[1918] | 1 | from asap.scantable import scantable
|
---|
[1827] | 2 | from asap.parameters import rcParams
|
---|
[1862] | 3 | from asap.logging import asaplog, asaplog_post_dec
|
---|
[1826] | 4 | from asap.selector import selector
|
---|
[1819] | 5 | from asap import asaplotgui
|
---|
[101] | 6 |
|
---|
[1862] | 7 | @asaplog_post_dec
|
---|
[143] | 8 | def average_time(*args, **kwargs):
|
---|
[101] | 9 | """
|
---|
[113] | 10 | Return the (time) average of a scan or list of scans. [in channels only]
|
---|
[305] | 11 | The cursor of the output scan is set to 0
|
---|
[113] | 12 | Parameters:
|
---|
[1361] | 13 | one scan or comma separated scans or a list of scans
|
---|
[143] | 14 | mask: an optional mask (only used for 'var' and 'tsys' weighting)
|
---|
[558] | 15 | scanav: True averages each scan separately.
|
---|
| 16 | False (default) averages all scans together,
|
---|
[1232] | 17 | weight: Weighting scheme.
|
---|
| 18 | 'none' (mean no weight)
|
---|
| 19 | 'var' (1/var(spec) weighted)
|
---|
| 20 | 'tsys' (1/Tsys**2 weighted)
|
---|
| 21 | 'tint' (integration time weighted)
|
---|
| 22 | 'tintsys' (Tint/Tsys**2)
|
---|
| 23 | 'median' ( median averaging)
|
---|
[930] | 24 | align: align the spectra in velocity before averaging. It takes
|
---|
| 25 | the time of the first spectrum in the first scantable
|
---|
| 26 | as reference time.
|
---|
[113] | 27 | Example:
|
---|
| 28 | # return a time averaged scan from scana and scanb
|
---|
| 29 | # without using a mask
|
---|
[129] | 30 | scanav = average_time(scana,scanb)
|
---|
[1589] | 31 | # or equivalent
|
---|
| 32 | # scanav = average_time([scana, scanb])
|
---|
[113] | 33 | # return the (time) averaged scan, i.e. the average of
|
---|
| 34 | # all correlator cycles
|
---|
[558] | 35 | scanav = average_time(scan, scanav=True)
|
---|
[101] | 36 | """
|
---|
[930] | 37 | scanav = False
|
---|
[143] | 38 | if kwargs.has_key('scanav'):
|
---|
[930] | 39 | scanav = kwargs.get('scanav')
|
---|
[524] | 40 | weight = 'tint'
|
---|
[143] | 41 | if kwargs.has_key('weight'):
|
---|
| 42 | weight = kwargs.get('weight')
|
---|
| 43 | mask = ()
|
---|
| 44 | if kwargs.has_key('mask'):
|
---|
| 45 | mask = kwargs.get('mask')
|
---|
[930] | 46 | align = False
|
---|
| 47 | if kwargs.has_key('align'):
|
---|
| 48 | align = kwargs.get('align')
|
---|
[1819] | 49 | compel = False
|
---|
| 50 | if kwargs.has_key('compel'):
|
---|
| 51 | compel = kwargs.get('compel')
|
---|
[489] | 52 | varlist = vars()
|
---|
[665] | 53 | if isinstance(args[0],list):
|
---|
[981] | 54 | lst = args[0]
|
---|
[665] | 55 | elif isinstance(args[0],tuple):
|
---|
[981] | 56 | lst = list(args[0])
|
---|
[665] | 57 | else:
|
---|
[981] | 58 | lst = list(args)
|
---|
[720] | 59 |
|
---|
[489] | 60 | del varlist["kwargs"]
|
---|
| 61 | varlist["args"] = "%d scantables" % len(lst)
|
---|
[981] | 62 | # need special formatting here for history...
|
---|
[720] | 63 |
|
---|
[876] | 64 | from asap._asap import stmath
|
---|
| 65 | stm = stmath()
|
---|
[113] | 66 | for s in lst:
|
---|
[101] | 67 | if not isinstance(s,scantable):
|
---|
[720] | 68 | msg = "Please give a list of scantables"
|
---|
[1859] | 69 | raise TypeError(msg)
|
---|
[945] | 70 | if scanav: scanav = "SCAN"
|
---|
| 71 | else: scanav = "NONE"
|
---|
[981] | 72 | alignedlst = []
|
---|
| 73 | if align:
|
---|
| 74 | refepoch = lst[0].get_time(0)
|
---|
| 75 | for scan in lst:
|
---|
| 76 | alignedlst.append(scan.freq_align(refepoch,insitu=False))
|
---|
| 77 | else:
|
---|
[1059] | 78 | alignedlst = lst
|
---|
[1232] | 79 | if weight.upper() == 'MEDIAN':
|
---|
| 80 | # median doesn't support list of scantables - merge first
|
---|
| 81 | merged = None
|
---|
| 82 | if len(alignedlst) > 1:
|
---|
| 83 | merged = merge(alignedlst)
|
---|
| 84 | else:
|
---|
| 85 | merged = alignedlst[0]
|
---|
| 86 | s = scantable(stm._averagechannel(merged, 'MEDIAN', scanav))
|
---|
| 87 | del merged
|
---|
| 88 | else:
|
---|
[1819] | 89 | #s = scantable(stm._average(alignedlst, mask, weight.upper(), scanav))
|
---|
| 90 | s = scantable(stm._new_average(alignedlst, compel, mask, weight.upper(), scanav))
|
---|
[489] | 91 | s._add_history("average_time",varlist)
|
---|
[1859] | 92 |
|
---|
[489] | 93 | return s
|
---|
[101] | 94 |
|
---|
[1862] | 95 | @asaplog_post_dec
|
---|
[1074] | 96 | def quotient(source, reference, preserve=True):
|
---|
| 97 | """
|
---|
| 98 | Return the quotient of a 'source' (signal) scan and a 'reference' scan.
|
---|
| 99 | The reference can have just one scan, even if the signal has many. Otherwise
|
---|
| 100 | they must have the same number of scans.
|
---|
| 101 | The cursor of the output scan is set to 0
|
---|
| 102 | Parameters:
|
---|
| 103 | source: the 'on' scan
|
---|
| 104 | reference: the 'off' scan
|
---|
| 105 | preserve: you can preserve (default) the continuum or
|
---|
| 106 | remove it. The equations used are
|
---|
| 107 | preserve: Output = Toff * (on/off) - Toff
|
---|
| 108 | remove: Output = Toff * (on/off) - Ton
|
---|
| 109 | """
|
---|
| 110 | varlist = vars()
|
---|
| 111 | from asap._asap import stmath
|
---|
| 112 | stm = stmath()
|
---|
| 113 | stm._setinsitu(False)
|
---|
| 114 | s = scantable(stm._quotient(source, reference, preserve))
|
---|
| 115 | s._add_history("quotient",varlist)
|
---|
| 116 | return s
|
---|
[101] | 117 |
|
---|
[1862] | 118 | @asaplog_post_dec
|
---|
[1391] | 119 | def dototalpower(calon, caloff, tcalval=0.0):
|
---|
| 120 | """
|
---|
| 121 | Do calibration for CAL on,off signals.
|
---|
| 122 | Adopted from GBTIDL dototalpower
|
---|
| 123 | Parameters:
|
---|
| 124 | calon: the 'cal on' subintegration
|
---|
| 125 | caloff: the 'cal off' subintegration
|
---|
| 126 | tcalval: user supplied Tcal value
|
---|
| 127 | """
|
---|
| 128 | varlist = vars()
|
---|
| 129 | from asap._asap import stmath
|
---|
| 130 | stm = stmath()
|
---|
| 131 | stm._setinsitu(False)
|
---|
| 132 | s = scantable(stm._dototalpower(calon, caloff, tcalval))
|
---|
| 133 | s._add_history("dototalpower",varlist)
|
---|
| 134 | return s
|
---|
| 135 |
|
---|
[1862] | 136 | @asaplog_post_dec
|
---|
[1391] | 137 | def dosigref(sig, ref, smooth, tsysval=0.0, tauval=0.0):
|
---|
| 138 | """
|
---|
| 139 | Calculate a quotient (sig-ref/ref * Tsys)
|
---|
| 140 | Adopted from GBTIDL dosigref
|
---|
| 141 | Parameters:
|
---|
| 142 | sig: on source data
|
---|
| 143 | ref: reference data
|
---|
| 144 | smooth: width of box car smoothing for reference
|
---|
| 145 | tsysval: user specified Tsys (scalar only)
|
---|
| 146 | tauval: user specified Tau (required if tsysval is set)
|
---|
| 147 | """
|
---|
| 148 | varlist = vars()
|
---|
| 149 | from asap._asap import stmath
|
---|
| 150 | stm = stmath()
|
---|
| 151 | stm._setinsitu(False)
|
---|
| 152 | s = scantable(stm._dosigref(sig, ref, smooth, tsysval, tauval))
|
---|
| 153 | s._add_history("dosigref",varlist)
|
---|
| 154 | return s
|
---|
| 155 |
|
---|
[1862] | 156 | @asaplog_post_dec
|
---|
[1819] | 157 | def calps(scantab, scannos, smooth=1, tsysval=0.0, tauval=0.0, tcalval=0.0, verify=False):
|
---|
[1391] | 158 | """
|
---|
| 159 | Calibrate GBT position switched data
|
---|
| 160 | Adopted from GBTIDL getps
|
---|
[1920] | 161 | Currently calps identify the scans as position switched data if source
|
---|
| 162 | type enum is pson or psoff. The data must contains 'CAL' signal
|
---|
| 163 | on/off in each integration. To identify 'CAL' on state, the source type
|
---|
| 164 | enum of poncal and poffcal need to be present in the source name field.
|
---|
[1391] | 165 | (GBT MS data reading process to scantable automatically append these
|
---|
| 166 | id names to the source names)
|
---|
| 167 |
|
---|
| 168 | Parameters:
|
---|
| 169 | scantab: scantable
|
---|
| 170 | scannos: list of scan numbers
|
---|
| 171 | smooth: optional box smoothing order for the reference
|
---|
| 172 | (default is 1 = no smoothing)
|
---|
| 173 | tsysval: optional user specified Tsys (default is 0.0,
|
---|
| 174 | use Tsys in the data)
|
---|
| 175 | tauval: optional user specified Tau
|
---|
| 176 | tcalval: optional user specified Tcal (default is 0.0,
|
---|
| 177 | use Tcal value in the data)
|
---|
[1920] | 178 | verify: Verify calibration if true
|
---|
[1391] | 179 | """
|
---|
| 180 | varlist = vars()
|
---|
| 181 | # check for the appropriate data
|
---|
[1819] | 182 | ## s = scantab.get_scan('*_ps*')
|
---|
| 183 | ## if s is None:
|
---|
| 184 | ## msg = "The input data appear to contain no position-switch mode data."
|
---|
[1859] | 185 | ## raise TypeError(msg)
|
---|
[1819] | 186 | s = scantab.copy()
|
---|
| 187 | from asap._asap import srctype
|
---|
| 188 | sel = selector()
|
---|
| 189 | sel.set_types( srctype.pson )
|
---|
| 190 | try:
|
---|
| 191 | scantab.set_selection( sel )
|
---|
| 192 | except Exception, e:
|
---|
[1391] | 193 | msg = "The input data appear to contain no position-switch mode data."
|
---|
[1859] | 194 | raise TypeError(msg)
|
---|
[1819] | 195 | s.set_selection()
|
---|
| 196 | sel.reset()
|
---|
[1391] | 197 | ssub = s.get_scan(scannos)
|
---|
| 198 | if ssub is None:
|
---|
| 199 | msg = "No data was found with given scan numbers!"
|
---|
[1859] | 200 | raise TypeError(msg)
|
---|
[1819] | 201 | #ssubon = ssub.get_scan('*calon')
|
---|
| 202 | #ssuboff = ssub.get_scan('*[^calon]')
|
---|
| 203 | sel.set_types( [srctype.poncal,srctype.poffcal] )
|
---|
| 204 | ssub.set_selection( sel )
|
---|
| 205 | ssubon = ssub.copy()
|
---|
| 206 | ssub.set_selection()
|
---|
| 207 | sel.reset()
|
---|
| 208 | sel.set_types( [srctype.pson,srctype.psoff] )
|
---|
| 209 | ssub.set_selection( sel )
|
---|
| 210 | ssuboff = ssub.copy()
|
---|
| 211 | ssub.set_selection()
|
---|
| 212 | sel.reset()
|
---|
[1391] | 213 | if ssubon.nrow() != ssuboff.nrow():
|
---|
| 214 | msg = "mismatch in numbers of CAL on/off scans. Cannot calibrate. Check the scan numbers."
|
---|
[1859] | 215 | raise TypeError(msg)
|
---|
[1391] | 216 | cals = dototalpower(ssubon, ssuboff, tcalval)
|
---|
[1819] | 217 | #sig = cals.get_scan('*ps')
|
---|
| 218 | #ref = cals.get_scan('*psr')
|
---|
| 219 | sel.set_types( srctype.pson )
|
---|
| 220 | cals.set_selection( sel )
|
---|
| 221 | sig = cals.copy()
|
---|
| 222 | cals.set_selection()
|
---|
| 223 | sel.reset()
|
---|
| 224 | sel.set_types( srctype.psoff )
|
---|
| 225 | cals.set_selection( sel )
|
---|
| 226 | ref = cals.copy()
|
---|
| 227 | cals.set_selection()
|
---|
| 228 | sel.reset()
|
---|
[1391] | 229 | if sig.nscan() != ref.nscan():
|
---|
| 230 | msg = "mismatch in numbers of on/off scans. Cannot calibrate. Check the scan numbers."
|
---|
[1859] | 231 | raise TypeError(msg)
|
---|
[1391] | 232 |
|
---|
| 233 | #for user supplied Tsys
|
---|
| 234 | if tsysval>0.0:
|
---|
| 235 | if tauval<=0.0:
|
---|
| 236 | msg = "Need to supply a valid tau to use the supplied Tsys"
|
---|
[1859] | 237 | raise TypeError(msg)
|
---|
[1391] | 238 | else:
|
---|
| 239 | sig.recalc_azel()
|
---|
| 240 | ref.recalc_azel()
|
---|
| 241 | #msg = "Use of user specified Tsys is not fully implemented yet."
|
---|
[1859] | 242 | #raise TypeError(msg)
|
---|
[1391] | 243 | # use get_elevation to get elevation and
|
---|
| 244 | # calculate a scaling factor using the formula
|
---|
| 245 | # -> tsys use to dosigref
|
---|
| 246 |
|
---|
| 247 | #ress = dosigref(sig, ref, smooth, tsysval)
|
---|
| 248 | ress = dosigref(sig, ref, smooth, tsysval, tauval)
|
---|
[1819] | 249 | ###
|
---|
| 250 | if verify:
|
---|
| 251 | # get data
|
---|
| 252 | import numpy
|
---|
| 253 | precal={}
|
---|
| 254 | postcal=[]
|
---|
| 255 | keys=['ps','ps_calon','psr','psr_calon']
|
---|
| 256 | types=[srctype.pson,srctype.poncal,srctype.psoff,srctype.poffcal]
|
---|
| 257 | ifnos=list(ssub.getifnos())
|
---|
| 258 | polnos=list(ssub.getpolnos())
|
---|
| 259 | sel=selector()
|
---|
| 260 | for i in range(2):
|
---|
| 261 | #ss=ssuboff.get_scan('*'+keys[2*i])
|
---|
| 262 | ll=[]
|
---|
| 263 | for j in range(len(ifnos)):
|
---|
| 264 | for k in range(len(polnos)):
|
---|
| 265 | sel.set_ifs(ifnos[j])
|
---|
| 266 | sel.set_polarizations(polnos[k])
|
---|
| 267 | sel.set_types(types[2*i])
|
---|
| 268 | try:
|
---|
| 269 | #ss.set_selection(sel)
|
---|
| 270 | ssuboff.set_selection(sel)
|
---|
| 271 | except:
|
---|
| 272 | continue
|
---|
| 273 | #ll.append(numpy.array(ss._getspectrum(0)))
|
---|
| 274 | ll.append(numpy.array(ssuboff._getspectrum(0)))
|
---|
| 275 | sel.reset()
|
---|
| 276 | ssuboff.set_selection()
|
---|
| 277 | precal[keys[2*i]]=ll
|
---|
| 278 | #del ss
|
---|
| 279 | #ss=ssubon.get_scan('*'+keys[2*i+1])
|
---|
| 280 | ll=[]
|
---|
| 281 | for j in range(len(ifnos)):
|
---|
| 282 | for k in range(len(polnos)):
|
---|
| 283 | sel.set_ifs(ifnos[j])
|
---|
| 284 | sel.set_polarizations(polnos[k])
|
---|
| 285 | sel.set_types(types[2*i+1])
|
---|
| 286 | try:
|
---|
| 287 | #ss.set_selection(sel)
|
---|
| 288 | ssubon.set_selection(sel)
|
---|
| 289 | except:
|
---|
| 290 | continue
|
---|
| 291 | #ll.append(numpy.array(ss._getspectrum(0)))
|
---|
| 292 | ll.append(numpy.array(ssubon._getspectrum(0)))
|
---|
| 293 | sel.reset()
|
---|
| 294 | ssubon.set_selection()
|
---|
| 295 | precal[keys[2*i+1]]=ll
|
---|
| 296 | #del ss
|
---|
| 297 | for j in range(len(ifnos)):
|
---|
| 298 | for k in range(len(polnos)):
|
---|
| 299 | sel.set_ifs(ifnos[j])
|
---|
| 300 | sel.set_polarizations(polnos[k])
|
---|
| 301 | try:
|
---|
| 302 | ress.set_selection(sel)
|
---|
| 303 | except:
|
---|
| 304 | continue
|
---|
| 305 | postcal.append(numpy.array(ress._getspectrum(0)))
|
---|
| 306 | sel.reset()
|
---|
| 307 | ress.set_selection()
|
---|
| 308 | del sel
|
---|
| 309 | # plot
|
---|
[1861] | 310 | asaplog.post()
|
---|
[1819] | 311 | asaplog.push('Plot only first spectrum for each [if,pol] pairs to verify calibration.')
|
---|
[1861] | 312 | asaplog.post('WARN')
|
---|
[1819] | 313 | p=asaplotgui.asaplotgui()
|
---|
| 314 | #nr=min(6,len(ifnos)*len(polnos))
|
---|
| 315 | nr=len(ifnos)*len(polnos)
|
---|
| 316 | titles=[]
|
---|
| 317 | btics=[]
|
---|
| 318 | if nr<4:
|
---|
| 319 | p.set_panels(rows=nr,cols=2,nplots=2*nr,ganged=False)
|
---|
| 320 | for i in range(2*nr):
|
---|
| 321 | b=False
|
---|
| 322 | if i >= 2*nr-2:
|
---|
| 323 | b=True
|
---|
| 324 | btics.append(b)
|
---|
| 325 | elif nr==4:
|
---|
| 326 | p.set_panels(rows=2,cols=4,nplots=8,ganged=False)
|
---|
| 327 | for i in range(2*nr):
|
---|
| 328 | b=False
|
---|
| 329 | if i >= 2*nr-4:
|
---|
| 330 | b=True
|
---|
| 331 | btics.append(b)
|
---|
| 332 | elif nr<7:
|
---|
| 333 | p.set_panels(rows=3,cols=4,nplots=2*nr,ganged=False)
|
---|
| 334 | for i in range(2*nr):
|
---|
| 335 | if i >= 2*nr-4:
|
---|
| 336 | b=True
|
---|
| 337 | btics.append(b)
|
---|
| 338 | else:
|
---|
[1861] | 339 | asaplog.post()
|
---|
[1819] | 340 | asaplog.push('Only first 6 [if,pol] pairs are plotted.')
|
---|
[1861] | 341 | asaplog.post('WARN')
|
---|
[1819] | 342 | nr=6
|
---|
| 343 | for i in range(2*nr):
|
---|
| 344 | b=False
|
---|
| 345 | if i >= 2*nr-4:
|
---|
| 346 | b=True
|
---|
| 347 | btics.append(b)
|
---|
| 348 | p.set_panels(rows=3,cols=4,nplots=2*nr,ganged=False)
|
---|
| 349 | for i in range(nr):
|
---|
| 350 | p.subplot(2*i)
|
---|
| 351 | p.color=0
|
---|
| 352 | title='raw data IF%s POL%s' % (ifnos[int(i/len(polnos))],polnos[i%len(polnos)])
|
---|
| 353 | titles.append(title)
|
---|
| 354 | #p.set_axes('title',title,fontsize=40)
|
---|
| 355 | ymin=1.0e100
|
---|
| 356 | ymax=-1.0e100
|
---|
| 357 | nchan=s.nchan()
|
---|
| 358 | edge=int(nchan*0.01)
|
---|
| 359 | for j in range(4):
|
---|
| 360 | spmin=min(precal[keys[j]][i][edge:nchan-edge])
|
---|
| 361 | spmax=max(precal[keys[j]][i][edge:nchan-edge])
|
---|
| 362 | ymin=min(ymin,spmin)
|
---|
| 363 | ymax=max(ymax,spmax)
|
---|
| 364 | for j in range(4):
|
---|
| 365 | if i==0:
|
---|
| 366 | p.set_line(label=keys[j])
|
---|
| 367 | else:
|
---|
| 368 | p.legend()
|
---|
| 369 | p.plot(precal[keys[j]][i])
|
---|
| 370 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
| 371 | if not btics[2*i]:
|
---|
| 372 | p.axes.set_xticks([])
|
---|
| 373 | p.subplot(2*i+1)
|
---|
| 374 | p.color=0
|
---|
| 375 | title='cal data IF%s POL%s' % (ifnos[int(i/len(polnos))],polnos[i%len(polnos)])
|
---|
| 376 | titles.append(title)
|
---|
| 377 | #p.set_axes('title',title)
|
---|
| 378 | p.legend()
|
---|
| 379 | ymin=postcal[i][edge:nchan-edge].min()
|
---|
| 380 | ymax=postcal[i][edge:nchan-edge].max()
|
---|
| 381 | p.plot(postcal[i])
|
---|
| 382 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
| 383 | if not btics[2*i+1]:
|
---|
| 384 | p.axes.set_xticks([])
|
---|
| 385 | for i in range(2*nr):
|
---|
| 386 | p.subplot(i)
|
---|
| 387 | p.set_axes('title',titles[i],fontsize='medium')
|
---|
| 388 | x=raw_input('Accept calibration ([y]/n): ' )
|
---|
| 389 | if x.upper() == 'N':
|
---|
| 390 | p.unmap()
|
---|
| 391 | del p
|
---|
| 392 | return scabtab
|
---|
| 393 | p.unmap()
|
---|
| 394 | del p
|
---|
| 395 | ###
|
---|
[1391] | 396 | ress._add_history("calps", varlist)
|
---|
| 397 | return ress
|
---|
| 398 |
|
---|
[1862] | 399 | @asaplog_post_dec
|
---|
[1819] | 400 | def calnod(scantab, scannos=[], smooth=1, tsysval=0.0, tauval=0.0, tcalval=0.0, verify=False):
|
---|
[1391] | 401 | """
|
---|
| 402 | Do full (but a pair of scans at time) processing of GBT Nod data
|
---|
| 403 | calibration.
|
---|
| 404 | Adopted from GBTIDL's getnod
|
---|
| 405 | Parameters:
|
---|
| 406 | scantab: scantable
|
---|
| 407 | scannos: a pair of scan numbers, or the first scan number of the pair
|
---|
| 408 | smooth: box car smoothing order
|
---|
| 409 | tsysval: optional user specified Tsys value
|
---|
| 410 | tauval: optional user specified tau value (not implemented yet)
|
---|
| 411 | tcalval: optional user specified Tcal value
|
---|
[1920] | 412 | verify: Verify calibration if true
|
---|
[1391] | 413 | """
|
---|
| 414 | varlist = vars()
|
---|
| 415 | from asap._asap import stmath
|
---|
[1819] | 416 | from asap._asap import srctype
|
---|
[1391] | 417 | stm = stmath()
|
---|
| 418 | stm._setinsitu(False)
|
---|
| 419 |
|
---|
| 420 | # check for the appropriate data
|
---|
[1819] | 421 | ## s = scantab.get_scan('*_nod*')
|
---|
| 422 | ## if s is None:
|
---|
| 423 | ## msg = "The input data appear to contain no Nod observing mode data."
|
---|
[1859] | 424 | ## raise TypeError(msg)
|
---|
[1819] | 425 | s = scantab.copy()
|
---|
| 426 | sel = selector()
|
---|
| 427 | sel.set_types( srctype.nod )
|
---|
| 428 | try:
|
---|
| 429 | s.set_selection( sel )
|
---|
| 430 | except Exception, e:
|
---|
[1391] | 431 | msg = "The input data appear to contain no Nod observing mode data."
|
---|
[1859] | 432 | raise TypeError(msg)
|
---|
[1819] | 433 | sel.reset()
|
---|
| 434 | del sel
|
---|
| 435 | del s
|
---|
[1391] | 436 |
|
---|
| 437 | # need check correspondance of each beam with sig-ref ...
|
---|
| 438 | # check for timestamps, scan numbers, subscan id (not available in
|
---|
| 439 | # ASAP data format...). Assume 1st scan of the pair (beam 0 - sig
|
---|
| 440 | # and beam 1 - ref...)
|
---|
| 441 | # First scan number of paired scans or list of pairs of
|
---|
| 442 | # scan numbers (has to have even number of pairs.)
|
---|
| 443 |
|
---|
| 444 | #data splitting
|
---|
| 445 | scan1no = scan2no = 0
|
---|
| 446 |
|
---|
| 447 | if len(scannos)==1:
|
---|
| 448 | scan1no = scannos[0]
|
---|
| 449 | scan2no = scannos[0]+1
|
---|
| 450 | pairScans = [scan1no, scan2no]
|
---|
| 451 | else:
|
---|
| 452 | #if len(scannos)>2:
|
---|
| 453 | # msg = "calnod can only process a pair of nod scans at time."
|
---|
[1859] | 454 | # raise TypeError(msg)
|
---|
[1391] | 455 | #
|
---|
| 456 | #if len(scannos)==2:
|
---|
| 457 | # scan1no = scannos[0]
|
---|
| 458 | # scan2no = scannos[1]
|
---|
| 459 | pairScans = list(scannos)
|
---|
| 460 |
|
---|
| 461 | if tsysval>0.0:
|
---|
| 462 | if tauval<=0.0:
|
---|
| 463 | msg = "Need to supply a valid tau to use the supplied Tsys"
|
---|
[1859] | 464 | raise TypeError(msg)
|
---|
[1391] | 465 | else:
|
---|
| 466 | scantab.recalc_azel()
|
---|
| 467 | resspec = scantable(stm._donod(scantab, pairScans, smooth, tsysval,tauval,tcalval))
|
---|
[1819] | 468 | ###
|
---|
| 469 | if verify:
|
---|
| 470 | # get data
|
---|
| 471 | import numpy
|
---|
| 472 | precal={}
|
---|
| 473 | postcal=[]
|
---|
| 474 | keys=['','_calon']
|
---|
| 475 | types=[srctype.nod,srctype.nodcal]
|
---|
| 476 | ifnos=list(scantab.getifnos())
|
---|
| 477 | polnos=list(scantab.getpolnos())
|
---|
| 478 | sel=selector()
|
---|
| 479 | ss = scantab.copy()
|
---|
| 480 | for i in range(2):
|
---|
| 481 | #ss=scantab.get_scan('*'+keys[i])
|
---|
| 482 | ll=[]
|
---|
| 483 | ll2=[]
|
---|
| 484 | for j in range(len(ifnos)):
|
---|
| 485 | for k in range(len(polnos)):
|
---|
| 486 | sel.set_ifs(ifnos[j])
|
---|
| 487 | sel.set_polarizations(polnos[k])
|
---|
| 488 | sel.set_scans(pairScans[0])
|
---|
| 489 | sel.set_types(types[i])
|
---|
| 490 | try:
|
---|
| 491 | ss.set_selection(sel)
|
---|
| 492 | except:
|
---|
| 493 | continue
|
---|
| 494 | ll.append(numpy.array(ss._getspectrum(0)))
|
---|
| 495 | sel.reset()
|
---|
| 496 | ss.set_selection()
|
---|
| 497 | sel.set_ifs(ifnos[j])
|
---|
| 498 | sel.set_polarizations(polnos[k])
|
---|
| 499 | sel.set_scans(pairScans[1])
|
---|
| 500 | sel.set_types(types[i])
|
---|
| 501 | try:
|
---|
| 502 | ss.set_selection(sel)
|
---|
| 503 | except:
|
---|
| 504 | ll.pop()
|
---|
| 505 | continue
|
---|
| 506 | ll2.append(numpy.array(ss._getspectrum(0)))
|
---|
| 507 | sel.reset()
|
---|
| 508 | ss.set_selection()
|
---|
| 509 | key='%s%s' %(pairScans[0],keys[i])
|
---|
| 510 | precal[key]=ll
|
---|
| 511 | key='%s%s' %(pairScans[1],keys[i])
|
---|
| 512 | precal[key]=ll2
|
---|
| 513 | #del ss
|
---|
| 514 | keys=precal.keys()
|
---|
| 515 | for j in range(len(ifnos)):
|
---|
| 516 | for k in range(len(polnos)):
|
---|
| 517 | sel.set_ifs(ifnos[j])
|
---|
| 518 | sel.set_polarizations(polnos[k])
|
---|
| 519 | sel.set_scans(pairScans[0])
|
---|
| 520 | try:
|
---|
| 521 | resspec.set_selection(sel)
|
---|
| 522 | except:
|
---|
| 523 | continue
|
---|
| 524 | postcal.append(numpy.array(resspec._getspectrum(0)))
|
---|
| 525 | sel.reset()
|
---|
| 526 | resspec.set_selection()
|
---|
| 527 | del sel
|
---|
| 528 | # plot
|
---|
[1861] | 529 | asaplog.post()
|
---|
[1819] | 530 | asaplog.push('Plot only first spectrum for each [if,pol] pairs to verify calibration.')
|
---|
[1861] | 531 | asaplog.post('WARN')
|
---|
[1819] | 532 | p=asaplotgui.asaplotgui()
|
---|
| 533 | #nr=min(6,len(ifnos)*len(polnos))
|
---|
| 534 | nr=len(ifnos)*len(polnos)
|
---|
| 535 | titles=[]
|
---|
| 536 | btics=[]
|
---|
| 537 | if nr<4:
|
---|
| 538 | p.set_panels(rows=nr,cols=2,nplots=2*nr,ganged=False)
|
---|
| 539 | for i in range(2*nr):
|
---|
| 540 | b=False
|
---|
| 541 | if i >= 2*nr-2:
|
---|
| 542 | b=True
|
---|
| 543 | btics.append(b)
|
---|
| 544 | elif nr==4:
|
---|
| 545 | p.set_panels(rows=2,cols=4,nplots=8,ganged=False)
|
---|
| 546 | for i in range(2*nr):
|
---|
| 547 | b=False
|
---|
| 548 | if i >= 2*nr-4:
|
---|
| 549 | b=True
|
---|
| 550 | btics.append(b)
|
---|
| 551 | elif nr<7:
|
---|
| 552 | p.set_panels(rows=3,cols=4,nplots=2*nr,ganged=False)
|
---|
| 553 | for i in range(2*nr):
|
---|
| 554 | if i >= 2*nr-4:
|
---|
| 555 | b=True
|
---|
| 556 | btics.append(b)
|
---|
| 557 | else:
|
---|
[1861] | 558 | asaplog.post()
|
---|
[1819] | 559 | asaplog.push('Only first 6 [if,pol] pairs are plotted.')
|
---|
[1861] | 560 | asaplog.post('WARN')
|
---|
[1819] | 561 | nr=6
|
---|
| 562 | for i in range(2*nr):
|
---|
| 563 | b=False
|
---|
| 564 | if i >= 2*nr-4:
|
---|
| 565 | b=True
|
---|
| 566 | btics.append(b)
|
---|
| 567 | p.set_panels(rows=3,cols=4,nplots=2*nr,ganged=False)
|
---|
| 568 | for i in range(nr):
|
---|
| 569 | p.subplot(2*i)
|
---|
| 570 | p.color=0
|
---|
| 571 | title='raw data IF%s POL%s' % (ifnos[int(i/len(polnos))],polnos[i%len(polnos)])
|
---|
| 572 | titles.append(title)
|
---|
| 573 | #p.set_axes('title',title,fontsize=40)
|
---|
| 574 | ymin=1.0e100
|
---|
| 575 | ymax=-1.0e100
|
---|
| 576 | nchan=scantab.nchan()
|
---|
| 577 | edge=int(nchan*0.01)
|
---|
| 578 | for j in range(4):
|
---|
| 579 | spmin=min(precal[keys[j]][i][edge:nchan-edge])
|
---|
| 580 | spmax=max(precal[keys[j]][i][edge:nchan-edge])
|
---|
| 581 | ymin=min(ymin,spmin)
|
---|
| 582 | ymax=max(ymax,spmax)
|
---|
| 583 | for j in range(4):
|
---|
| 584 | if i==0:
|
---|
| 585 | p.set_line(label=keys[j])
|
---|
| 586 | else:
|
---|
| 587 | p.legend()
|
---|
| 588 | p.plot(precal[keys[j]][i])
|
---|
| 589 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
| 590 | if not btics[2*i]:
|
---|
| 591 | p.axes.set_xticks([])
|
---|
| 592 | p.subplot(2*i+1)
|
---|
| 593 | p.color=0
|
---|
| 594 | title='cal data IF%s POL%s' % (ifnos[int(i/len(polnos))],polnos[i%len(polnos)])
|
---|
| 595 | titles.append(title)
|
---|
| 596 | #p.set_axes('title',title)
|
---|
| 597 | p.legend()
|
---|
| 598 | ymin=postcal[i][edge:nchan-edge].min()
|
---|
| 599 | ymax=postcal[i][edge:nchan-edge].max()
|
---|
| 600 | p.plot(postcal[i])
|
---|
| 601 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
| 602 | if not btics[2*i+1]:
|
---|
| 603 | p.axes.set_xticks([])
|
---|
| 604 | for i in range(2*nr):
|
---|
| 605 | p.subplot(i)
|
---|
| 606 | p.set_axes('title',titles[i],fontsize='medium')
|
---|
| 607 | x=raw_input('Accept calibration ([y]/n): ' )
|
---|
| 608 | if x.upper() == 'N':
|
---|
| 609 | p.unmap()
|
---|
| 610 | del p
|
---|
| 611 | return scabtab
|
---|
| 612 | p.unmap()
|
---|
| 613 | del p
|
---|
| 614 | ###
|
---|
[1391] | 615 | resspec._add_history("calnod",varlist)
|
---|
| 616 | return resspec
|
---|
| 617 |
|
---|
[1862] | 618 | @asaplog_post_dec
|
---|
[1819] | 619 | def calfs(scantab, scannos=[], smooth=1, tsysval=0.0, tauval=0.0, tcalval=0.0, verify=False):
|
---|
[1391] | 620 | """
|
---|
| 621 | Calibrate GBT frequency switched data.
|
---|
| 622 | Adopted from GBTIDL getfs.
|
---|
[1920] | 623 | Currently calfs identify the scans as frequency switched data if source
|
---|
| 624 | type enum is fson and fsoff. The data must contains 'CAL' signal
|
---|
| 625 | on/off in each integration. To identify 'CAL' on state, the source type
|
---|
| 626 | enum of foncal and foffcal need to be present in the source name field.
|
---|
[1391] | 627 | (GBT MS data reading via scantable automatically append these
|
---|
| 628 | id names to the source names)
|
---|
| 629 |
|
---|
| 630 | Parameters:
|
---|
| 631 | scantab: scantable
|
---|
| 632 | scannos: list of scan numbers
|
---|
| 633 | smooth: optional box smoothing order for the reference
|
---|
| 634 | (default is 1 = no smoothing)
|
---|
| 635 | tsysval: optional user specified Tsys (default is 0.0,
|
---|
| 636 | use Tsys in the data)
|
---|
| 637 | tauval: optional user specified Tau
|
---|
[1920] | 638 | verify: Verify calibration if true
|
---|
[1391] | 639 | """
|
---|
| 640 | varlist = vars()
|
---|
| 641 | from asap._asap import stmath
|
---|
[1819] | 642 | from asap._asap import srctype
|
---|
[1391] | 643 | stm = stmath()
|
---|
| 644 | stm._setinsitu(False)
|
---|
| 645 |
|
---|
| 646 | # check = scantab.get_scan('*_fs*')
|
---|
| 647 | # if check is None:
|
---|
| 648 | # msg = "The input data appear to contain no Nod observing mode data."
|
---|
[1859] | 649 | # raise TypeError(msg)
|
---|
[1391] | 650 | s = scantab.get_scan(scannos)
|
---|
| 651 | del scantab
|
---|
| 652 |
|
---|
| 653 | resspec = scantable(stm._dofs(s, scannos, smooth, tsysval,tauval,tcalval))
|
---|
[1819] | 654 | ###
|
---|
| 655 | if verify:
|
---|
| 656 | # get data
|
---|
| 657 | ssub = s.get_scan(scannos)
|
---|
| 658 | #ssubon = ssub.get_scan('*calon')
|
---|
| 659 | #ssuboff = ssub.get_scan('*[^calon]')
|
---|
| 660 | sel = selector()
|
---|
| 661 | sel.set_types( [srctype.foncal,srctype.foffcal] )
|
---|
| 662 | ssub.set_selection( sel )
|
---|
| 663 | ssubon = ssub.copy()
|
---|
| 664 | ssub.set_selection()
|
---|
| 665 | sel.reset()
|
---|
| 666 | sel.set_types( [srctype.fson,srctype.fsoff] )
|
---|
| 667 | ssub.set_selection( sel )
|
---|
| 668 | ssuboff = ssub.copy()
|
---|
| 669 | ssub.set_selection()
|
---|
| 670 | sel.reset()
|
---|
| 671 | import numpy
|
---|
| 672 | precal={}
|
---|
| 673 | postcal=[]
|
---|
| 674 | keys=['fs','fs_calon','fsr','fsr_calon']
|
---|
| 675 | types=[srctype.fson,srctype.foncal,srctype.fsoff,srctype.foffcal]
|
---|
| 676 | ifnos=list(ssub.getifnos())
|
---|
| 677 | polnos=list(ssub.getpolnos())
|
---|
| 678 | for i in range(2):
|
---|
| 679 | #ss=ssuboff.get_scan('*'+keys[2*i])
|
---|
| 680 | ll=[]
|
---|
| 681 | for j in range(len(ifnos)):
|
---|
| 682 | for k in range(len(polnos)):
|
---|
| 683 | sel.set_ifs(ifnos[j])
|
---|
| 684 | sel.set_polarizations(polnos[k])
|
---|
| 685 | sel.set_types(types[2*i])
|
---|
| 686 | try:
|
---|
| 687 | #ss.set_selection(sel)
|
---|
| 688 | ssuboff.set_selection(sel)
|
---|
| 689 | except:
|
---|
| 690 | continue
|
---|
| 691 | ll.append(numpy.array(ss._getspectrum(0)))
|
---|
| 692 | sel.reset()
|
---|
| 693 | #ss.set_selection()
|
---|
| 694 | ssuboff.set_selection()
|
---|
| 695 | precal[keys[2*i]]=ll
|
---|
| 696 | #del ss
|
---|
| 697 | #ss=ssubon.get_scan('*'+keys[2*i+1])
|
---|
| 698 | ll=[]
|
---|
| 699 | for j in range(len(ifnos)):
|
---|
| 700 | for k in range(len(polnos)):
|
---|
| 701 | sel.set_ifs(ifnos[j])
|
---|
| 702 | sel.set_polarizations(polnos[k])
|
---|
| 703 | sel.set_types(types[2*i+1])
|
---|
| 704 | try:
|
---|
| 705 | #ss.set_selection(sel)
|
---|
| 706 | ssubon.set_selection(sel)
|
---|
| 707 | except:
|
---|
| 708 | continue
|
---|
| 709 | ll.append(numpy.array(ss._getspectrum(0)))
|
---|
| 710 | sel.reset()
|
---|
| 711 | #ss.set_selection()
|
---|
| 712 | ssubon.set_selection()
|
---|
| 713 | precal[keys[2*i+1]]=ll
|
---|
| 714 | #del ss
|
---|
| 715 | #sig=resspec.get_scan('*_fs')
|
---|
| 716 | #ref=resspec.get_scan('*_fsr')
|
---|
| 717 | sel.set_types( srctype.fson )
|
---|
| 718 | resspec.set_selection( sel )
|
---|
| 719 | sig=resspec.copy()
|
---|
| 720 | resspec.set_selection()
|
---|
| 721 | sel.reset()
|
---|
| 722 | sel.set_type( srctype.fsoff )
|
---|
| 723 | resspec.set_selection( sel )
|
---|
| 724 | ref=resspec.copy()
|
---|
| 725 | resspec.set_selection()
|
---|
| 726 | sel.reset()
|
---|
| 727 | for k in range(len(polnos)):
|
---|
| 728 | for j in range(len(ifnos)):
|
---|
| 729 | sel.set_ifs(ifnos[j])
|
---|
| 730 | sel.set_polarizations(polnos[k])
|
---|
| 731 | try:
|
---|
| 732 | sig.set_selection(sel)
|
---|
| 733 | postcal.append(numpy.array(sig._getspectrum(0)))
|
---|
| 734 | except:
|
---|
| 735 | ref.set_selection(sel)
|
---|
| 736 | postcal.append(numpy.array(ref._getspectrum(0)))
|
---|
| 737 | sel.reset()
|
---|
| 738 | resspec.set_selection()
|
---|
| 739 | del sel
|
---|
| 740 | # plot
|
---|
[1861] | 741 | asaplog.post()
|
---|
[1819] | 742 | asaplog.push('Plot only first spectrum for each [if,pol] pairs to verify calibration.')
|
---|
[1861] | 743 | asaplog.post('WARN')
|
---|
[1819] | 744 | p=asaplotgui.asaplotgui()
|
---|
| 745 | #nr=min(6,len(ifnos)*len(polnos))
|
---|
| 746 | nr=len(ifnos)/2*len(polnos)
|
---|
| 747 | titles=[]
|
---|
| 748 | btics=[]
|
---|
| 749 | if nr>3:
|
---|
[1861] | 750 | asaplog.post()
|
---|
[1819] | 751 | asaplog.push('Only first 3 [if,pol] pairs are plotted.')
|
---|
[1861] | 752 | asaplog.post('WARN')
|
---|
[1819] | 753 | nr=3
|
---|
| 754 | p.set_panels(rows=nr,cols=3,nplots=3*nr,ganged=False)
|
---|
| 755 | for i in range(3*nr):
|
---|
| 756 | b=False
|
---|
| 757 | if i >= 3*nr-3:
|
---|
| 758 | b=True
|
---|
| 759 | btics.append(b)
|
---|
| 760 | for i in range(nr):
|
---|
| 761 | p.subplot(3*i)
|
---|
| 762 | p.color=0
|
---|
| 763 | 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)])
|
---|
| 764 | titles.append(title)
|
---|
| 765 | #p.set_axes('title',title,fontsize=40)
|
---|
| 766 | ymin=1.0e100
|
---|
| 767 | ymax=-1.0e100
|
---|
| 768 | nchan=s.nchan()
|
---|
| 769 | edge=int(nchan*0.01)
|
---|
| 770 | for j in range(4):
|
---|
| 771 | spmin=min(precal[keys[j]][i][edge:nchan-edge])
|
---|
| 772 | spmax=max(precal[keys[j]][i][edge:nchan-edge])
|
---|
| 773 | ymin=min(ymin,spmin)
|
---|
| 774 | ymax=max(ymax,spmax)
|
---|
| 775 | for j in range(4):
|
---|
| 776 | if i==0:
|
---|
| 777 | p.set_line(label=keys[j])
|
---|
| 778 | else:
|
---|
| 779 | p.legend()
|
---|
| 780 | p.plot(precal[keys[j]][i])
|
---|
| 781 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
| 782 | if not btics[3*i]:
|
---|
| 783 | p.axes.set_xticks([])
|
---|
| 784 | p.subplot(3*i+1)
|
---|
| 785 | p.color=0
|
---|
| 786 | title='sig data IF%s POL%s' % (ifnos[2*int(i/len(polnos))],polnos[i%len(polnos)])
|
---|
| 787 | titles.append(title)
|
---|
| 788 | #p.set_axes('title',title)
|
---|
| 789 | p.legend()
|
---|
| 790 | ymin=postcal[2*i][edge:nchan-edge].min()
|
---|
| 791 | ymax=postcal[2*i][edge:nchan-edge].max()
|
---|
| 792 | p.plot(postcal[2*i])
|
---|
| 793 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
| 794 | if not btics[3*i+1]:
|
---|
| 795 | p.axes.set_xticks([])
|
---|
| 796 | p.subplot(3*i+2)
|
---|
| 797 | p.color=0
|
---|
| 798 | title='ref data IF%s POL%s' % (ifnos[2*int(i/len(polnos))+1],polnos[i%len(polnos)])
|
---|
| 799 | titles.append(title)
|
---|
| 800 | #p.set_axes('title',title)
|
---|
| 801 | p.legend()
|
---|
| 802 | ymin=postcal[2*i+1][edge:nchan-edge].min()
|
---|
| 803 | ymax=postcal[2*i+1][edge:nchan-edge].max()
|
---|
| 804 | p.plot(postcal[2*i+1])
|
---|
| 805 | p.axes.set_ylim(ymin-0.1*abs(ymin),ymax+0.1*abs(ymax))
|
---|
| 806 | if not btics[3*i+2]:
|
---|
| 807 | p.axes.set_xticks([])
|
---|
| 808 | for i in range(3*nr):
|
---|
| 809 | p.subplot(i)
|
---|
| 810 | p.set_axes('title',titles[i],fontsize='medium')
|
---|
| 811 | x=raw_input('Accept calibration ([y]/n): ' )
|
---|
| 812 | if x.upper() == 'N':
|
---|
| 813 | p.unmap()
|
---|
| 814 | del p
|
---|
| 815 | return scabtab
|
---|
| 816 | p.unmap()
|
---|
| 817 | del p
|
---|
| 818 | ###
|
---|
[1391] | 819 | resspec._add_history("calfs",varlist)
|
---|
| 820 | return resspec
|
---|
| 821 |
|
---|
[1862] | 822 | @asaplog_post_dec
|
---|
[918] | 823 | def merge(*args):
|
---|
[945] | 824 | """
|
---|
[1362] | 825 | Merge a list of scanatables, or comma-sperated scantables into one
|
---|
| 826 | scnatble.
|
---|
| 827 | Parameters:
|
---|
| 828 | A list [scan1, scan2] or scan1, scan2.
|
---|
| 829 | Example:
|
---|
| 830 | myscans = [scan1, scan2]
|
---|
[1589] | 831 | allscans = merge(myscans)
|
---|
| 832 | # or equivalent
|
---|
| 833 | sameallscans = merge(scan1, scan2)
|
---|
[945] | 834 | """
|
---|
[918] | 835 | varlist = vars()
|
---|
| 836 | if isinstance(args[0],list):
|
---|
| 837 | lst = tuple(args[0])
|
---|
| 838 | elif isinstance(args[0],tuple):
|
---|
| 839 | lst = args[0]
|
---|
| 840 | else:
|
---|
| 841 | lst = tuple(args)
|
---|
| 842 | varlist["args"] = "%d scantables" % len(lst)
|
---|
| 843 | # need special formatting her for history...
|
---|
| 844 | from asap._asap import stmath
|
---|
| 845 | stm = stmath()
|
---|
| 846 | for s in lst:
|
---|
| 847 | if not isinstance(s,scantable):
|
---|
| 848 | msg = "Please give a list of scantables"
|
---|
[1859] | 849 | raise TypeError(msg)
|
---|
[918] | 850 | s = scantable(stm._merge(lst))
|
---|
| 851 | s._add_history("merge", varlist)
|
---|
| 852 | return s
|
---|
[1819] | 853 |
|
---|
[1862] | 854 | @asaplog_post_dec
|
---|
[1819] | 855 | def calibrate( scantab, scannos=[], calmode='none', verify=None ):
|
---|
| 856 | """
|
---|
| 857 | Calibrate data.
|
---|
[1826] | 858 |
|
---|
[1819] | 859 | Parameters:
|
---|
| 860 | scantab: scantable
|
---|
| 861 | scannos: list of scan number
|
---|
| 862 | calmode: calibration mode
|
---|
[1826] | 863 | verify: verify calibration
|
---|
[1819] | 864 | """
|
---|
[2103] | 865 | import re
|
---|
[1819] | 866 | antname = scantab.get_antennaname()
|
---|
| 867 | if ( calmode == 'nod' ):
|
---|
| 868 | asaplog.push( 'Calibrating nod data.' )
|
---|
| 869 | scal = calnod( scantab, scannos=scannos, verify=verify )
|
---|
| 870 | elif ( calmode == 'quotient' ):
|
---|
| 871 | asaplog.push( 'Calibrating using quotient.' )
|
---|
| 872 | scal = scantab.auto_quotient( verify=verify )
|
---|
| 873 | elif ( calmode == 'ps' ):
|
---|
| 874 | asaplog.push( 'Calibrating %s position-switched data.' % antname )
|
---|
| 875 | if ( antname.find( 'APEX' ) != -1 ):
|
---|
| 876 | scal = apexcal( scantab, scannos, calmode, verify )
|
---|
[2103] | 877 | elif ( antname.find( 'ALMA' ) != -1 or antname.find( 'OSF' ) != -1
|
---|
| 878 | or re.match('DV[0-9][0-9]$',antname) is not None
|
---|
| 879 | or re.match('PM[0-9][0-9]$',antname) is not None
|
---|
| 880 | or re.match('CM[0-9][0-9]$',antname) is not None
|
---|
| 881 | or re.match('DA[0-9][0-9]$',antname) is not None ):
|
---|
[1819] | 882 | scal = almacal( scantab, scannos, calmode, verify )
|
---|
| 883 | else:
|
---|
| 884 | scal = calps( scantab, scannos=scannos, verify=verify )
|
---|
| 885 | elif ( calmode == 'fs' or calmode == 'fsotf' ):
|
---|
| 886 | asaplog.push( 'Calibrating %s frequency-switched data.' % antname )
|
---|
| 887 | if ( antname.find( 'APEX' ) != -1 ):
|
---|
| 888 | scal = apexcal( scantab, scannos, calmode, verify )
|
---|
| 889 | elif ( antname.find( 'ALMA' ) != -1 or antname.find( 'OSF' ) != -1 ):
|
---|
| 890 | scal = almacal( scantab, scannos, calmode, verify )
|
---|
| 891 | else:
|
---|
| 892 | scal = calfs( scantab, scannos=scannos, verify=verify )
|
---|
| 893 | elif ( calmode == 'otf' ):
|
---|
| 894 | asaplog.push( 'Calibrating %s On-The-Fly data.' % antname )
|
---|
| 895 | scal = almacal( scantab, scannos, calmode, verify )
|
---|
| 896 | else:
|
---|
| 897 | asaplog.push( 'No calibration.' )
|
---|
| 898 | scal = scantab.copy()
|
---|
| 899 |
|
---|
[1826] | 900 | return scal
|
---|
[1819] | 901 |
|
---|
| 902 | def apexcal( scantab, scannos=[], calmode='none', verify=False ):
|
---|
| 903 | """
|
---|
| 904 | Calibrate APEX data
|
---|
| 905 |
|
---|
| 906 | Parameters:
|
---|
| 907 | scantab: scantable
|
---|
| 908 | scannos: list of scan number
|
---|
| 909 | calmode: calibration mode
|
---|
| 910 |
|
---|
[1826] | 911 | verify: verify calibration
|
---|
[1819] | 912 | """
|
---|
| 913 | from asap._asap import stmath
|
---|
| 914 | stm = stmath()
|
---|
| 915 | antname = scantab.get_antennaname()
|
---|
| 916 | ssub = scantab.get_scan( scannos )
|
---|
| 917 | scal = scantable( stm.cwcal( ssub, calmode, antname ) )
|
---|
| 918 | return scal
|
---|
| 919 |
|
---|
| 920 | def almacal( scantab, scannos=[], calmode='none', verify=False ):
|
---|
| 921 | """
|
---|
| 922 | Calibrate ALMA data
|
---|
| 923 |
|
---|
| 924 | Parameters:
|
---|
| 925 | scantab: scantable
|
---|
| 926 | scannos: list of scan number
|
---|
| 927 | calmode: calibration mode
|
---|
| 928 |
|
---|
[1826] | 929 | verify: verify calibration
|
---|
[1819] | 930 | """
|
---|
| 931 | from asap._asap import stmath
|
---|
| 932 | stm = stmath()
|
---|
| 933 | ssub = scantab.get_scan( scannos )
|
---|
| 934 | scal = scantable( stm.almacal( ssub, calmode ) )
|
---|
| 935 | return scal
|
---|
| 936 |
|
---|
[1862] | 937 | @asaplog_post_dec
|
---|
[1819] | 938 | def splitant(filename, outprefix='',overwrite=False):
|
---|
| 939 | """
|
---|
| 940 | Split Measurement set by antenna name, save data as a scantables,
|
---|
| 941 | and return a list of filename.
|
---|
[1826] | 942 | Notice this method can only be available from CASA.
|
---|
[1819] | 943 | Prameter
|
---|
[1826] | 944 | filename: the name of Measurement set to be read.
|
---|
[1819] | 945 | outprefix: the prefix of output scantable name.
|
---|
| 946 | the names of output scantable will be
|
---|
| 947 | outprefix.antenna1, outprefix.antenna2, ....
|
---|
| 948 | If not specified, outprefix = filename is assumed.
|
---|
| 949 | overwrite If the file should be overwritten if it exists.
|
---|
| 950 | The default False is to return with warning
|
---|
| 951 | without writing the output. USE WITH CARE.
|
---|
[1826] | 952 |
|
---|
[1819] | 953 | """
|
---|
| 954 | # Import the table toolkit from CASA
|
---|
[1859] | 955 | import casac
|
---|
[1918] | 956 | from asap.scantable import is_ms
|
---|
[1859] | 957 | tbtool = casac.homefinder.find_home_by_name('tableHome')
|
---|
| 958 | tb = tbtool.create()
|
---|
[1819] | 959 | # Check the input filename
|
---|
| 960 | if isinstance(filename, str):
|
---|
| 961 | import os.path
|
---|
| 962 | filename = os.path.expandvars(filename)
|
---|
| 963 | filename = os.path.expanduser(filename)
|
---|
| 964 | if not os.path.exists(filename):
|
---|
| 965 | s = "File '%s' not found." % (filename)
|
---|
| 966 | raise IOError(s)
|
---|
| 967 | # check if input file is MS
|
---|
[1883] | 968 | #if not os.path.isdir(filename) \
|
---|
| 969 | # or not os.path.exists(filename+'/ANTENNA') \
|
---|
| 970 | # or not os.path.exists(filename+'/table.f1'):
|
---|
| 971 | if not is_ms(filename):
|
---|
[1819] | 972 | s = "File '%s' is not a Measurement set." % (filename)
|
---|
| 973 | raise IOError(s)
|
---|
| 974 | else:
|
---|
| 975 | s = "The filename should be string. "
|
---|
| 976 | raise TypeError(s)
|
---|
| 977 | # Check out put file name
|
---|
| 978 | outname=''
|
---|
| 979 | if len(outprefix) > 0: prefix=outprefix+'.'
|
---|
| 980 | else:
|
---|
| 981 | prefix=filename.rstrip('/')
|
---|
| 982 | # Now do the actual splitting.
|
---|
| 983 | outfiles=[]
|
---|
[2034] | 984 | tb.open(tablename=filename,nomodify=True)
|
---|
| 985 | ant1=tb.getcol('ANTENNA1',0,-1,1)
|
---|
| 986 | anttab=tb.getkeyword('ANTENNA').split()[-1]
|
---|
| 987 | tb.close()
|
---|
| 988 | #tb.open(tablename=filename+'/ANTENNA',nomodify=True)
|
---|
| 989 | tb.open(tablename=anttab,nomodify=True)
|
---|
[1819] | 990 | nant=tb.nrows()
|
---|
| 991 | antnames=tb.getcol('NAME',0,nant,1)
|
---|
| 992 | tb.close()
|
---|
[1880] | 993 | tmpname='asapmath.splitant.tmp'
|
---|
[1819] | 994 | for antid in set(ant1):
|
---|
[1883] | 995 | tb.open(tablename=filename,nomodify=True)
|
---|
| 996 | tbsel=tb.query('ANTENNA1 == %s && ANTENNA2 == %s'%(antid,antid),tmpname)
|
---|
[1918] | 997 | scan=scantable(tmpname,average=False,getpt=True,antenna=int(antid))
|
---|
| 998 | outname=prefix+antnames[antid]+'.asap'
|
---|
| 999 | scan.save(outname,format='ASAP',overwrite=overwrite)
|
---|
[1880] | 1000 | tbsel.close()
|
---|
[1883] | 1001 | tb.close()
|
---|
[1880] | 1002 | del tbsel
|
---|
[1819] | 1003 | del scan
|
---|
| 1004 | outfiles.append(outname)
|
---|
[1880] | 1005 | os.system('rm -rf '+tmpname)
|
---|
| 1006 | del tb
|
---|
[1819] | 1007 | return outfiles
|
---|
| 1008 |
|
---|
[1862] | 1009 | @asaplog_post_dec
|
---|
[1819] | 1010 | def _array2dOp( scan, value, mode="ADD", tsys=False ):
|
---|
| 1011 | """
|
---|
| 1012 | This function is workaround on the basic operation of scantable
|
---|
| 1013 | with 2 dimensional float list.
|
---|
| 1014 |
|
---|
| 1015 | scan: scantable operand
|
---|
| 1016 | value: float list operand
|
---|
| 1017 | mode: operation mode (ADD, SUB, MUL, DIV)
|
---|
[1826] | 1018 | tsys: if True, operate tsys as well
|
---|
[1819] | 1019 | """
|
---|
| 1020 | nrow = scan.nrow()
|
---|
| 1021 | s = None
|
---|
| 1022 | if len( value ) == 1:
|
---|
| 1023 | from asap._asap import stmath
|
---|
| 1024 | stm = stmath()
|
---|
| 1025 | s = scantable( stm._arrayop( scan.copy(), value[0], mode, tsys ) )
|
---|
| 1026 | del stm
|
---|
| 1027 | elif len( value ) != nrow:
|
---|
[1859] | 1028 | raise ValueError( 'len(value) must be 1 or conform to scan.nrow()' )
|
---|
[1819] | 1029 | else:
|
---|
| 1030 | from asap._asap import stmath
|
---|
| 1031 | stm = stmath()
|
---|
[1826] | 1032 | # insitu must be True
|
---|
[1819] | 1033 | stm._setinsitu( True )
|
---|
| 1034 | s = scan.copy()
|
---|
| 1035 | sel = selector()
|
---|
| 1036 | for irow in range( nrow ):
|
---|
| 1037 | sel.set_rows( irow )
|
---|
| 1038 | s.set_selection( sel )
|
---|
| 1039 | if len( value[irow] ) == 1:
|
---|
| 1040 | stm._unaryop( s, value[irow][0], mode, tsys )
|
---|
| 1041 | else:
|
---|
| 1042 | stm._arrayop( s, value[irow], mode, tsys, 'channel' )
|
---|
| 1043 | s.set_selection()
|
---|
| 1044 | sel.reset()
|
---|
| 1045 | del sel
|
---|
| 1046 | del stm
|
---|
| 1047 | return s
|
---|