ifnos)):
|
---|
488 | for k in range(len(polnos)):
|
---|
489 | sel.set_ifs(ifnos[j])
|
---|
490 | sel.set_polarizations(polnos[k])
|
---|
491 | sel.set_scans(pairScans[0])
|
---|
492 | sel.set_types(types[i])
|
---|
493 | try:
|
---|
494 | ss.set_selection(sel)
|
---|
495 | except:
|
---|
496 | continue
|
---|
497 | ll.append(numpy.array(ss._getspectrum(0)))
|
---|
498 | sel.reset()
|
---|
499 | ss.set_selection()
|
---|
500 | sel.set_ifs(ifnos[j])
|
---|
501 | sel.set_polarizations(polnos[k])
|
---|
502 | sel.set_scans(pairScans[1])
|
---|
503 | sel.set_types(types[i])
|
---|
504 | try:
|
---|
505 | ss.set_selection(sel)
|
---|
506 | except:
|
---|
507 | ll.pop()
|
---|
508 | continue
|
---|
509 | ll2.append(numpy.array(ss._getspectrum(0)))
|
---|
510 | sel.reset()
|
---|
511 | ss.set_selection()
|
---|
512 | key='%s%s' %(pairScans[0],keys[i])
|
---|
513 | precal[key]=ll
|
---|
514 | key='%s%s' %(pairScans[1],keys[i])
|
---|
515 | precal[key]=ll2
|
---|
516 | #del ss
|
---|
517 | keys=precal.keys()
|
---|
518 | for j in range(len(ifnos)):
|
---|
519 | for k in range(len(polnos)):
|
---|
520 | sel.set_ifs(ifnos[j])
|
---|
521 | sel.set_polarizations(polnos[k])
|
---|
522 | sel.set_scans(pairScans[0])
|
---|
523 | try:
|
---|
524 | resspec.set_selection(sel)
|
---|
525 | except:
|
---|
526 | continue
|
---|
527 | postcal.append(numpy.array(resspec._getspectrum(0)))
|
---|
528 | sel.reset()
|
---|
529 | resspec.set_selection()
|
---|
530 | del sel
|
---|
531 | # plot
|
---|
532 | asaplog.post()
|
---|
533 | asaplog.push('Plot only first spectrum for each [if,pol] pairs to verify calibration.')
|
---|
534 | asaplog.post('WARN')
|
---|
535 | p=new_asaplot()
|
---|
536 | rcp('lines', linewidth=1)
|
---|
537 | #nr=min(6,len(ifnos)*len(polnos))
|
---|
538 | nr=len(ifnos)*len(polnos)
|
---|
539 | titles=[]
|
---|
540 | btics=[]
|
---|
541 | if nr<4:
|
---|
542 | p.set_panels(rows=nr,cols=2,nplots=2*nr,ganged=False)
|
---|
543 | for i in range(2*nr):
|
---|
544 | b=False
|
---|
545 | if i >= 2*nr-2:
|
---|
546 | b=True
|
---|
547 | btics.append(b)
|
---|
548 | elif nr==4:
|
---|
549 | p.set_panels(rows=2,cols=4,nplots=8,ganged=False)
|
---|
550 | for i in range(2*nr):
|
---|
551 | b=False
|
---|
552 | if i >= 2*nr-4:
|
---|
553 | b=True
|
---|
554 | btics.append(b)
|
---|
555 | elif nr<7:
|
---|
556 | p.set_panels(rows=3,cols=4,nplots=2*nr,ganged=False)
|
---|
557 | for i in range(2*nr):
|
---|
558 | if i >= 2*nr-4:
|
---|
559 | b=True
|
---|
560 | btics.append(b)
|
---|
561 | else:
|
---|
562 | asaplog.post()
|
---|
563 | asaplog.push('Only first 6 [if,pol] pairs are plotted.')
|
---|
564 | asaplog.post('WARN')
|
---|
565 | nr=6
|
---|
566 | for i in range(2*nr):
|
---|
567 | b=False
|
---|
568 | if i >= 2*nr-4:
|
---|
569 | b=True
|
---|
570 | btics.append(b)
|
---|
571 | p.set_panels(rows=3,cols=4,nplots=2*nr,ganged=False)
|
---|
572 | for i in range(nr):
|
---|
573 | p.subplot(2*i)
|
---|
574 | p.color=0
|
---|
575 | title='raw data IF%s POL%s' % (ifnos[int(i/len(polnos))],polnos[i%len(polnos)])
|
---|
576 | titles.append(title)
|
---|
577 | #p.set_axes('title',title,fontsize=40)
|
---|
578 | ymin=1.0e100
|
---|
579 | ymax=-1.0e100
|
---|
580 | nchan=scantab.nchan(ifnos[int(i/len(polnos))])
|
---|
581 | edge=int(nchan*0.01)
|
---|
582 | for j in range(4):
|
---|
583 | spmin=min(precal[keys[j]][i][edge:nchan-edge])
|
---|
584 | spmax=max(precal[keys[j]][i][edge:nchan-edge])
|
---|
585 | ymin=min(ymin,spmin)
|
---|
586 | ymax=max(ymax,spmax)
|
---|
587 | for j in range(4):
|
---|
588 | if i==0:
|
---|
589 | p.set_line(label=keys[j])
|
---|
590 | else:
|
---|
591 | p.legend()
|
---|
592 | p.plot(precal[keys[j]][i])
|
---|
593 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
594 | if not btics[2*i]:
|
---|
595 | p.axes.set_xticks([])
|
---|
596 | p.subplot(2*i+1)
|
---|
597 | p.color=0
|
---|
598 | title='cal data IF%s POL%s' % (ifnos[int(i/len(polnos))],polnos[i%len(polnos)])
|
---|
599 | titles.append(title)
|
---|
600 | #p.set_axes('title',title)
|
---|
601 | p.legend()
|
---|
602 | ymin=postcal[i][edge:nchan-edge].min()
|
---|
603 | ymax=postcal[i][edge:nchan-edge].max()
|
---|
604 | p.plot(postcal[i])
|
---|
605 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
606 | if not btics[2*i+1]:
|
---|
607 | p.axes.set_xticks([])
|
---|
608 | for i in range(2*nr):
|
---|
609 | p.subplot(i)
|
---|
610 | p.set_axes('title',titles[i],fontsize='medium')
|
---|
611 | x=raw_input('Accept calibration ([y]/n): ' )
|
---|
612 | if x.upper() == 'N':
|
---|
613 | p.quit()
|
---|
614 | del p
|
---|
615 | return scantab
|
---|
616 | p.quit()
|
---|
617 | del p
|
---|
618 | ###
|
---|
619 | resspec._add_history("calnod",varlist)
|
---|
620 | return resspec
|
---|
621 |
|
---|
622 | @asaplog_post_dec
|
---|
623 | def calfs(scantab, scannos=[], smooth=1, tsysval=0.0, tauval=0.0, tcalval=0.0, verify=False):
|
---|
624 | """
|
---|
625 | Calibrate GBT frequency switched data.
|
---|
626 | Adopted from GBTIDL getfs.
|
---|
627 | Currently calfs identify the scans as frequency switched data if source
|
---|
628 | type enum is fson and fsoff. The data must contains 'CAL' signal
|
---|
629 | on/off in each integration. To identify 'CAL' on state, the source type
|
---|
630 | enum of foncal and foffcal need to be present.
|
---|
631 |
|
---|
632 | Parameters:
|
---|
633 | scantab: scantable
|
---|
634 | scannos: list of scan numbers
|
---|
635 | smooth: optional box smoothing order for the reference
|
---|
636 | (default is 1 = no smoothing)
|
---|
637 | tsysval: optional user specified Tsys (default is 0.0,
|
---|
638 | use Tsys in the data)
|
---|
639 | tauval: optional user specified Tau
|
---|
640 | verify: Verify calibration if true
|
---|
641 | """
|
---|
642 | varlist = vars()
|
---|
643 | from asap._asap import stmath
|
---|
644 | from asap._asap import srctype
|
---|
645 | stm = stmath()
|
---|
646 | stm._setinsitu(False)
|
---|
647 |
|
---|
648 | # check = scantab.get_scan('*_fs*')
|
---|
649 | # if check is None:
|
---|
650 | # msg = "The input data appear to contain no Nod observing mode data."
|
---|
651 | # raise TypeError(msg)
|
---|
652 | s = scantab.get_scan(scannos)
|
---|
653 | del scantab
|
---|
654 |
|
---|
655 | resspec = scantable(stm._dofs(s, scannos, smooth, tsysval,tauval,tcalval))
|
---|
656 | ###
|
---|
657 | if verify:
|
---|
658 | # get data
|
---|
659 | ssub = s.get_scan(scannos)
|
---|
660 | #ssubon = ssub.get_scan('*calon')
|
---|
661 | #ssuboff = ssub.get_scan('*[^calon]')
|
---|
662 | sel = selector()
|
---|
663 | sel.set_types( [srctype.foncal,srctype.foffcal] )
|
---|
664 | ssub.set_selection( sel )
|
---|
665 | ssubon = ssub.copy()
|
---|
666 | ssub.set_selection()
|
---|
667 | sel.reset()
|
---|
668 | sel.set_types( [srctype.fson,srctype.fsoff] )
|
---|
669 | ssub.set_selection( sel )
|
---|
670 | ssuboff = ssub.copy()
|
---|
671 | ssub.set_selection()
|
---|
672 | sel.reset()
|
---|
673 | import numpy
|
---|
674 | precal={}
|
---|
675 | postcal=[]
|
---|
676 | keys=['fs','fs_calon','fsr','fsr_calon']
|
---|
677 | types=[srctype.fson,srctype.foncal,srctype.fsoff,srctype.foffcal]
|
---|
678 | ifnos=list(ssub.getifnos())
|
---|
679 | polnos=list(ssub.getpolnos())
|
---|
680 | for i in range(2):
|
---|
681 | #ss=ssuboff.get_scan('*'+keys[2*i])
|
---|
682 | ll=[]
|
---|
683 | for j in range(len(ifnos)):
|
---|
684 | for k in range(len(polnos)):
|
---|
685 | sel.set_ifs(ifnos[j])
|
---|
686 | sel.set_polarizations(polnos[k])
|
---|
687 | sel.set_types(types[2*i])
|
---|
688 | try:
|
---|
689 | #ss.set_selection(sel)
|
---|
690 | ssuboff.set_selection(sel)
|
---|
691 | except:
|
---|
692 | continue
|
---|
693 | ll.append(numpy.array(ss._getspectrum(0)))
|
---|
694 | sel.reset()
|
---|
695 | #ss.set_selection()
|
---|
696 | ssuboff.set_selection()
|
---|
697 | precal[keys[2*i]]=ll
|
---|
698 | #del ss
|
---|
699 | #ss=ssubon.get_scan('*'+keys[2*i+1])
|
---|
700 | ll=[]
|
---|
701 | for j in range(len(ifnos)):
|
---|
702 | for k in range(len(polnos)):
|
---|
703 | sel.set_ifs(ifnos[j])
|
---|
704 | sel.set_polarizations(polnos[k])
|
---|
705 | sel.set_types(types[2*i+1])
|
---|
706 | try:
|
---|
707 | #ss.set_selection(sel)
|
---|
708 | ssubon.set_selection(sel)
|
---|
709 | except:
|
---|
710 | continue
|
---|
711 | ll.append(numpy.array(ss._getspectrum(0)))
|
---|
712 | sel.reset()
|
---|
713 | #ss.set_selection()
|
---|
714 | ssubon.set_selection()
|
---|
715 | precal[keys[2*i+1]]=ll
|
---|
716 | #del ss
|
---|
717 | #sig=resspec.get_scan('*_fs')
|
---|
718 | #ref=resspec.get_scan('*_fsr')
|
---|
719 | sel.set_types( srctype.fson )
|
---|
720 | resspec.set_selection( sel )
|
---|
721 | sig=resspec.copy()
|
---|
722 | resspec.set_selection()
|
---|
723 | sel.reset()
|
---|
724 | sel.set_type( srctype.fsoff )
|
---|
725 | resspec.set_selection( sel )
|
---|
726 | ref=resspec.copy()
|
---|
727 | resspec.set_selection()
|
---|
728 | sel.reset()
|
---|
729 | for k in range(len(polnos)):
|
---|
730 | for j in range(len(ifnos)):
|
---|
731 | sel.set_ifs(ifnos[j])
|
---|
732 | sel.set_polarizations(polnos[k])
|
---|
733 | try:
|
---|
734 | sig.set_selection(sel)
|
---|
735 | postcal.append(numpy.array(sig._getspectrum(0)))
|
---|
736 | except:
|
---|
737 | ref.set_selection(sel)
|
---|
738 | postcal.append(numpy.array(ref._getspectrum(0)))
|
---|
739 | sel.reset()
|
---|
740 | resspec.set_selection()
|
---|
741 | del sel
|
---|
742 | # plot
|
---|
743 | asaplog.post()
|
---|
744 | asaplog.push('Plot only first spectrum for each [if,pol] pairs to verify calibration.')
|
---|
745 | asaplog.post('WARN')
|
---|
746 | p=new_asaplot()
|
---|
747 | rcp('lines', linewidth=1)
|
---|
748 | #nr=min(6,len(ifnos)*len(polnos))
|
---|
749 | nr=len(ifnos)/2*len(polnos)
|
---|
750 | titles=[]
|
---|
751 | btics=[]
|
---|
752 | if nr>3:
|
---|
753 | asaplog.post()
|
---|
754 | asaplog.push('Only first 3 [if,pol] pairs are plotted.')
|
---|
755 | asaplog.post('WARN')
|
---|
756 | nr=3
|
---|
757 | p.set_panels(rows=nr,cols=3,nplots=3*nr,ganged=False)
|
---|
758 | for i in range(3*nr):
|
---|
759 | b=False
|
---|
760 | if i >= 3*nr-3:
|
---|
761 | b=True
|
---|
762 | btics.append(b)
|
---|
763 | for i in range(nr):
|
---|
764 | p.subplot(3*i)
|
---|
765 | p.color=0
|
---|
766 | title='raw data IF%s,%s POL%s' % (ifnos[2*int(i/len(polnos))],ifnos[2*int(i/len(polnos))+1],polnos[i%len(polnos)])
|
---|
767 | titles.append(title)
|
---|
768 | #p.set_axes('title',title,fontsize=40)
|
---|
769 | ymin=1.0e100
|
---|
770 | ymax=-1.0e100
|
---|
771 | nchan=s.nchan(ifnos[2*int(i/len(polnos))])
|
---|
772 | edge=int(nchan*0.01)
|
---|
773 | for j in range(4):
|
---|
774 | spmin=min(precal[keys[j]][i][edge:nchan-edge])
|
---|
775 | spmax=max(precal[keys[j]][i][edge:nchan-edge])
|
---|
776 | ymin=min(ymin,spmin)
|
---|
777 | ymax=max(ymax,spmax)
|
---|
778 | for j in range(4):
|
---|
779 | if i==0:
|
---|
780 | p.set_line(label=keys[j])
|
---|
781 | else:
|
---|
782 | p.legend()
|
---|
783 | p.plot(precal[keys[j]][i])
|
---|
784 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
785 | if not btics[3*i]:
|
---|
786 | p.axes.set_xticks([])
|
---|
787 | p.subplot(3*i+1)
|
---|
788 | p.color=0
|
---|
789 | title='sig data IF%s POL%s' % (ifnos[2*int(i/len(polnos))],polnos[i%len(polnos)])
|
---|
790 | titles.append(title)
|
---|
791 | #p.set_axes('title',title)
|
---|
792 | p.legend()
|
---|
793 | ymin=postcal[2*i][edge:nchan-edge].min()
|
---|
794 | ymax=postcal[2*i][edge:nchan-edge].max()
|
---|
795 | p.plot(postcal[2*i])
|
---|
796 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
797 | if not btics[3*i+1]:
|
---|
798 | p.axes.set_xticks([])
|
---|
799 | p.subplot(3*i+2)
|
---|
800 | p.color=0
|
---|
801 | title='ref data IF%s POL%s' % (ifnos[2*int(i/len(polnos))+1],polnos[i%len(polnos)])
|
---|
802 | titles.append(title)
|
---|
803 | #p.set_axes('title',title)
|
---|
804 | p.legend()
|
---|
805 | ymin=postcal[2*i+1][edge:nchan-edge].min()
|
---|
806 | ymax=postcal[2*i+1][edge:nchan-edge].max()
|
---|
807 | p.plot(postcal[2*i+1])
|
---|
808 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
809 | if not btics[3*i+2]:
|
---|
810 | p.axes.set_xticks([])
|
---|
811 | for i in range(3*nr):
|
---|
812 | p.subplot(i)
|
---|
813 | p.set_axes('title',titles[i],fontsize='medium')
|
---|
814 | x=raw_input('Accept calibration ([y]/n): ' )
|
---|
815 | if x.upper() == 'N':
|
---|
816 | p.quit()
|
---|
817 | del p
|
---|
818 | return scantab
|
---|
819 | p.quit()
|
---|
820 | del p
|
---|
821 | ###
|
---|
822 | resspec._add_history("calfs",varlist)
|
---|
823 | return resspec
|
---|
824 |
|
---|
825 | @asaplog_post_dec
|
---|
826 | def merge(*args, **kwargs):
|
---|
827 | """
|
---|
828 | Merge a list of scanatables, or comma-sperated scantables into one
|
---|
829 | scnatble.
|
---|
830 | Parameters:
|
---|
831 | A list [scan1, scan2] or scan1, scan2.
|
---|
832 | freq_tol: frequency tolerance for merging IFs. numeric values
|
---|
833 | in units of Hz (1.0e6 -> 1MHz) and string ('1MHz')
|
---|
834 | is allowed.
|
---|
835 | Example:
|
---|
836 | myscans = [scan1, scan2]
|
---|
837 | allscans = merge(myscans)
|
---|
838 | # or equivalent
|
---|
839 | sameallscans = merge(scan1, scan2)
|
---|
840 | # with freqtol
|
---|
841 | allscans = merge(scan1, scan2, freq_tol=1.0e6)
|
---|
842 | # or equivalently
|
---|
843 | allscans = merge(scan1, scan2, freq_tol='1MHz')
|
---|
844 | """
|
---|
845 | varlist = vars()
|
---|
846 | if isinstance(args[0],list):
|
---|
847 | lst = tuple(args[0])
|
---|
848 | elif isinstance(args[0],tuple):
|
---|
849 | lst = args[0]
|
---|
850 | else:
|
---|
851 | lst = tuple(args)
|
---|
852 | if kwargs.has_key('freq_tol'):
|
---|
853 | freq_tol = str(kwargs['freq_tol'])
|
---|
854 | if len(freq_tol) > 0 and re.match('.+[GMk]Hz$', freq_tol) is None:
|
---|
855 | freq_tol += 'Hz'
|
---|
856 | else:
|
---|
857 | freq_tol = ''
|
---|
858 | varlist["args"] = "%d scantables" % len(lst)
|
---|
859 | # need special formatting her for history...
|
---|
860 | from asap._asap import stmath
|
---|
861 | stm = stmath()
|
---|
862 | for s in lst:
|
---|
863 | if not isinstance(s,scantable):
|
---|
864 | msg = "Please give a list of scantables"
|
---|
865 | raise TypeError(msg)
|
---|
866 | s = scantable(stm._merge(lst, freq_tol))
|
---|
867 | s._add_history("merge", varlist)
|
---|
868 | return s
|
---|
869 |
|
---|
870 | @asaplog_post_dec
|
---|
871 | def calibrate( scantab, scannos=[], calmode='none', verify=None ):
|
---|
872 | """
|
---|
873 | Calibrate data.
|
---|
874 |
|
---|
875 | Parameters:
|
---|
876 | scantab: scantable
|
---|
877 | scannos: list of scan number
|
---|
878 | calmode: calibration mode
|
---|
879 | verify: verify calibration
|
---|
880 | """
|
---|
881 | import re
|
---|
882 | antname = scantab.get_antennaname()
|
---|
883 | if ( calmode == 'nod' ):
|
---|
884 | asaplog.push( 'Calibrating nod data.' )
|
---|
885 | scal = calnod( scantab, scannos=scannos, verify=verify )
|
---|
886 | elif ( calmode == 'quotient' ):
|
---|
887 | asaplog.push( 'Calibrating using quotient.' )
|
---|
888 | scal = scantab.auto_quotient( verify=verify )
|
---|
889 | elif ( calmode == 'ps' ):
|
---|
890 | asaplog.push( 'Calibrating %s position-switched data.' % antname )
|
---|
891 | if ( antname.find( 'APEX' ) != -1 ):
|
---|
892 | scal = apexcal( scantab, scannos, calmode, verify )
|
---|
893 | elif ( antname.find( 'ALMA' ) != -1 or antname.find( 'OSF' ) != -1
|
---|
894 | or re.match('DV[0-9][0-9]$',antname) is not None
|
---|
895 | or re.match('PM[0-9][0-9]$',antname) is not None
|
---|
896 | or re.match('CM[0-9][0-9]$',antname) is not None
|
---|
897 | or re.match('DA[0-9][0-9]$',antname) is not None ):
|
---|
898 | scal = almacal( scantab, scannos, calmode, verify )
|
---|
899 | else:
|
---|
900 | scal = calps( scantab, scannos=scannos, verify=verify )
|
---|
901 | elif ( calmode == 'fs' or calmode == 'fsotf' ):
|
---|
902 | asaplog.push( 'Calibrating %s frequency-switched data.' % antname )
|
---|
903 | if ( antname.find( 'APEX' ) != -1 ):
|
---|
904 | scal = apexcal( scantab, scannos, calmode, verify )
|
---|
905 | elif ( antname.find( 'ALMA' ) != -1 or antname.find( 'OSF' ) != -1 ):
|
---|
906 | scal = almacal( scantab, scannos, calmode, verify )
|
---|
907 | else:
|
---|
908 | scal = calfs( scantab, scannos=scannos, verify=verify )
|
---|
909 | elif ( calmode == 'otf' ):
|
---|
910 | asaplog.push( 'Calibrating %s On-The-Fly data.' % antname )
|
---|
911 | scal = almacal( scantab, scannos, calmode, verify )
|
---|
912 | else:
|
---|
913 | asaplog.push( 'No calibration.' )
|
---|
914 | scal = scantab.copy()
|
---|
915 |
|
---|
916 | return scal
|
---|
917 |
|
---|
918 | def apexcal( scantab, scannos=[], calmode='none', verify=False ):
|
---|
919 | """
|
---|
920 | Calibrate APEX data
|
---|
921 |
|
---|
922 | Parameters:
|
---|
923 | scantab: scantable
|
---|
924 | scannos: list of scan number
|
---|
925 | calmode: calibration mode
|
---|
926 |
|
---|
927 | verify: verify calibration
|
---|
928 | """
|
---|
929 | from asap._asap import stmath
|
---|
930 | stm = stmath()
|
---|
931 | antname = scantab.get_antennaname()
|
---|
932 | selection=selector()
|
---|
933 | selection.set_scans(scannos)
|
---|
934 | orig = scantab.get_selection()
|
---|
935 | scantab.set_selection(orig+selection)
|
---|
936 | ## ssub = scantab.get_scan( scannos )
|
---|
937 | ## scal = scantable( stm.cwcal( ssub, calmode, antname ) )
|
---|
938 | scal = scantable( stm.cwcal( scantab, calmode, antname ) )
|
---|
939 | scantab.set_selection(orig)
|
---|
940 | return scal
|
---|
941 |
|
---|
942 | def almacal( scantab, scannos=[], calmode='none', verify=False ):
|
---|
943 | """
|
---|
944 | Calibrate ALMA data
|
---|
945 |
|
---|
946 | Parameters:
|
---|
947 | scantab: scantable
|
---|
948 | scannos: list of scan number
|
---|
949 | calmode: calibration mode
|
---|
950 |
|
---|
951 | verify: verify calibration
|
---|
952 | """
|
---|
953 | from asap._asap import stmath
|
---|
954 | stm = stmath()
|
---|
955 | selection=selector()
|
---|
956 | selection.set_scans(scannos)
|
---|
957 | orig = scantab.get_selection()
|
---|
958 | scantab.set_selection(orig+selection)
|
---|
959 | ## ssub = scantab.get_scan( scannos )
|
---|
960 | ## scal = scantable( stm.almacal( ssub, calmode ) )
|
---|
961 | scal = scantable( stm.almacal( scantab, calmode ) )
|
---|
962 | scantab.set_selection(orig)
|
---|
963 | return scal
|
---|
964 |
|
---|
965 | @asaplog_post_dec
|
---|
966 | def splitant(filename, outprefix='',overwrite=False, getpt=True):
|
---|
967 | """
|
---|
968 | Split Measurement set by antenna name, save data as a scantables,
|
---|
969 | and return a list of filename. Note that frequency reference frame
|
---|
970 | is imported as it is in Measurement set.
|
---|
971 | Notice this method can only be available from CASA.
|
---|
972 | Prameter
|
---|
973 | filename: the name of Measurement set to be read.
|
---|
974 | outprefix: the prefix of output scantable name.
|
---|
975 | the names of output scantable will be
|
---|
976 | outprefix.antenna1, outprefix.antenna2, ....
|
---|
977 | If not specified, outprefix = filename is assumed.
|
---|
978 | overwrite If the file should be overwritten if it exists.
|
---|
979 | The default False is to return with warning
|
---|
980 | without writing the output. USE WITH CARE.
|
---|
981 | getpt Whether to import direction from MS/POINTING
|
---|
982 | table or not. Default is True (import direction).
|
---|
983 | """
|
---|
984 | # Import the table toolkit from CASA
|
---|
985 | from taskinit import gentools
|
---|
986 | from asap.scantable import is_ms
|
---|
987 | tb = gentools(['tb'])[0]
|
---|
988 | # Check the input filename
|
---|
989 | if isinstance(filename, str):
|
---|
990 | import os.path
|
---|
991 | filename = os.path.expandvars(filename)
|
---|
992 | filename = os.path.expanduser(filename)
|
---|
993 | if not os.path.exists(filename):
|
---|
994 | s = "File '%s' not found." % (filename)
|
---|
995 | raise IOError(s)
|
---|
996 | # check if input file is MS
|
---|
997 | if not is_ms(filename):
|
---|
998 | s = "File '%s' is not a Measurement set." % (filename)
|
---|
999 | raise IOError(s)
|
---|
1000 | else:
|
---|
1001 | s = "The filename should be string. "
|
---|
1002 | raise TypeError(s)
|
---|
1003 | # Check out put file name
|
---|
1004 | outname=''
|
---|
1005 | if len(outprefix) > 0: prefix=outprefix+'.'
|
---|
1006 | else:
|
---|
1007 | prefix=filename.rstrip('/')
|
---|
1008 | # Now do the actual splitting.
|
---|
1009 | outfiles=[]
|
---|
1010 | tb.open(tablename=filename,nomodify=True)
|
---|
1011 | ant1=tb.getcol('ANTENNA1',0,-1,1)
|
---|
1012 | anttab=tb.getkeyword('ANTENNA').lstrip('Table: ')
|
---|
1013 | tb.close()
|
---|
1014 | tb.open(tablename=anttab,nomodify=True)
|
---|
1015 | nant=tb.nrows()
|
---|
1016 | antnames=tb.getcol('NAME',0,nant,1)
|
---|
1017 | tb.close()
|
---|
1018 | for antid in set(ant1):
|
---|
1019 | scan=scantable(filename,average=False,antenna=int(antid),getpt=getpt)
|
---|
1020 | outname=prefix+antnames[antid]+'.asap'
|
---|
1021 | scan.save(outname,format='ASAP',overwrite=overwrite)
|
---|
1022 | del scan
|
---|
1023 | outfiles.append(outname)
|
---|
1024 | return outfiles
|
---|
1025 |
|
---|
1026 | @asaplog_post_dec
|
---|
1027 | def _array2dOp( scan, value, mode="ADD", tsys=False, insitu=None):
|
---|
1028 | """
|
---|
1029 | This function is workaround on the basic operation of scantable
|
---|
1030 | with 2 dimensional float list.
|
---|
1031 |
|
---|
1032 | scan: scantable operand
|
---|
1033 | value: float list operand
|
---|
1034 | mode: operation mode (ADD, SUB, MUL, DIV)
|
---|
1035 | tsys: if True, operate tsys as well
|
---|
1036 | insitu: if False, a new scantable is returned.
|
---|
1037 | Otherwise, the array operation is done in-sitsu.
|
---|
1038 | """
|
---|
1039 | if insitu is None: insitu = rcParams['insitu']
|
---|
1040 | nrow = scan.nrow()
|
---|
1041 | s = None
|
---|
1042 | from asap._asap import stmath
|
---|
1043 | stm = stmath()
|
---|
1044 | stm._setinsitu(insitu)
|
---|
1045 | if len( value ) == 1:
|
---|
1046 | s = scantable( stm._arrayop( scan, value[0], mode, tsys ) )
|
---|
1047 | elif len( value ) != nrow:
|
---|
1048 | raise ValueError( 'len(value) must be 1 or conform to scan.nrow()' )
|
---|
1049 | else:
|
---|
1050 | from asap._asap import stmath
|
---|
1051 | if not insitu:
|
---|
1052 | s = scan.copy()
|
---|
1053 | else:
|
---|
1054 | s = scan
|
---|
1055 | # insitu must be True as we go row by row on the same data
|
---|
1056 | stm._setinsitu( True )
|
---|
1057 | basesel = s.get_selection()
|
---|
1058 | # generate a new selector object based on basesel
|
---|
1059 | sel = selector(basesel)
|
---|
1060 | for irow in range( nrow ):
|
---|
1061 | sel.set_rows( irow )
|
---|
1062 | s.set_selection( sel )
|
---|
1063 | if len( value[irow] ) == 1:
|
---|
1064 | stm._unaryop( s, value[irow][0], mode, tsys )
|
---|
1065 | else:
|
---|
1066 | #stm._arrayop( s, value[irow], mode, tsys, 'channel' )
|
---|
1067 | stm._arrayop( s, value[irow], mode, tsys )
|
---|
1068 | s.set_selection(basesel)
|
---|
1069 | return s
|
---|