[1846] | 1 | """This module defines the scantable class."""
|
---|
| 2 |
|
---|
[1697] | 3 | import os
|
---|
[1691] | 4 | try:
|
---|
| 5 | from functools import wraps as wraps_dec
|
---|
| 6 | except ImportError:
|
---|
| 7 | from asap.compatibility import wraps as wraps_dec
|
---|
| 8 |
|
---|
[1824] | 9 | from asap.env import is_casapy
|
---|
[876] | 10 | from asap._asap import Scantable
|
---|
[1843] | 11 | from asap._asap import filler
|
---|
[1824] | 12 | from asap.parameters import rcParams
|
---|
[1862] | 13 | from asap.logging import asaplog, asaplog_post_dec
|
---|
[1824] | 14 | from asap.selector import selector
|
---|
| 15 | from asap.linecatalog import linecatalog
|
---|
[1600] | 16 | from asap.coordinate import coordinate
|
---|
[1859] | 17 | from asap.utils import _n_bools, mask_not, mask_and, mask_or, page
|
---|
[1907] | 18 | from asap.asapfitter import fitter
|
---|
[102] | 19 |
|
---|
[1689] | 20 |
|
---|
| 21 | def preserve_selection(func):
|
---|
[1691] | 22 | @wraps_dec(func)
|
---|
[1689] | 23 | def wrap(obj, *args, **kw):
|
---|
| 24 | basesel = obj.get_selection()
|
---|
[1857] | 25 | try:
|
---|
| 26 | val = func(obj, *args, **kw)
|
---|
| 27 | finally:
|
---|
| 28 | obj.set_selection(basesel)
|
---|
[1689] | 29 | return val
|
---|
| 30 | return wrap
|
---|
| 31 |
|
---|
[1846] | 32 | def is_scantable(filename):
|
---|
| 33 | """Is the given file a scantable?
|
---|
[1689] | 34 |
|
---|
[1846] | 35 | Parameters:
|
---|
| 36 |
|
---|
| 37 | filename: the name of the file/directory to test
|
---|
| 38 |
|
---|
| 39 | """
|
---|
[1883] | 40 | if ( os.path.isdir(filename)
|
---|
| 41 | and os.path.exists(filename+'/table.info')
|
---|
| 42 | and os.path.exists(filename+'/table.dat') ):
|
---|
| 43 | f=open(filename+'/table.info')
|
---|
| 44 | l=f.readline()
|
---|
| 45 | f.close()
|
---|
| 46 | #if ( l.find('Scantable') != -1 ):
|
---|
| 47 | if ( l.find('Measurement Set') == -1 ):
|
---|
| 48 | return True
|
---|
| 49 | else:
|
---|
| 50 | return False
|
---|
| 51 | else:
|
---|
| 52 | return False
|
---|
| 53 | ## return (os.path.isdir(filename)
|
---|
| 54 | ## and not os.path.exists(filename+'/table.f1')
|
---|
| 55 | ## and os.path.exists(filename+'/table.info'))
|
---|
[1697] | 56 |
|
---|
[1883] | 57 | def is_ms(filename):
|
---|
| 58 | """Is the given file a MeasurementSet?
|
---|
[1697] | 59 |
|
---|
[1883] | 60 | Parameters:
|
---|
| 61 |
|
---|
| 62 | filename: the name of the file/directory to test
|
---|
| 63 |
|
---|
| 64 | """
|
---|
| 65 | if ( os.path.isdir(filename)
|
---|
| 66 | and os.path.exists(filename+'/table.info')
|
---|
| 67 | and os.path.exists(filename+'/table.dat') ):
|
---|
| 68 | f=open(filename+'/table.info')
|
---|
| 69 | l=f.readline()
|
---|
| 70 | f.close()
|
---|
| 71 | if ( l.find('Measurement Set') != -1 ):
|
---|
| 72 | return True
|
---|
| 73 | else:
|
---|
| 74 | return False
|
---|
| 75 | else:
|
---|
| 76 | return False
|
---|
| 77 |
|
---|
[876] | 78 | class scantable(Scantable):
|
---|
[1846] | 79 | """\
|
---|
| 80 | The ASAP container for scans (single-dish data).
|
---|
[102] | 81 | """
|
---|
[1819] | 82 |
|
---|
[1862] | 83 | @asaplog_post_dec
|
---|
[1916] | 84 | #def __init__(self, filename, average=None, unit=None, getpt=None,
|
---|
| 85 | # antenna=None, parallactify=None):
|
---|
| 86 | def __init__(self, filename, average=None, unit=None, parallactify=None, **args):
|
---|
[1846] | 87 | """\
|
---|
[102] | 88 | Create a scantable from a saved one or make a reference
|
---|
[1846] | 89 |
|
---|
[102] | 90 | Parameters:
|
---|
[1846] | 91 |
|
---|
| 92 | filename: the name of an asap table on disk
|
---|
| 93 | or
|
---|
| 94 | the name of a rpfits/sdfits/ms file
|
---|
| 95 | (integrations within scans are auto averaged
|
---|
| 96 | and the whole file is read) or
|
---|
| 97 | [advanced] a reference to an existing scantable
|
---|
| 98 |
|
---|
| 99 | average: average all integrations withinb a scan on read.
|
---|
| 100 | The default (True) is taken from .asaprc.
|
---|
| 101 |
|
---|
[484] | 102 | unit: brightness unit; must be consistent with K or Jy.
|
---|
[1846] | 103 | Over-rides the default selected by the filler
|
---|
| 104 | (input rpfits/sdfits/ms) or replaces the value
|
---|
| 105 | in existing scantables
|
---|
| 106 |
|
---|
| 107 | getpt: for MeasurementSet input data only:
|
---|
| 108 | If True, all pointing data are filled.
|
---|
| 109 | The deafult is False, which makes time to load
|
---|
| 110 | the MS data faster in some cases.
|
---|
| 111 |
|
---|
| 112 | antenna: Antenna selection. integer (id) or string (name or id).
|
---|
| 113 |
|
---|
| 114 | parallactify: Indicate that the data had been parallatified. Default
|
---|
| 115 | is taken from rc file.
|
---|
| 116 |
|
---|
[710] | 117 | """
|
---|
[976] | 118 | if average is None:
|
---|
[710] | 119 | average = rcParams['scantable.autoaverage']
|
---|
[1916] | 120 | #if getpt is None:
|
---|
| 121 | # getpt = True
|
---|
| 122 | #if antenna is not None:
|
---|
| 123 | # asaplog.push("Antenna selection currently unsupported."
|
---|
| 124 | # "Using ''")
|
---|
| 125 | # asaplog.post('WARN')
|
---|
| 126 | #if antenna is None:
|
---|
| 127 | # antenna = ''
|
---|
| 128 | #elif type(antenna) == int:
|
---|
| 129 | # antenna = '%s' % antenna
|
---|
| 130 | #elif type(antenna) == list:
|
---|
| 131 | # tmpstr = ''
|
---|
| 132 | # for i in range( len(antenna) ):
|
---|
| 133 | # if type(antenna[i]) == int:
|
---|
| 134 | # tmpstr = tmpstr + ('%s,'%(antenna[i]))
|
---|
| 135 | # elif type(antenna[i]) == str:
|
---|
| 136 | # tmpstr=tmpstr+antenna[i]+','
|
---|
| 137 | # else:
|
---|
| 138 | # raise TypeError('Bad antenna selection.')
|
---|
| 139 | # antenna = tmpstr.rstrip(',')
|
---|
[1593] | 140 | parallactify = parallactify or rcParams['scantable.parallactify']
|
---|
[1259] | 141 | varlist = vars()
|
---|
[876] | 142 | from asap._asap import stmath
|
---|
[1819] | 143 | self._math = stmath( rcParams['insitu'] )
|
---|
[876] | 144 | if isinstance(filename, Scantable):
|
---|
| 145 | Scantable.__init__(self, filename)
|
---|
[181] | 146 | else:
|
---|
[1697] | 147 | if isinstance(filename, str):
|
---|
[976] | 148 | filename = os.path.expandvars(filename)
|
---|
| 149 | filename = os.path.expanduser(filename)
|
---|
| 150 | if not os.path.exists(filename):
|
---|
| 151 | s = "File '%s' not found." % (filename)
|
---|
| 152 | raise IOError(s)
|
---|
[1697] | 153 | if is_scantable(filename):
|
---|
| 154 | ondisk = rcParams['scantable.storage'] == 'disk'
|
---|
| 155 | Scantable.__init__(self, filename, ondisk)
|
---|
| 156 | if unit is not None:
|
---|
| 157 | self.set_fluxunit(unit)
|
---|
[1819] | 158 | # do not reset to the default freqframe
|
---|
| 159 | #self.set_freqframe(rcParams['scantable.freqframe'])
|
---|
[1883] | 160 | #elif os.path.isdir(filename) \
|
---|
| 161 | # and not os.path.exists(filename+'/table.f1'):
|
---|
| 162 | elif is_ms(filename):
|
---|
[1916] | 163 | # Measurement Set
|
---|
| 164 | opts={'ms': {}}
|
---|
| 165 | mskeys=['getpt','antenna']
|
---|
| 166 | for key in mskeys:
|
---|
| 167 | if key in args.keys():
|
---|
| 168 | opts['ms'][key] = args[key]
|
---|
| 169 | #self._fill([filename], unit, average, getpt, antenna)
|
---|
| 170 | self._fill([filename], unit, average, opts)
|
---|
[1893] | 171 | elif os.path.isfile(filename):
|
---|
[1916] | 172 | #self._fill([filename], unit, average, getpt, antenna)
|
---|
| 173 | self._fill([filename], unit, average)
|
---|
[1883] | 174 | else:
|
---|
[1819] | 175 | msg = "The given file '%s'is not a valid " \
|
---|
| 176 | "asap table." % (filename)
|
---|
[1859] | 177 | raise IOError(msg)
|
---|
[1118] | 178 | elif (isinstance(filename, list) or isinstance(filename, tuple)) \
|
---|
[976] | 179 | and isinstance(filename[-1], str):
|
---|
[1916] | 180 | #self._fill(filename, unit, average, getpt, antenna)
|
---|
| 181 | self._fill(filename, unit, average)
|
---|
[1586] | 182 | self.parallactify(parallactify)
|
---|
[1259] | 183 | self._add_history("scantable", varlist)
|
---|
[102] | 184 |
|
---|
[1862] | 185 | @asaplog_post_dec
|
---|
[876] | 186 | def save(self, name=None, format=None, overwrite=False):
|
---|
[1846] | 187 | """\
|
---|
[1280] | 188 | Store the scantable on disk. This can be an asap (aips++) Table,
|
---|
| 189 | SDFITS or MS2 format.
|
---|
[1846] | 190 |
|
---|
[116] | 191 | Parameters:
|
---|
[1846] | 192 |
|
---|
[1093] | 193 | name: the name of the outputfile. For format "ASCII"
|
---|
| 194 | this is the root file name (data in 'name'.txt
|
---|
[497] | 195 | and header in 'name'_header.txt)
|
---|
[1855] | 196 |
|
---|
[116] | 197 | format: an optional file format. Default is ASAP.
|
---|
[1855] | 198 | Allowed are:
|
---|
| 199 |
|
---|
| 200 | * 'ASAP' (save as ASAP [aips++] Table),
|
---|
| 201 | * 'SDFITS' (save as SDFITS file)
|
---|
| 202 | * 'ASCII' (saves as ascii text file)
|
---|
| 203 | * 'MS2' (saves as an casacore MeasurementSet V2)
|
---|
| 204 | * 'FITS' (save as image FITS - not readable by class)
|
---|
| 205 | * 'CLASS' (save as FITS readable by CLASS)
|
---|
| 206 |
|
---|
[411] | 207 | overwrite: If the file should be overwritten if it exists.
|
---|
[256] | 208 | The default False is to return with warning
|
---|
[411] | 209 | without writing the output. USE WITH CARE.
|
---|
[1855] | 210 |
|
---|
[1846] | 211 | Example::
|
---|
| 212 |
|
---|
[116] | 213 | scan.save('myscan.asap')
|
---|
[1118] | 214 | scan.save('myscan.sdfits', 'SDFITS')
|
---|
[1846] | 215 |
|
---|
[116] | 216 | """
|
---|
[411] | 217 | from os import path
|
---|
[1593] | 218 | format = format or rcParams['scantable.save']
|
---|
[256] | 219 | suffix = '.'+format.lower()
|
---|
[1118] | 220 | if name is None or name == "":
|
---|
[256] | 221 | name = 'scantable'+suffix
|
---|
[718] | 222 | msg = "No filename given. Using default name %s..." % name
|
---|
| 223 | asaplog.push(msg)
|
---|
[411] | 224 | name = path.expandvars(name)
|
---|
[256] | 225 | if path.isfile(name) or path.isdir(name):
|
---|
| 226 | if not overwrite:
|
---|
[718] | 227 | msg = "File %s exists." % name
|
---|
[1859] | 228 | raise IOError(msg)
|
---|
[451] | 229 | format2 = format.upper()
|
---|
| 230 | if format2 == 'ASAP':
|
---|
[116] | 231 | self._save(name)
|
---|
| 232 | else:
|
---|
[989] | 233 | from asap._asap import stwriter as stw
|
---|
[1118] | 234 | writer = stw(format2)
|
---|
| 235 | writer.write(self, name)
|
---|
[116] | 236 | return
|
---|
| 237 |
|
---|
[102] | 238 | def copy(self):
|
---|
[1846] | 239 | """Return a copy of this scantable.
|
---|
| 240 |
|
---|
| 241 | *Note*:
|
---|
| 242 |
|
---|
[1348] | 243 | This makes a full (deep) copy. scan2 = scan1 makes a reference.
|
---|
[1846] | 244 |
|
---|
| 245 | Example::
|
---|
| 246 |
|
---|
[102] | 247 | copiedscan = scan.copy()
|
---|
[1846] | 248 |
|
---|
[102] | 249 | """
|
---|
[876] | 250 | sd = scantable(Scantable._copy(self))
|
---|
[113] | 251 | return sd
|
---|
| 252 |
|
---|
[1093] | 253 | def drop_scan(self, scanid=None):
|
---|
[1846] | 254 | """\
|
---|
[1093] | 255 | Return a new scantable where the specified scan number(s) has(have)
|
---|
| 256 | been dropped.
|
---|
[1846] | 257 |
|
---|
[1093] | 258 | Parameters:
|
---|
[1846] | 259 |
|
---|
[1093] | 260 | scanid: a (list of) scan number(s)
|
---|
[1846] | 261 |
|
---|
[1093] | 262 | """
|
---|
| 263 | from asap import _is_sequence_or_number as _is_valid
|
---|
| 264 | from asap import _to_list
|
---|
| 265 | from asap import unique
|
---|
| 266 | if not _is_valid(scanid):
|
---|
[1859] | 267 | raise RuntimeError( 'Please specify a scanno to drop from the scantable' )
|
---|
| 268 | scanid = _to_list(scanid)
|
---|
| 269 | allscans = unique([ self.getscan(i) for i in range(self.nrow())])
|
---|
| 270 | for sid in scanid: allscans.remove(sid)
|
---|
| 271 | if len(allscans) == 0:
|
---|
| 272 | raise ValueError("Can't remove all scans")
|
---|
| 273 | sel = selector(scans=allscans)
|
---|
| 274 | return self._select_copy(sel)
|
---|
[1093] | 275 |
|
---|
[1594] | 276 | def _select_copy(self, selection):
|
---|
| 277 | orig = self.get_selection()
|
---|
| 278 | self.set_selection(orig+selection)
|
---|
| 279 | cp = self.copy()
|
---|
| 280 | self.set_selection(orig)
|
---|
| 281 | return cp
|
---|
| 282 |
|
---|
[102] | 283 | def get_scan(self, scanid=None):
|
---|
[1855] | 284 | """\
|
---|
[102] | 285 | Return a specific scan (by scanno) or collection of scans (by
|
---|
| 286 | source name) in a new scantable.
|
---|
[1846] | 287 |
|
---|
| 288 | *Note*:
|
---|
| 289 |
|
---|
[1348] | 290 | See scantable.drop_scan() for the inverse operation.
|
---|
[1846] | 291 |
|
---|
[102] | 292 | Parameters:
|
---|
[1846] | 293 |
|
---|
[513] | 294 | scanid: a (list of) scanno or a source name, unix-style
|
---|
| 295 | patterns are accepted for source name matching, e.g.
|
---|
| 296 | '*_R' gets all 'ref scans
|
---|
[1846] | 297 |
|
---|
| 298 | Example::
|
---|
| 299 |
|
---|
[513] | 300 | # get all scans containing the source '323p459'
|
---|
| 301 | newscan = scan.get_scan('323p459')
|
---|
| 302 | # get all 'off' scans
|
---|
| 303 | refscans = scan.get_scan('*_R')
|
---|
| 304 | # get a susbset of scans by scanno (as listed in scan.summary())
|
---|
[1118] | 305 | newscan = scan.get_scan([0, 2, 7, 10])
|
---|
[1846] | 306 |
|
---|
[102] | 307 | """
|
---|
| 308 | if scanid is None:
|
---|
[1859] | 309 | raise RuntimeError( 'Please specify a scan no or name to '
|
---|
| 310 | 'retrieve from the scantable' )
|
---|
[102] | 311 | try:
|
---|
[946] | 312 | bsel = self.get_selection()
|
---|
| 313 | sel = selector()
|
---|
[102] | 314 | if type(scanid) is str:
|
---|
[946] | 315 | sel.set_name(scanid)
|
---|
[1594] | 316 | return self._select_copy(sel)
|
---|
[102] | 317 | elif type(scanid) is int:
|
---|
[946] | 318 | sel.set_scans([scanid])
|
---|
[1594] | 319 | return self._select_copy(sel)
|
---|
[381] | 320 | elif type(scanid) is list:
|
---|
[946] | 321 | sel.set_scans(scanid)
|
---|
[1594] | 322 | return self._select_copy(sel)
|
---|
[381] | 323 | else:
|
---|
[718] | 324 | msg = "Illegal scanid type, use 'int' or 'list' if ints."
|
---|
[1859] | 325 | raise TypeError(msg)
|
---|
[102] | 326 | except RuntimeError:
|
---|
[1859] | 327 | raise
|
---|
[102] | 328 |
|
---|
| 329 | def __str__(self):
|
---|
[1118] | 330 | return Scantable._summary(self, True)
|
---|
[102] | 331 |
|
---|
[976] | 332 | def summary(self, filename=None):
|
---|
[1846] | 333 | """\
|
---|
[102] | 334 | Print a summary of the contents of this scantable.
|
---|
[1846] | 335 |
|
---|
[102] | 336 | Parameters:
|
---|
[1846] | 337 |
|
---|
[102] | 338 | filename: the name of a file to write the putput to
|
---|
| 339 | Default - no file output
|
---|
[1846] | 340 |
|
---|
[102] | 341 | """
|
---|
[976] | 342 | info = Scantable._summary(self, True)
|
---|
[102] | 343 | if filename is not None:
|
---|
[256] | 344 | if filename is "":
|
---|
| 345 | filename = 'scantable_summary.txt'
|
---|
[415] | 346 | from os.path import expandvars, isdir
|
---|
[411] | 347 | filename = expandvars(filename)
|
---|
[415] | 348 | if not isdir(filename):
|
---|
[413] | 349 | data = open(filename, 'w')
|
---|
| 350 | data.write(info)
|
---|
| 351 | data.close()
|
---|
| 352 | else:
|
---|
[718] | 353 | msg = "Illegal file name '%s'." % (filename)
|
---|
[1859] | 354 | raise IOError(msg)
|
---|
| 355 | return page(info)
|
---|
[710] | 356 |
|
---|
[1512] | 357 | def get_spectrum(self, rowno):
|
---|
[1471] | 358 | """Return the spectrum for the current row in the scantable as a list.
|
---|
[1846] | 359 |
|
---|
[1471] | 360 | Parameters:
|
---|
[1846] | 361 |
|
---|
[1573] | 362 | rowno: the row number to retrieve the spectrum from
|
---|
[1846] | 363 |
|
---|
[1471] | 364 | """
|
---|
| 365 | return self._getspectrum(rowno)
|
---|
[946] | 366 |
|
---|
[1471] | 367 | def get_mask(self, rowno):
|
---|
| 368 | """Return the mask for the current row in the scantable as a list.
|
---|
[1846] | 369 |
|
---|
[1471] | 370 | Parameters:
|
---|
[1846] | 371 |
|
---|
[1573] | 372 | rowno: the row number to retrieve the mask from
|
---|
[1846] | 373 |
|
---|
[1471] | 374 | """
|
---|
| 375 | return self._getmask(rowno)
|
---|
| 376 |
|
---|
| 377 | def set_spectrum(self, spec, rowno):
|
---|
| 378 | """Return the spectrum for the current row in the scantable as a list.
|
---|
[1846] | 379 |
|
---|
[1471] | 380 | Parameters:
|
---|
[1846] | 381 |
|
---|
[1855] | 382 | spec: the new spectrum
|
---|
[1846] | 383 |
|
---|
[1855] | 384 | rowno: the row number to set the spectrum for
|
---|
| 385 |
|
---|
[1471] | 386 | """
|
---|
| 387 | assert(len(spec) == self.nchan())
|
---|
| 388 | return self._setspectrum(spec, rowno)
|
---|
| 389 |
|
---|
[1600] | 390 | def get_coordinate(self, rowno):
|
---|
| 391 | """Return the (spectral) coordinate for a a given 'rowno'.
|
---|
[1846] | 392 |
|
---|
| 393 | *Note*:
|
---|
| 394 |
|
---|
[1600] | 395 | * This coordinate is only valid until a scantable method modifies
|
---|
| 396 | the frequency axis.
|
---|
| 397 | * This coordinate does contain the original frequency set-up
|
---|
| 398 | NOT the new frame. The conversions however are done using the user
|
---|
| 399 | specified frame (e.g. LSRK/TOPO). To get the 'real' coordinate,
|
---|
| 400 | use scantable.freq_align first. Without it there is no closure,
|
---|
[1846] | 401 | i.e.::
|
---|
[1600] | 402 |
|
---|
[1846] | 403 | c = myscan.get_coordinate(0)
|
---|
| 404 | c.to_frequency(c.get_reference_pixel()) != c.get_reference_value()
|
---|
| 405 |
|
---|
[1600] | 406 | Parameters:
|
---|
[1846] | 407 |
|
---|
[1600] | 408 | rowno: the row number for the spectral coordinate
|
---|
| 409 |
|
---|
| 410 | """
|
---|
| 411 | return coordinate(Scantable.get_coordinate(self, rowno))
|
---|
| 412 |
|
---|
[946] | 413 | def get_selection(self):
|
---|
[1846] | 414 | """\
|
---|
[1005] | 415 | Get the selection object currently set on this scantable.
|
---|
[1846] | 416 |
|
---|
| 417 | Example::
|
---|
| 418 |
|
---|
[1005] | 419 | sel = scan.get_selection()
|
---|
| 420 | sel.set_ifs(0) # select IF 0
|
---|
| 421 | scan.set_selection(sel) # apply modified selection
|
---|
[1846] | 422 |
|
---|
[946] | 423 | """
|
---|
| 424 | return selector(self._getselection())
|
---|
| 425 |
|
---|
[1576] | 426 | def set_selection(self, selection=None, **kw):
|
---|
[1846] | 427 | """\
|
---|
[1005] | 428 | Select a subset of the data. All following operations on this scantable
|
---|
| 429 | are only applied to thi selection.
|
---|
[1846] | 430 |
|
---|
[1005] | 431 | Parameters:
|
---|
[1697] | 432 |
|
---|
[1846] | 433 | selection: a selector object (default unset the selection), or
|
---|
| 434 | any combination of "pols", "ifs", "beams", "scans",
|
---|
| 435 | "cycles", "name", "query"
|
---|
[1697] | 436 |
|
---|
[1846] | 437 | Examples::
|
---|
[1697] | 438 |
|
---|
[1005] | 439 | sel = selector() # create a selection object
|
---|
[1118] | 440 | self.set_scans([0, 3]) # select SCANNO 0 and 3
|
---|
[1005] | 441 | scan.set_selection(sel) # set the selection
|
---|
| 442 | scan.summary() # will only print summary of scanno 0 an 3
|
---|
| 443 | scan.set_selection() # unset the selection
|
---|
[1697] | 444 | # or the equivalent
|
---|
| 445 | scan.set_selection(scans=[0,3])
|
---|
| 446 | scan.summary() # will only print summary of scanno 0 an 3
|
---|
| 447 | scan.set_selection() # unset the selection
|
---|
[1846] | 448 |
|
---|
[946] | 449 | """
|
---|
[1576] | 450 | if selection is None:
|
---|
| 451 | # reset
|
---|
| 452 | if len(kw) == 0:
|
---|
| 453 | selection = selector()
|
---|
| 454 | else:
|
---|
| 455 | # try keywords
|
---|
| 456 | for k in kw:
|
---|
| 457 | if k not in selector.fields:
|
---|
| 458 | raise KeyError("Invalid selection key '%s', valid keys are %s" % (k, selector.fields))
|
---|
| 459 | selection = selector(**kw)
|
---|
[946] | 460 | self._setselection(selection)
|
---|
| 461 |
|
---|
[1819] | 462 | def get_row(self, row=0, insitu=None):
|
---|
[1846] | 463 | """\
|
---|
[1819] | 464 | Select a row in the scantable.
|
---|
| 465 | Return a scantable with single row.
|
---|
[1846] | 466 |
|
---|
[1819] | 467 | Parameters:
|
---|
[1846] | 468 |
|
---|
| 469 | row: row no of integration, default is 0.
|
---|
| 470 | insitu: if False a new scantable is returned. Otherwise, the
|
---|
| 471 | scaling is done in-situ. The default is taken from .asaprc
|
---|
| 472 | (False)
|
---|
| 473 |
|
---|
[1819] | 474 | """
|
---|
| 475 | if insitu is None: insitu = rcParams['insitu']
|
---|
| 476 | if not insitu:
|
---|
| 477 | workscan = self.copy()
|
---|
| 478 | else:
|
---|
| 479 | workscan = self
|
---|
| 480 | # Select a row
|
---|
| 481 | sel=selector()
|
---|
| 482 | sel.set_scans([workscan.getscan(row)])
|
---|
| 483 | sel.set_cycles([workscan.getcycle(row)])
|
---|
| 484 | sel.set_beams([workscan.getbeam(row)])
|
---|
| 485 | sel.set_ifs([workscan.getif(row)])
|
---|
| 486 | sel.set_polarisations([workscan.getpol(row)])
|
---|
| 487 | sel.set_name(workscan._getsourcename(row))
|
---|
| 488 | workscan.set_selection(sel)
|
---|
| 489 | if not workscan.nrow() == 1:
|
---|
| 490 | msg = "Cloud not identify single row. %d rows selected."%(workscan.nrow())
|
---|
| 491 | raise RuntimeError(msg)
|
---|
| 492 | del sel
|
---|
| 493 | if insitu:
|
---|
| 494 | self._assign(workscan)
|
---|
| 495 | else:
|
---|
| 496 | return workscan
|
---|
| 497 |
|
---|
[1862] | 498 | @asaplog_post_dec
|
---|
[1907] | 499 | def stats(self, stat='stddev', mask=None, form='3.3f', row=None):
|
---|
[1846] | 500 | """\
|
---|
[135] | 501 | Determine the specified statistic of the current beam/if/pol
|
---|
[102] | 502 | Takes a 'mask' as an optional parameter to specify which
|
---|
| 503 | channels should be excluded.
|
---|
[1846] | 504 |
|
---|
[102] | 505 | Parameters:
|
---|
[1846] | 506 |
|
---|
[1819] | 507 | stat: 'min', 'max', 'min_abc', 'max_abc', 'sumsq', 'sum',
|
---|
| 508 | 'mean', 'var', 'stddev', 'avdev', 'rms', 'median'
|
---|
[1855] | 509 |
|
---|
[135] | 510 | mask: an optional mask specifying where the statistic
|
---|
[102] | 511 | should be determined.
|
---|
[1855] | 512 |
|
---|
[1819] | 513 | form: format string to print statistic values
|
---|
[1846] | 514 |
|
---|
[1907] | 515 | row: row number of spectrum to process.
|
---|
| 516 | (default is None: for all rows)
|
---|
[1846] | 517 |
|
---|
[1907] | 518 | Example:
|
---|
[113] | 519 | scan.set_unit('channel')
|
---|
[1118] | 520 | msk = scan.create_mask([100, 200], [500, 600])
|
---|
[135] | 521 | scan.stats(stat='mean', mask=m)
|
---|
[1846] | 522 |
|
---|
[102] | 523 | """
|
---|
[1593] | 524 | mask = mask or []
|
---|
[876] | 525 | if not self._check_ifs():
|
---|
[1118] | 526 | raise ValueError("Cannot apply mask as the IFs have different "
|
---|
| 527 | "number of channels. Please use setselection() "
|
---|
| 528 | "to select individual IFs")
|
---|
[1819] | 529 | rtnabc = False
|
---|
| 530 | if stat.lower().endswith('_abc'): rtnabc = True
|
---|
| 531 | getchan = False
|
---|
| 532 | if stat.lower().startswith('min') or stat.lower().startswith('max'):
|
---|
| 533 | chan = self._math._minmaxchan(self, mask, stat)
|
---|
| 534 | getchan = True
|
---|
| 535 | statvals = []
|
---|
[1907] | 536 | if not rtnabc:
|
---|
| 537 | if row == None:
|
---|
| 538 | statvals = self._math._stats(self, mask, stat)
|
---|
| 539 | else:
|
---|
| 540 | statvals = self._math._statsrow(self, mask, stat, int(row))
|
---|
[256] | 541 |
|
---|
[1819] | 542 | #def cb(i):
|
---|
| 543 | # return statvals[i]
|
---|
[256] | 544 |
|
---|
[1819] | 545 | #return self._row_callback(cb, stat)
|
---|
[102] | 546 |
|
---|
[1819] | 547 | label=stat
|
---|
| 548 | #callback=cb
|
---|
| 549 | out = ""
|
---|
| 550 | #outvec = []
|
---|
| 551 | sep = '-'*50
|
---|
[1907] | 552 |
|
---|
| 553 | if row == None:
|
---|
| 554 | rows = xrange(self.nrow())
|
---|
| 555 | elif isinstance(row, int):
|
---|
| 556 | rows = [ row ]
|
---|
| 557 |
|
---|
| 558 | for i in rows:
|
---|
[1819] | 559 | refstr = ''
|
---|
| 560 | statunit= ''
|
---|
| 561 | if getchan:
|
---|
| 562 | qx, qy = self.chan2data(rowno=i, chan=chan[i])
|
---|
| 563 | if rtnabc:
|
---|
| 564 | statvals.append(qx['value'])
|
---|
| 565 | refstr = ('(value: %'+form) % (qy['value'])+' ['+qy['unit']+'])'
|
---|
| 566 | statunit= '['+qx['unit']+']'
|
---|
| 567 | else:
|
---|
| 568 | refstr = ('(@ %'+form) % (qx['value'])+' ['+qx['unit']+'])'
|
---|
| 569 |
|
---|
| 570 | tm = self._gettime(i)
|
---|
| 571 | src = self._getsourcename(i)
|
---|
| 572 | out += 'Scan[%d] (%s) ' % (self.getscan(i), src)
|
---|
| 573 | out += 'Time[%s]:\n' % (tm)
|
---|
[1907] | 574 | if self.nbeam(-1) > 1: out += ' Beam[%d] ' % (self.getbeam(i))
|
---|
| 575 | if self.nif(-1) > 1: out += ' IF[%d] ' % (self.getif(i))
|
---|
| 576 | if self.npol(-1) > 1: out += ' Pol[%d] ' % (self.getpol(i))
|
---|
[1819] | 577 | #outvec.append(callback(i))
|
---|
[1907] | 578 | if len(rows) > 1:
|
---|
| 579 | # out += ('= %'+form) % (outvec[i]) +' '+refstr+'\n'
|
---|
| 580 | out += ('= %'+form) % (statvals[i]) +' '+refstr+'\n'
|
---|
| 581 | else:
|
---|
| 582 | # out += ('= %'+form) % (outvec[0]) +' '+refstr+'\n'
|
---|
| 583 | out += ('= %'+form) % (statvals[0]) +' '+refstr+'\n'
|
---|
[1819] | 584 | out += sep+"\n"
|
---|
| 585 |
|
---|
[1859] | 586 | import os
|
---|
| 587 | if os.environ.has_key( 'USER' ):
|
---|
| 588 | usr = os.environ['USER']
|
---|
| 589 | else:
|
---|
| 590 | import commands
|
---|
| 591 | usr = commands.getoutput( 'whoami' )
|
---|
| 592 | tmpfile = '/tmp/tmp_'+usr+'_casapy_asap_scantable_stats'
|
---|
| 593 | f = open(tmpfile,'w')
|
---|
| 594 | print >> f, sep
|
---|
| 595 | print >> f, ' %s %s' % (label, statunit)
|
---|
| 596 | print >> f, sep
|
---|
| 597 | print >> f, out
|
---|
| 598 | f.close()
|
---|
| 599 | f = open(tmpfile,'r')
|
---|
| 600 | x = f.readlines()
|
---|
| 601 | f.close()
|
---|
| 602 | asaplog.push(''.join(x), False)
|
---|
| 603 |
|
---|
[1819] | 604 | return statvals
|
---|
| 605 |
|
---|
| 606 | def chan2data(self, rowno=0, chan=0):
|
---|
[1846] | 607 | """\
|
---|
[1819] | 608 | Returns channel/frequency/velocity and spectral value
|
---|
| 609 | at an arbitrary row and channel in the scantable.
|
---|
[1846] | 610 |
|
---|
[1819] | 611 | Parameters:
|
---|
[1846] | 612 |
|
---|
[1819] | 613 | rowno: a row number in the scantable. Default is the
|
---|
| 614 | first row, i.e. rowno=0
|
---|
[1855] | 615 |
|
---|
[1819] | 616 | chan: a channel in the scantable. Default is the first
|
---|
| 617 | channel, i.e. pos=0
|
---|
[1846] | 618 |
|
---|
[1819] | 619 | """
|
---|
| 620 | if isinstance(rowno, int) and isinstance(chan, int):
|
---|
| 621 | qx = {'unit': self.get_unit(),
|
---|
| 622 | 'value': self._getabcissa(rowno)[chan]}
|
---|
| 623 | qy = {'unit': self.get_fluxunit(),
|
---|
| 624 | 'value': self._getspectrum(rowno)[chan]}
|
---|
| 625 | return qx, qy
|
---|
| 626 |
|
---|
[1118] | 627 | def stddev(self, mask=None):
|
---|
[1846] | 628 | """\
|
---|
[135] | 629 | Determine the standard deviation of the current beam/if/pol
|
---|
| 630 | Takes a 'mask' as an optional parameter to specify which
|
---|
| 631 | channels should be excluded.
|
---|
[1846] | 632 |
|
---|
[135] | 633 | Parameters:
|
---|
[1846] | 634 |
|
---|
[135] | 635 | mask: an optional mask specifying where the standard
|
---|
| 636 | deviation should be determined.
|
---|
| 637 |
|
---|
[1846] | 638 | Example::
|
---|
| 639 |
|
---|
[135] | 640 | scan.set_unit('channel')
|
---|
[1118] | 641 | msk = scan.create_mask([100, 200], [500, 600])
|
---|
[135] | 642 | scan.stddev(mask=m)
|
---|
[1846] | 643 |
|
---|
[135] | 644 | """
|
---|
[1118] | 645 | return self.stats(stat='stddev', mask=mask);
|
---|
[135] | 646 |
|
---|
[1003] | 647 |
|
---|
[1259] | 648 | def get_column_names(self):
|
---|
[1846] | 649 | """\
|
---|
[1003] | 650 | Return a list of column names, which can be used for selection.
|
---|
| 651 | """
|
---|
[1259] | 652 | return list(Scantable.get_column_names(self))
|
---|
[1003] | 653 |
|
---|
[1730] | 654 | def get_tsys(self, row=-1):
|
---|
[1846] | 655 | """\
|
---|
[113] | 656 | Return the System temperatures.
|
---|
[1846] | 657 |
|
---|
| 658 | Parameters:
|
---|
| 659 |
|
---|
| 660 | row: the rowno to get the information for. (default all rows)
|
---|
| 661 |
|
---|
[113] | 662 | Returns:
|
---|
[1846] | 663 |
|
---|
[876] | 664 | a list of Tsys values for the current selection
|
---|
[1846] | 665 |
|
---|
[113] | 666 | """
|
---|
[1730] | 667 | if row > -1:
|
---|
| 668 | return self._get_column(self._gettsys, row)
|
---|
[876] | 669 | return self._row_callback(self._gettsys, "Tsys")
|
---|
[256] | 670 |
|
---|
[1730] | 671 |
|
---|
| 672 | def get_weather(self, row=-1):
|
---|
[1846] | 673 | """\
|
---|
| 674 | Return the weather informations.
|
---|
| 675 |
|
---|
| 676 | Parameters:
|
---|
| 677 |
|
---|
| 678 | row: the rowno to get the information for. (default all rows)
|
---|
| 679 |
|
---|
| 680 | Returns:
|
---|
| 681 |
|
---|
| 682 | a dict or list of of dicts of values for the current selection
|
---|
| 683 |
|
---|
| 684 | """
|
---|
| 685 |
|
---|
[1730] | 686 | values = self._get_column(self._get_weather, row)
|
---|
| 687 | if row > -1:
|
---|
| 688 | return {'temperature': values[0],
|
---|
| 689 | 'pressure': values[1], 'humidity' : values[2],
|
---|
| 690 | 'windspeed' : values[3], 'windaz' : values[4]
|
---|
| 691 | }
|
---|
| 692 | else:
|
---|
| 693 | out = []
|
---|
| 694 | for r in values:
|
---|
| 695 |
|
---|
| 696 | out.append({'temperature': r[0],
|
---|
| 697 | 'pressure': r[1], 'humidity' : r[2],
|
---|
| 698 | 'windspeed' : r[3], 'windaz' : r[4]
|
---|
| 699 | })
|
---|
| 700 | return out
|
---|
| 701 |
|
---|
[876] | 702 | def _row_callback(self, callback, label):
|
---|
| 703 | out = ""
|
---|
[1118] | 704 | outvec = []
|
---|
[1590] | 705 | sep = '-'*50
|
---|
[876] | 706 | for i in range(self.nrow()):
|
---|
| 707 | tm = self._gettime(i)
|
---|
| 708 | src = self._getsourcename(i)
|
---|
[1590] | 709 | out += 'Scan[%d] (%s) ' % (self.getscan(i), src)
|
---|
[876] | 710 | out += 'Time[%s]:\n' % (tm)
|
---|
[1590] | 711 | if self.nbeam(-1) > 1:
|
---|
| 712 | out += ' Beam[%d] ' % (self.getbeam(i))
|
---|
| 713 | if self.nif(-1) > 1: out += ' IF[%d] ' % (self.getif(i))
|
---|
| 714 | if self.npol(-1) > 1: out += ' Pol[%d] ' % (self.getpol(i))
|
---|
[876] | 715 | outvec.append(callback(i))
|
---|
| 716 | out += '= %3.3f\n' % (outvec[i])
|
---|
[1590] | 717 | out += sep+'\n'
|
---|
[1859] | 718 |
|
---|
| 719 | asaplog.push(sep)
|
---|
| 720 | asaplog.push(" %s" % (label))
|
---|
| 721 | asaplog.push(sep)
|
---|
| 722 | asaplog.push(out)
|
---|
[1861] | 723 | asaplog.post()
|
---|
[1175] | 724 | return outvec
|
---|
[256] | 725 |
|
---|
[1070] | 726 | def _get_column(self, callback, row=-1):
|
---|
| 727 | """
|
---|
| 728 | """
|
---|
| 729 | if row == -1:
|
---|
| 730 | return [callback(i) for i in range(self.nrow())]
|
---|
| 731 | else:
|
---|
[1819] | 732 | if 0 <= row < self.nrow():
|
---|
[1070] | 733 | return callback(row)
|
---|
[256] | 734 |
|
---|
[1070] | 735 |
|
---|
[1348] | 736 | def get_time(self, row=-1, asdatetime=False):
|
---|
[1846] | 737 | """\
|
---|
[113] | 738 | Get a list of time stamps for the observations.
|
---|
[1348] | 739 | Return a datetime object for each integration time stamp in the scantable.
|
---|
[1846] | 740 |
|
---|
[113] | 741 | Parameters:
|
---|
[1846] | 742 |
|
---|
[1348] | 743 | row: row no of integration. Default -1 return all rows
|
---|
[1855] | 744 |
|
---|
[1348] | 745 | asdatetime: return values as datetime objects rather than strings
|
---|
[1846] | 746 |
|
---|
[113] | 747 | """
|
---|
[1175] | 748 | from time import strptime
|
---|
| 749 | from datetime import datetime
|
---|
[1392] | 750 | times = self._get_column(self._gettime, row)
|
---|
[1348] | 751 | if not asdatetime:
|
---|
[1392] | 752 | return times
|
---|
[1175] | 753 | format = "%Y/%m/%d/%H:%M:%S"
|
---|
| 754 | if isinstance(times, list):
|
---|
| 755 | return [datetime(*strptime(i, format)[:6]) for i in times]
|
---|
| 756 | else:
|
---|
| 757 | return datetime(*strptime(times, format)[:6])
|
---|
[102] | 758 |
|
---|
[1348] | 759 |
|
---|
| 760 | def get_inttime(self, row=-1):
|
---|
[1846] | 761 | """\
|
---|
[1348] | 762 | Get a list of integration times for the observations.
|
---|
| 763 | Return a time in seconds for each integration in the scantable.
|
---|
[1846] | 764 |
|
---|
[1348] | 765 | Parameters:
|
---|
[1846] | 766 |
|
---|
[1348] | 767 | row: row no of integration. Default -1 return all rows.
|
---|
[1846] | 768 |
|
---|
[1348] | 769 | """
|
---|
[1573] | 770 | return self._get_column(self._getinttime, row)
|
---|
[1348] | 771 |
|
---|
[1573] | 772 |
|
---|
[714] | 773 | def get_sourcename(self, row=-1):
|
---|
[1846] | 774 | """\
|
---|
[794] | 775 | Get a list source names for the observations.
|
---|
[714] | 776 | Return a string for each integration in the scantable.
|
---|
| 777 | Parameters:
|
---|
[1846] | 778 |
|
---|
[1348] | 779 | row: row no of integration. Default -1 return all rows.
|
---|
[1846] | 780 |
|
---|
[714] | 781 | """
|
---|
[1070] | 782 | return self._get_column(self._getsourcename, row)
|
---|
[714] | 783 |
|
---|
[794] | 784 | def get_elevation(self, row=-1):
|
---|
[1846] | 785 | """\
|
---|
[794] | 786 | Get a list of elevations for the observations.
|
---|
| 787 | Return a float for each integration in the scantable.
|
---|
[1846] | 788 |
|
---|
[794] | 789 | Parameters:
|
---|
[1846] | 790 |
|
---|
[1348] | 791 | row: row no of integration. Default -1 return all rows.
|
---|
[1846] | 792 |
|
---|
[794] | 793 | """
|
---|
[1070] | 794 | return self._get_column(self._getelevation, row)
|
---|
[794] | 795 |
|
---|
| 796 | def get_azimuth(self, row=-1):
|
---|
[1846] | 797 | """\
|
---|
[794] | 798 | Get a list of azimuths for the observations.
|
---|
| 799 | Return a float for each integration in the scantable.
|
---|
[1846] | 800 |
|
---|
[794] | 801 | Parameters:
|
---|
[1348] | 802 | row: row no of integration. Default -1 return all rows.
|
---|
[1846] | 803 |
|
---|
[794] | 804 | """
|
---|
[1070] | 805 | return self._get_column(self._getazimuth, row)
|
---|
[794] | 806 |
|
---|
| 807 | def get_parangle(self, row=-1):
|
---|
[1846] | 808 | """\
|
---|
[794] | 809 | Get a list of parallactic angles for the observations.
|
---|
| 810 | Return a float for each integration in the scantable.
|
---|
[1846] | 811 |
|
---|
[794] | 812 | Parameters:
|
---|
[1846] | 813 |
|
---|
[1348] | 814 | row: row no of integration. Default -1 return all rows.
|
---|
[1846] | 815 |
|
---|
[794] | 816 | """
|
---|
[1070] | 817 | return self._get_column(self._getparangle, row)
|
---|
[794] | 818 |
|
---|
[1070] | 819 | def get_direction(self, row=-1):
|
---|
| 820 | """
|
---|
| 821 | Get a list of Positions on the sky (direction) for the observations.
|
---|
[1594] | 822 | Return a string for each integration in the scantable.
|
---|
[1855] | 823 |
|
---|
[1070] | 824 | Parameters:
|
---|
[1855] | 825 |
|
---|
[1070] | 826 | row: row no of integration. Default -1 return all rows
|
---|
[1855] | 827 |
|
---|
[1070] | 828 | """
|
---|
| 829 | return self._get_column(self._getdirection, row)
|
---|
| 830 |
|
---|
[1391] | 831 | def get_directionval(self, row=-1):
|
---|
[1846] | 832 | """\
|
---|
[1391] | 833 | Get a list of Positions on the sky (direction) for the observations.
|
---|
| 834 | Return a float for each integration in the scantable.
|
---|
[1846] | 835 |
|
---|
[1391] | 836 | Parameters:
|
---|
[1846] | 837 |
|
---|
[1391] | 838 | row: row no of integration. Default -1 return all rows
|
---|
[1846] | 839 |
|
---|
[1391] | 840 | """
|
---|
| 841 | return self._get_column(self._getdirectionvec, row)
|
---|
| 842 |
|
---|
[1862] | 843 | @asaplog_post_dec
|
---|
[102] | 844 | def set_unit(self, unit='channel'):
|
---|
[1846] | 845 | """\
|
---|
[102] | 846 | Set the unit for all following operations on this scantable
|
---|
[1846] | 847 |
|
---|
[102] | 848 | Parameters:
|
---|
[1846] | 849 |
|
---|
| 850 | unit: optional unit, default is 'channel'. Use one of '*Hz',
|
---|
| 851 | 'km/s', 'channel' or equivalent ''
|
---|
| 852 |
|
---|
[102] | 853 | """
|
---|
[484] | 854 | varlist = vars()
|
---|
[1118] | 855 | if unit in ['', 'pixel', 'channel']:
|
---|
[113] | 856 | unit = ''
|
---|
| 857 | inf = list(self._getcoordinfo())
|
---|
| 858 | inf[0] = unit
|
---|
| 859 | self._setcoordinfo(inf)
|
---|
[1118] | 860 | self._add_history("set_unit", varlist)
|
---|
[113] | 861 |
|
---|
[1862] | 862 | @asaplog_post_dec
|
---|
[484] | 863 | def set_instrument(self, instr):
|
---|
[1846] | 864 | """\
|
---|
[1348] | 865 | Set the instrument for subsequent processing.
|
---|
[1846] | 866 |
|
---|
[358] | 867 | Parameters:
|
---|
[1846] | 868 |
|
---|
[710] | 869 | instr: Select from 'ATPKSMB', 'ATPKSHOH', 'ATMOPRA',
|
---|
[407] | 870 | 'DSS-43' (Tid), 'CEDUNA', and 'HOBART'
|
---|
[1846] | 871 |
|
---|
[358] | 872 | """
|
---|
| 873 | self._setInstrument(instr)
|
---|
[1118] | 874 | self._add_history("set_instument", vars())
|
---|
[358] | 875 |
|
---|
[1862] | 876 | @asaplog_post_dec
|
---|
[1190] | 877 | def set_feedtype(self, feedtype):
|
---|
[1846] | 878 | """\
|
---|
[1190] | 879 | Overwrite the feed type, which might not be set correctly.
|
---|
[1846] | 880 |
|
---|
[1190] | 881 | Parameters:
|
---|
[1846] | 882 |
|
---|
[1190] | 883 | feedtype: 'linear' or 'circular'
|
---|
[1846] | 884 |
|
---|
[1190] | 885 | """
|
---|
| 886 | self._setfeedtype(feedtype)
|
---|
| 887 | self._add_history("set_feedtype", vars())
|
---|
| 888 |
|
---|
[1862] | 889 | @asaplog_post_dec
|
---|
[276] | 890 | def set_doppler(self, doppler='RADIO'):
|
---|
[1846] | 891 | """\
|
---|
[276] | 892 | Set the doppler for all following operations on this scantable.
|
---|
[1846] | 893 |
|
---|
[276] | 894 | Parameters:
|
---|
[1846] | 895 |
|
---|
[276] | 896 | doppler: One of 'RADIO', 'OPTICAL', 'Z', 'BETA', 'GAMMA'
|
---|
[1846] | 897 |
|
---|
[276] | 898 | """
|
---|
[484] | 899 | varlist = vars()
|
---|
[276] | 900 | inf = list(self._getcoordinfo())
|
---|
| 901 | inf[2] = doppler
|
---|
| 902 | self._setcoordinfo(inf)
|
---|
[1118] | 903 | self._add_history("set_doppler", vars())
|
---|
[710] | 904 |
|
---|
[1862] | 905 | @asaplog_post_dec
|
---|
[226] | 906 | def set_freqframe(self, frame=None):
|
---|
[1846] | 907 | """\
|
---|
[113] | 908 | Set the frame type of the Spectral Axis.
|
---|
[1846] | 909 |
|
---|
[113] | 910 | Parameters:
|
---|
[1846] | 911 |
|
---|
[591] | 912 | frame: an optional frame type, default 'LSRK'. Valid frames are:
|
---|
[1819] | 913 | 'TOPO', 'LSRD', 'LSRK', 'BARY',
|
---|
[1118] | 914 | 'GEO', 'GALACTO', 'LGROUP', 'CMB'
|
---|
[1846] | 915 |
|
---|
| 916 | Example::
|
---|
| 917 |
|
---|
[113] | 918 | scan.set_freqframe('BARY')
|
---|
[1846] | 919 |
|
---|
[113] | 920 | """
|
---|
[1593] | 921 | frame = frame or rcParams['scantable.freqframe']
|
---|
[484] | 922 | varlist = vars()
|
---|
[1819] | 923 | # "REST" is not implemented in casacore
|
---|
| 924 | #valid = ['REST', 'TOPO', 'LSRD', 'LSRK', 'BARY', \
|
---|
| 925 | # 'GEO', 'GALACTO', 'LGROUP', 'CMB']
|
---|
| 926 | valid = ['TOPO', 'LSRD', 'LSRK', 'BARY', \
|
---|
[1118] | 927 | 'GEO', 'GALACTO', 'LGROUP', 'CMB']
|
---|
[591] | 928 |
|
---|
[989] | 929 | if frame in valid:
|
---|
[113] | 930 | inf = list(self._getcoordinfo())
|
---|
| 931 | inf[1] = frame
|
---|
| 932 | self._setcoordinfo(inf)
|
---|
[1118] | 933 | self._add_history("set_freqframe", varlist)
|
---|
[102] | 934 | else:
|
---|
[1118] | 935 | msg = "Please specify a valid freq type. Valid types are:\n", valid
|
---|
[1859] | 936 | raise TypeError(msg)
|
---|
[710] | 937 |
|
---|
[1862] | 938 | @asaplog_post_dec
|
---|
[989] | 939 | def set_dirframe(self, frame=""):
|
---|
[1846] | 940 | """\
|
---|
[989] | 941 | Set the frame type of the Direction on the sky.
|
---|
[1846] | 942 |
|
---|
[989] | 943 | Parameters:
|
---|
[1846] | 944 |
|
---|
[989] | 945 | frame: an optional frame type, default ''. Valid frames are:
|
---|
| 946 | 'J2000', 'B1950', 'GALACTIC'
|
---|
[1846] | 947 |
|
---|
| 948 | Example:
|
---|
| 949 |
|
---|
[989] | 950 | scan.set_dirframe('GALACTIC')
|
---|
[1846] | 951 |
|
---|
[989] | 952 | """
|
---|
| 953 | varlist = vars()
|
---|
[1859] | 954 | Scantable.set_dirframe(self, frame)
|
---|
[1118] | 955 | self._add_history("set_dirframe", varlist)
|
---|
[989] | 956 |
|
---|
[113] | 957 | def get_unit(self):
|
---|
[1846] | 958 | """\
|
---|
[113] | 959 | Get the default unit set in this scantable
|
---|
[1846] | 960 |
|
---|
[113] | 961 | Returns:
|
---|
[1846] | 962 |
|
---|
[113] | 963 | A unit string
|
---|
[1846] | 964 |
|
---|
[113] | 965 | """
|
---|
| 966 | inf = self._getcoordinfo()
|
---|
| 967 | unit = inf[0]
|
---|
| 968 | if unit == '': unit = 'channel'
|
---|
| 969 | return unit
|
---|
[102] | 970 |
|
---|
[1862] | 971 | @asaplog_post_dec
|
---|
[158] | 972 | def get_abcissa(self, rowno=0):
|
---|
[1846] | 973 | """\
|
---|
[158] | 974 | Get the abcissa in the current coordinate setup for the currently
|
---|
[113] | 975 | selected Beam/IF/Pol
|
---|
[1846] | 976 |
|
---|
[113] | 977 | Parameters:
|
---|
[1846] | 978 |
|
---|
[226] | 979 | rowno: an optional row number in the scantable. Default is the
|
---|
| 980 | first row, i.e. rowno=0
|
---|
[1846] | 981 |
|
---|
[113] | 982 | Returns:
|
---|
[1846] | 983 |
|
---|
[1348] | 984 | The abcissa values and the format string (as a dictionary)
|
---|
[1846] | 985 |
|
---|
[113] | 986 | """
|
---|
[256] | 987 | abc = self._getabcissa(rowno)
|
---|
[710] | 988 | lbl = self._getabcissalabel(rowno)
|
---|
[158] | 989 | return abc, lbl
|
---|
[113] | 990 |
|
---|
[1862] | 991 | @asaplog_post_dec
|
---|
[1819] | 992 | def flag(self, mask=None, unflag=False):
|
---|
[1846] | 993 | """\
|
---|
[1001] | 994 | Flag the selected data using an optional channel mask.
|
---|
[1846] | 995 |
|
---|
[1001] | 996 | Parameters:
|
---|
[1846] | 997 |
|
---|
[1001] | 998 | mask: an optional channel mask, created with create_mask. Default
|
---|
| 999 | (no mask) is all channels.
|
---|
[1855] | 1000 |
|
---|
[1819] | 1001 | unflag: if True, unflag the data
|
---|
[1846] | 1002 |
|
---|
[1001] | 1003 | """
|
---|
| 1004 | varlist = vars()
|
---|
[1593] | 1005 | mask = mask or []
|
---|
[1859] | 1006 | self._flag(mask, unflag)
|
---|
[1001] | 1007 | self._add_history("flag", varlist)
|
---|
| 1008 |
|
---|
[1862] | 1009 | @asaplog_post_dec
|
---|
[1819] | 1010 | def flag_row(self, rows=[], unflag=False):
|
---|
[1846] | 1011 | """\
|
---|
[1819] | 1012 | Flag the selected data in row-based manner.
|
---|
[1846] | 1013 |
|
---|
[1819] | 1014 | Parameters:
|
---|
[1846] | 1015 |
|
---|
[1843] | 1016 | rows: list of row numbers to be flagged. Default is no row
|
---|
| 1017 | (must be explicitly specified to execute row-based flagging).
|
---|
[1855] | 1018 |
|
---|
[1819] | 1019 | unflag: if True, unflag the data.
|
---|
[1846] | 1020 |
|
---|
[1819] | 1021 | """
|
---|
| 1022 | varlist = vars()
|
---|
[1859] | 1023 | self._flag_row(rows, unflag)
|
---|
[1819] | 1024 | self._add_history("flag_row", varlist)
|
---|
| 1025 |
|
---|
[1862] | 1026 | @asaplog_post_dec
|
---|
[1819] | 1027 | def clip(self, uthres=None, dthres=None, clipoutside=True, unflag=False):
|
---|
[1846] | 1028 | """\
|
---|
[1819] | 1029 | Flag the selected data outside a specified range (in channel-base)
|
---|
[1846] | 1030 |
|
---|
[1819] | 1031 | Parameters:
|
---|
[1846] | 1032 |
|
---|
[1819] | 1033 | uthres: upper threshold.
|
---|
[1855] | 1034 |
|
---|
[1819] | 1035 | dthres: lower threshold
|
---|
[1846] | 1036 |
|
---|
[1819] | 1037 | clipoutside: True for flagging data outside the range [dthres:uthres].
|
---|
| 1038 | False for glagging data inside the range.
|
---|
[1855] | 1039 |
|
---|
[1846] | 1040 | unflag: if True, unflag the data.
|
---|
| 1041 |
|
---|
[1819] | 1042 | """
|
---|
| 1043 | varlist = vars()
|
---|
[1859] | 1044 | self._clip(uthres, dthres, clipoutside, unflag)
|
---|
[1819] | 1045 | self._add_history("clip", varlist)
|
---|
| 1046 |
|
---|
[1862] | 1047 | @asaplog_post_dec
|
---|
[1584] | 1048 | def lag_flag(self, start, end, unit="MHz", insitu=None):
|
---|
[1846] | 1049 | """\
|
---|
[1192] | 1050 | Flag the data in 'lag' space by providing a frequency to remove.
|
---|
[1584] | 1051 | Flagged data in the scantable gets interpolated over the region.
|
---|
[1192] | 1052 | No taper is applied.
|
---|
[1846] | 1053 |
|
---|
[1192] | 1054 | Parameters:
|
---|
[1846] | 1055 |
|
---|
[1579] | 1056 | start: the start frequency (really a period within the
|
---|
| 1057 | bandwidth) or period to remove
|
---|
[1855] | 1058 |
|
---|
[1579] | 1059 | end: the end frequency or period to remove
|
---|
[1855] | 1060 |
|
---|
[1584] | 1061 | unit: the frequency unit (default "MHz") or "" for
|
---|
[1579] | 1062 | explicit lag channels
|
---|
[1846] | 1063 |
|
---|
| 1064 | *Notes*:
|
---|
| 1065 |
|
---|
[1579] | 1066 | It is recommended to flag edges of the band or strong
|
---|
[1348] | 1067 | signals beforehand.
|
---|
[1846] | 1068 |
|
---|
[1192] | 1069 | """
|
---|
| 1070 | if insitu is None: insitu = rcParams['insitu']
|
---|
| 1071 | self._math._setinsitu(insitu)
|
---|
| 1072 | varlist = vars()
|
---|
[1579] | 1073 | base = { "GHz": 1000000000., "MHz": 1000000., "kHz": 1000., "Hz": 1.}
|
---|
| 1074 | if not (unit == "" or base.has_key(unit)):
|
---|
[1192] | 1075 | raise ValueError("%s is not a valid unit." % unit)
|
---|
[1859] | 1076 | if unit == "":
|
---|
| 1077 | s = scantable(self._math._lag_flag(self, start, end, "lags"))
|
---|
| 1078 | else:
|
---|
| 1079 | s = scantable(self._math._lag_flag(self, start*base[unit],
|
---|
| 1080 | end*base[unit], "frequency"))
|
---|
[1192] | 1081 | s._add_history("lag_flag", varlist)
|
---|
| 1082 | if insitu:
|
---|
| 1083 | self._assign(s)
|
---|
| 1084 | else:
|
---|
| 1085 | return s
|
---|
[1001] | 1086 |
|
---|
[1862] | 1087 | @asaplog_post_dec
|
---|
[113] | 1088 | def create_mask(self, *args, **kwargs):
|
---|
[1846] | 1089 | """\
|
---|
[1118] | 1090 | Compute and return a mask based on [min, max] windows.
|
---|
[189] | 1091 | The specified windows are to be INCLUDED, when the mask is
|
---|
[113] | 1092 | applied.
|
---|
[1846] | 1093 |
|
---|
[102] | 1094 | Parameters:
|
---|
[1846] | 1095 |
|
---|
[1118] | 1096 | [min, max], [min2, max2], ...
|
---|
[1024] | 1097 | Pairs of start/end points (inclusive)specifying the regions
|
---|
[102] | 1098 | to be masked
|
---|
[1855] | 1099 |
|
---|
[189] | 1100 | invert: optional argument. If specified as True,
|
---|
| 1101 | return an inverted mask, i.e. the regions
|
---|
| 1102 | specified are EXCLUDED
|
---|
[1855] | 1103 |
|
---|
[513] | 1104 | row: create the mask using the specified row for
|
---|
| 1105 | unit conversions, default is row=0
|
---|
| 1106 | only necessary if frequency varies over rows.
|
---|
[1846] | 1107 |
|
---|
| 1108 | Examples::
|
---|
| 1109 |
|
---|
[113] | 1110 | scan.set_unit('channel')
|
---|
[1846] | 1111 | # a)
|
---|
[1118] | 1112 | msk = scan.create_mask([400, 500], [800, 900])
|
---|
[189] | 1113 | # masks everything outside 400 and 500
|
---|
[113] | 1114 | # and 800 and 900 in the unit 'channel'
|
---|
| 1115 |
|
---|
[1846] | 1116 | # b)
|
---|
[1118] | 1117 | msk = scan.create_mask([400, 500], [800, 900], invert=True)
|
---|
[189] | 1118 | # masks the regions between 400 and 500
|
---|
[113] | 1119 | # and 800 and 900 in the unit 'channel'
|
---|
[1846] | 1120 |
|
---|
| 1121 | # c)
|
---|
| 1122 | #mask only channel 400
|
---|
[1554] | 1123 | msk = scan.create_mask([400])
|
---|
[1846] | 1124 |
|
---|
[102] | 1125 | """
|
---|
[1554] | 1126 | row = kwargs.get("row", 0)
|
---|
[513] | 1127 | data = self._getabcissa(row)
|
---|
[113] | 1128 | u = self._getcoordinfo()[0]
|
---|
[1859] | 1129 | if u == "":
|
---|
| 1130 | u = "channel"
|
---|
| 1131 | msg = "The current mask window unit is %s" % u
|
---|
| 1132 | i = self._check_ifs()
|
---|
| 1133 | if not i:
|
---|
| 1134 | msg += "\nThis mask is only valid for IF=%d" % (self.getif(i))
|
---|
| 1135 | asaplog.push(msg)
|
---|
[102] | 1136 | n = self.nchan()
|
---|
[1295] | 1137 | msk = _n_bools(n, False)
|
---|
[710] | 1138 | # test if args is a 'list' or a 'normal *args - UGLY!!!
|
---|
| 1139 |
|
---|
[1118] | 1140 | ws = (isinstance(args[-1][-1], int) or isinstance(args[-1][-1], float)) \
|
---|
| 1141 | and args or args[0]
|
---|
[710] | 1142 | for window in ws:
|
---|
[1554] | 1143 | if len(window) == 1:
|
---|
| 1144 | window = [window[0], window[0]]
|
---|
| 1145 | if len(window) == 0 or len(window) > 2:
|
---|
| 1146 | raise ValueError("A window needs to be defined as [start(, end)]")
|
---|
[1545] | 1147 | if window[0] > window[1]:
|
---|
| 1148 | tmp = window[0]
|
---|
| 1149 | window[0] = window[1]
|
---|
| 1150 | window[1] = tmp
|
---|
[102] | 1151 | for i in range(n):
|
---|
[1024] | 1152 | if data[i] >= window[0] and data[i] <= window[1]:
|
---|
[1295] | 1153 | msk[i] = True
|
---|
[113] | 1154 | if kwargs.has_key('invert'):
|
---|
| 1155 | if kwargs.get('invert'):
|
---|
[1295] | 1156 | msk = mask_not(msk)
|
---|
[102] | 1157 | return msk
|
---|
[710] | 1158 |
|
---|
[1819] | 1159 | def get_masklist(self, mask=None, row=0):
|
---|
[1846] | 1160 | """\
|
---|
[1819] | 1161 | Compute and return a list of mask windows, [min, max].
|
---|
[1846] | 1162 |
|
---|
[1819] | 1163 | Parameters:
|
---|
[1846] | 1164 |
|
---|
[1819] | 1165 | mask: channel mask, created with create_mask.
|
---|
[1855] | 1166 |
|
---|
[1819] | 1167 | row: calcutate the masklist using the specified row
|
---|
| 1168 | for unit conversions, default is row=0
|
---|
| 1169 | only necessary if frequency varies over rows.
|
---|
[1846] | 1170 |
|
---|
[1819] | 1171 | Returns:
|
---|
[1846] | 1172 |
|
---|
[1819] | 1173 | [min, max], [min2, max2], ...
|
---|
| 1174 | Pairs of start/end points (inclusive)specifying
|
---|
| 1175 | the masked regions
|
---|
[1846] | 1176 |
|
---|
[1819] | 1177 | """
|
---|
| 1178 | if not (isinstance(mask,list) or isinstance(mask, tuple)):
|
---|
| 1179 | raise TypeError("The mask should be list or tuple.")
|
---|
| 1180 | if len(mask) < 2:
|
---|
| 1181 | raise TypeError("The mask elements should be > 1")
|
---|
| 1182 | if self.nchan() != len(mask):
|
---|
| 1183 | msg = "Number of channels in scantable != number of mask elements"
|
---|
| 1184 | raise TypeError(msg)
|
---|
| 1185 | data = self._getabcissa(row)
|
---|
| 1186 | u = self._getcoordinfo()[0]
|
---|
[1859] | 1187 | if u == "":
|
---|
| 1188 | u = "channel"
|
---|
| 1189 | msg = "The current mask window unit is %s" % u
|
---|
| 1190 | i = self._check_ifs()
|
---|
| 1191 | if not i:
|
---|
| 1192 | msg += "\nThis mask is only valid for IF=%d" % (self.getif(i))
|
---|
| 1193 | asaplog.push(msg)
|
---|
[1819] | 1194 | masklist=[]
|
---|
| 1195 | ist, ien = None, None
|
---|
| 1196 | ist, ien=self.get_mask_indices(mask)
|
---|
| 1197 | if ist is not None and ien is not None:
|
---|
| 1198 | for i in xrange(len(ist)):
|
---|
| 1199 | range=[data[ist[i]],data[ien[i]]]
|
---|
| 1200 | range.sort()
|
---|
| 1201 | masklist.append([range[0],range[1]])
|
---|
| 1202 | return masklist
|
---|
| 1203 |
|
---|
| 1204 | def get_mask_indices(self, mask=None):
|
---|
[1846] | 1205 | """\
|
---|
[1819] | 1206 | Compute and Return lists of mask start indices and mask end indices.
|
---|
[1855] | 1207 |
|
---|
| 1208 | Parameters:
|
---|
| 1209 |
|
---|
[1819] | 1210 | mask: channel mask, created with create_mask.
|
---|
[1846] | 1211 |
|
---|
[1819] | 1212 | Returns:
|
---|
[1846] | 1213 |
|
---|
[1819] | 1214 | List of mask start indices and that of mask end indices,
|
---|
| 1215 | i.e., [istart1,istart2,....], [iend1,iend2,....].
|
---|
[1846] | 1216 |
|
---|
[1819] | 1217 | """
|
---|
| 1218 | if not (isinstance(mask,list) or isinstance(mask, tuple)):
|
---|
| 1219 | raise TypeError("The mask should be list or tuple.")
|
---|
| 1220 | if len(mask) < 2:
|
---|
| 1221 | raise TypeError("The mask elements should be > 1")
|
---|
| 1222 | istart=[]
|
---|
| 1223 | iend=[]
|
---|
| 1224 | if mask[0]: istart.append(0)
|
---|
| 1225 | for i in range(len(mask)-1):
|
---|
| 1226 | if not mask[i] and mask[i+1]:
|
---|
| 1227 | istart.append(i+1)
|
---|
| 1228 | elif mask[i] and not mask[i+1]:
|
---|
| 1229 | iend.append(i)
|
---|
| 1230 | if mask[len(mask)-1]: iend.append(len(mask)-1)
|
---|
| 1231 | if len(istart) != len(iend):
|
---|
| 1232 | raise RuntimeError("Numbers of mask start != mask end.")
|
---|
| 1233 | for i in range(len(istart)):
|
---|
| 1234 | if istart[i] > iend[i]:
|
---|
| 1235 | raise RuntimeError("Mask start index > mask end index")
|
---|
| 1236 | break
|
---|
| 1237 | return istart,iend
|
---|
| 1238 |
|
---|
| 1239 | # def get_restfreqs(self):
|
---|
| 1240 | # """
|
---|
| 1241 | # Get the restfrequency(s) stored in this scantable.
|
---|
| 1242 | # The return value(s) are always of unit 'Hz'
|
---|
| 1243 | # Parameters:
|
---|
| 1244 | # none
|
---|
| 1245 | # Returns:
|
---|
| 1246 | # a list of doubles
|
---|
| 1247 | # """
|
---|
| 1248 | # return list(self._getrestfreqs())
|
---|
| 1249 |
|
---|
| 1250 | def get_restfreqs(self, ids=None):
|
---|
[1846] | 1251 | """\
|
---|
[256] | 1252 | Get the restfrequency(s) stored in this scantable.
|
---|
| 1253 | The return value(s) are always of unit 'Hz'
|
---|
[1846] | 1254 |
|
---|
[256] | 1255 | Parameters:
|
---|
[1846] | 1256 |
|
---|
[1819] | 1257 | ids: (optional) a list of MOLECULE_ID for that restfrequency(s) to
|
---|
| 1258 | be retrieved
|
---|
[1846] | 1259 |
|
---|
[256] | 1260 | Returns:
|
---|
[1846] | 1261 |
|
---|
[1819] | 1262 | dictionary containing ids and a list of doubles for each id
|
---|
[1846] | 1263 |
|
---|
[256] | 1264 | """
|
---|
[1819] | 1265 | if ids is None:
|
---|
| 1266 | rfreqs={}
|
---|
| 1267 | idlist = self.getmolnos()
|
---|
| 1268 | for i in idlist:
|
---|
| 1269 | rfreqs[i]=list(self._getrestfreqs(i))
|
---|
| 1270 | return rfreqs
|
---|
| 1271 | else:
|
---|
| 1272 | if type(ids)==list or type(ids)==tuple:
|
---|
| 1273 | rfreqs={}
|
---|
| 1274 | for i in ids:
|
---|
| 1275 | rfreqs[i]=list(self._getrestfreqs(i))
|
---|
| 1276 | return rfreqs
|
---|
| 1277 | else:
|
---|
| 1278 | return list(self._getrestfreqs(ids))
|
---|
| 1279 | #return list(self._getrestfreqs(ids))
|
---|
[102] | 1280 |
|
---|
[931] | 1281 | def set_restfreqs(self, freqs=None, unit='Hz'):
|
---|
[1846] | 1282 | """\
|
---|
[931] | 1283 | Set or replace the restfrequency specified and
|
---|
| 1284 | If the 'freqs' argument holds a scalar,
|
---|
| 1285 | then that rest frequency will be applied to all the selected
|
---|
| 1286 | data. If the 'freqs' argument holds
|
---|
| 1287 | a vector, then it MUST be of equal or smaller length than
|
---|
| 1288 | the number of IFs (and the available restfrequencies will be
|
---|
| 1289 | replaced by this vector). In this case, *all* data have
|
---|
| 1290 | the restfrequency set per IF according
|
---|
| 1291 | to the corresponding value you give in the 'freqs' vector.
|
---|
[1118] | 1292 | E.g. 'freqs=[1e9, 2e9]' would mean IF 0 gets restfreq 1e9 and
|
---|
[931] | 1293 | IF 1 gets restfreq 2e9.
|
---|
[1846] | 1294 |
|
---|
[1395] | 1295 | You can also specify the frequencies via a linecatalog.
|
---|
[1153] | 1296 |
|
---|
[931] | 1297 | Parameters:
|
---|
[1846] | 1298 |
|
---|
[931] | 1299 | freqs: list of rest frequency values or string idenitfiers
|
---|
[1855] | 1300 |
|
---|
[931] | 1301 | unit: unit for rest frequency (default 'Hz')
|
---|
[402] | 1302 |
|
---|
[1846] | 1303 |
|
---|
| 1304 | Example::
|
---|
| 1305 |
|
---|
[1819] | 1306 | # set the given restfrequency for the all currently selected IFs
|
---|
[931] | 1307 | scan.set_restfreqs(freqs=1.4e9)
|
---|
[1845] | 1308 | # set restfrequencies for the n IFs (n > 1) in the order of the
|
---|
| 1309 | # list, i.e
|
---|
| 1310 | # IF0 -> 1.4e9, IF1 -> 1.41e9, IF3 -> 1.42e9
|
---|
| 1311 | # len(list_of_restfreqs) == nIF
|
---|
| 1312 | # for nIF == 1 the following will set multiple restfrequency for
|
---|
| 1313 | # that IF
|
---|
[1819] | 1314 | scan.set_restfreqs(freqs=[1.4e9, 1.41e9, 1.42e9])
|
---|
[1845] | 1315 | # set multiple restfrequencies per IF. as a list of lists where
|
---|
| 1316 | # the outer list has nIF elements, the inner s arbitrary
|
---|
| 1317 | scan.set_restfreqs(freqs=[[1.4e9, 1.41e9], [1.67e9]])
|
---|
[391] | 1318 |
|
---|
[1846] | 1319 | *Note*:
|
---|
[1845] | 1320 |
|
---|
[931] | 1321 | To do more sophisticate Restfrequency setting, e.g. on a
|
---|
| 1322 | source and IF basis, use scantable.set_selection() before using
|
---|
[1846] | 1323 | this function::
|
---|
[931] | 1324 |
|
---|
[1846] | 1325 | # provided your scantable is called scan
|
---|
| 1326 | selection = selector()
|
---|
| 1327 | selection.set_name("ORION*")
|
---|
| 1328 | selection.set_ifs([1])
|
---|
| 1329 | scan.set_selection(selection)
|
---|
| 1330 | scan.set_restfreqs(freqs=86.6e9)
|
---|
| 1331 |
|
---|
[931] | 1332 | """
|
---|
| 1333 | varlist = vars()
|
---|
[1157] | 1334 | from asap import linecatalog
|
---|
| 1335 | # simple value
|
---|
[1118] | 1336 | if isinstance(freqs, int) or isinstance(freqs, float):
|
---|
[1845] | 1337 | self._setrestfreqs([freqs], [""], unit)
|
---|
[1157] | 1338 | # list of values
|
---|
[1118] | 1339 | elif isinstance(freqs, list) or isinstance(freqs, tuple):
|
---|
[1157] | 1340 | # list values are scalars
|
---|
[1118] | 1341 | if isinstance(freqs[-1], int) or isinstance(freqs[-1], float):
|
---|
[1845] | 1342 | if len(freqs) == 1:
|
---|
| 1343 | self._setrestfreqs(freqs, [""], unit)
|
---|
| 1344 | else:
|
---|
| 1345 | # allow the 'old' mode of setting mulitple IFs
|
---|
| 1346 | sel = selector()
|
---|
| 1347 | savesel = self._getselection()
|
---|
| 1348 | iflist = self.getifnos()
|
---|
| 1349 | if len(freqs)>len(iflist):
|
---|
| 1350 | raise ValueError("number of elements in list of list "
|
---|
| 1351 | "exeeds the current IF selections")
|
---|
| 1352 | iflist = self.getifnos()
|
---|
| 1353 | for i, fval in enumerate(freqs):
|
---|
| 1354 | sel.set_ifs(iflist[i])
|
---|
| 1355 | self._setselection(sel)
|
---|
| 1356 | self._setrestfreqs([fval], [""], unit)
|
---|
| 1357 | self._setselection(savesel)
|
---|
| 1358 |
|
---|
| 1359 | # list values are dict, {'value'=, 'name'=)
|
---|
[1157] | 1360 | elif isinstance(freqs[-1], dict):
|
---|
[1845] | 1361 | values = []
|
---|
| 1362 | names = []
|
---|
| 1363 | for d in freqs:
|
---|
| 1364 | values.append(d["value"])
|
---|
| 1365 | names.append(d["name"])
|
---|
| 1366 | self._setrestfreqs(values, names, unit)
|
---|
[1819] | 1367 | elif isinstance(freqs[-1], list) or isinstance(freqs[-1], tuple):
|
---|
[1157] | 1368 | sel = selector()
|
---|
| 1369 | savesel = self._getselection()
|
---|
[1322] | 1370 | iflist = self.getifnos()
|
---|
[1819] | 1371 | if len(freqs)>len(iflist):
|
---|
[1845] | 1372 | raise ValueError("number of elements in list of list exeeds"
|
---|
| 1373 | " the current IF selections")
|
---|
| 1374 | for i, fval in enumerate(freqs):
|
---|
[1322] | 1375 | sel.set_ifs(iflist[i])
|
---|
[1259] | 1376 | self._setselection(sel)
|
---|
[1845] | 1377 | self._setrestfreqs(fval, [""], unit)
|
---|
[1157] | 1378 | self._setselection(savesel)
|
---|
| 1379 | # freqs are to be taken from a linecatalog
|
---|
[1153] | 1380 | elif isinstance(freqs, linecatalog):
|
---|
| 1381 | sel = selector()
|
---|
| 1382 | savesel = self._getselection()
|
---|
| 1383 | for i in xrange(freqs.nrow()):
|
---|
[1322] | 1384 | sel.set_ifs(iflist[i])
|
---|
[1153] | 1385 | self._setselection(sel)
|
---|
[1845] | 1386 | self._setrestfreqs([freqs.get_frequency(i)],
|
---|
| 1387 | [freqs.get_name(i)], "MHz")
|
---|
[1153] | 1388 | # ensure that we are not iterating past nIF
|
---|
| 1389 | if i == self.nif()-1: break
|
---|
| 1390 | self._setselection(savesel)
|
---|
[931] | 1391 | else:
|
---|
| 1392 | return
|
---|
| 1393 | self._add_history("set_restfreqs", varlist)
|
---|
| 1394 |
|
---|
[1360] | 1395 | def shift_refpix(self, delta):
|
---|
[1846] | 1396 | """\
|
---|
[1589] | 1397 | Shift the reference pixel of the Spectra Coordinate by an
|
---|
| 1398 | integer amount.
|
---|
[1846] | 1399 |
|
---|
[1589] | 1400 | Parameters:
|
---|
[1846] | 1401 |
|
---|
[1589] | 1402 | delta: the amount to shift by
|
---|
[1846] | 1403 |
|
---|
| 1404 | *Note*:
|
---|
| 1405 |
|
---|
[1589] | 1406 | Be careful using this with broadband data.
|
---|
[1846] | 1407 |
|
---|
[1360] | 1408 | """
|
---|
[1731] | 1409 | Scantable.shift_refpix(self, delta)
|
---|
[931] | 1410 |
|
---|
[1862] | 1411 | @asaplog_post_dec
|
---|
[1259] | 1412 | def history(self, filename=None):
|
---|
[1846] | 1413 | """\
|
---|
[1259] | 1414 | Print the history. Optionally to a file.
|
---|
[1846] | 1415 |
|
---|
[1348] | 1416 | Parameters:
|
---|
[1846] | 1417 |
|
---|
[1348] | 1418 | filename: The name of the file to save the history to.
|
---|
[1846] | 1419 |
|
---|
[1259] | 1420 | """
|
---|
[484] | 1421 | hist = list(self._gethistory())
|
---|
[794] | 1422 | out = "-"*80
|
---|
[484] | 1423 | for h in hist:
|
---|
[489] | 1424 | if h.startswith("---"):
|
---|
[1857] | 1425 | out = "\n".join([out, h])
|
---|
[489] | 1426 | else:
|
---|
| 1427 | items = h.split("##")
|
---|
| 1428 | date = items[0]
|
---|
| 1429 | func = items[1]
|
---|
| 1430 | items = items[2:]
|
---|
[794] | 1431 | out += "\n"+date+"\n"
|
---|
| 1432 | out += "Function: %s\n Parameters:" % (func)
|
---|
[489] | 1433 | for i in items:
|
---|
| 1434 | s = i.split("=")
|
---|
[1118] | 1435 | out += "\n %s = %s" % (s[0], s[1])
|
---|
[1857] | 1436 | out = "\n".join([out, "-"*80])
|
---|
[1259] | 1437 | if filename is not None:
|
---|
| 1438 | if filename is "":
|
---|
| 1439 | filename = 'scantable_history.txt'
|
---|
| 1440 | import os
|
---|
| 1441 | filename = os.path.expandvars(os.path.expanduser(filename))
|
---|
| 1442 | if not os.path.isdir(filename):
|
---|
| 1443 | data = open(filename, 'w')
|
---|
| 1444 | data.write(out)
|
---|
| 1445 | data.close()
|
---|
| 1446 | else:
|
---|
| 1447 | msg = "Illegal file name '%s'." % (filename)
|
---|
[1859] | 1448 | raise IOError(msg)
|
---|
| 1449 | return page(out)
|
---|
[513] | 1450 | #
|
---|
| 1451 | # Maths business
|
---|
| 1452 | #
|
---|
[1862] | 1453 | @asaplog_post_dec
|
---|
[931] | 1454 | def average_time(self, mask=None, scanav=False, weight='tint', align=False):
|
---|
[1846] | 1455 | """\
|
---|
[1070] | 1456 | Return the (time) weighted average of a scan.
|
---|
[1846] | 1457 |
|
---|
| 1458 | *Note*:
|
---|
| 1459 |
|
---|
[1070] | 1460 | in channels only - align if necessary
|
---|
[1846] | 1461 |
|
---|
[513] | 1462 | Parameters:
|
---|
[1846] | 1463 |
|
---|
[513] | 1464 | mask: an optional mask (only used for 'var' and 'tsys'
|
---|
| 1465 | weighting)
|
---|
[1855] | 1466 |
|
---|
[558] | 1467 | scanav: True averages each scan separately
|
---|
| 1468 | False (default) averages all scans together,
|
---|
[1855] | 1469 |
|
---|
[1099] | 1470 | weight: Weighting scheme.
|
---|
| 1471 | 'none' (mean no weight)
|
---|
| 1472 | 'var' (1/var(spec) weighted)
|
---|
| 1473 | 'tsys' (1/Tsys**2 weighted)
|
---|
| 1474 | 'tint' (integration time weighted)
|
---|
| 1475 | 'tintsys' (Tint/Tsys**2)
|
---|
| 1476 | 'median' ( median averaging)
|
---|
[535] | 1477 | The default is 'tint'
|
---|
[1855] | 1478 |
|
---|
[931] | 1479 | align: align the spectra in velocity before averaging. It takes
|
---|
| 1480 | the time of the first spectrum as reference time.
|
---|
[1846] | 1481 |
|
---|
| 1482 | Example::
|
---|
| 1483 |
|
---|
[513] | 1484 | # time average the scantable without using a mask
|
---|
[710] | 1485 | newscan = scan.average_time()
|
---|
[1846] | 1486 |
|
---|
[513] | 1487 | """
|
---|
| 1488 | varlist = vars()
|
---|
[1593] | 1489 | weight = weight or 'TINT'
|
---|
| 1490 | mask = mask or ()
|
---|
| 1491 | scanav = (scanav and 'SCAN') or 'NONE'
|
---|
[1118] | 1492 | scan = (self, )
|
---|
[1859] | 1493 |
|
---|
| 1494 | if align:
|
---|
| 1495 | scan = (self.freq_align(insitu=False), )
|
---|
| 1496 | s = None
|
---|
| 1497 | if weight.upper() == 'MEDIAN':
|
---|
| 1498 | s = scantable(self._math._averagechannel(scan[0], 'MEDIAN',
|
---|
| 1499 | scanav))
|
---|
| 1500 | else:
|
---|
| 1501 | s = scantable(self._math._average(scan, mask, weight.upper(),
|
---|
| 1502 | scanav))
|
---|
[1099] | 1503 | s._add_history("average_time", varlist)
|
---|
[513] | 1504 | return s
|
---|
[710] | 1505 |
|
---|
[1862] | 1506 | @asaplog_post_dec
|
---|
[876] | 1507 | def convert_flux(self, jyperk=None, eta=None, d=None, insitu=None):
|
---|
[1846] | 1508 | """\
|
---|
[513] | 1509 | Return a scan where all spectra are converted to either
|
---|
| 1510 | Jansky or Kelvin depending upon the flux units of the scan table.
|
---|
| 1511 | By default the function tries to look the values up internally.
|
---|
| 1512 | If it can't find them (or if you want to over-ride), you must
|
---|
| 1513 | specify EITHER jyperk OR eta (and D which it will try to look up
|
---|
| 1514 | also if you don't set it). jyperk takes precedence if you set both.
|
---|
[1846] | 1515 |
|
---|
[513] | 1516 | Parameters:
|
---|
[1846] | 1517 |
|
---|
[513] | 1518 | jyperk: the Jy / K conversion factor
|
---|
[1855] | 1519 |
|
---|
[513] | 1520 | eta: the aperture efficiency
|
---|
[1855] | 1521 |
|
---|
[513] | 1522 | d: the geomtric diameter (metres)
|
---|
[1855] | 1523 |
|
---|
[513] | 1524 | insitu: if False a new scantable is returned.
|
---|
| 1525 | Otherwise, the scaling is done in-situ
|
---|
| 1526 | The default is taken from .asaprc (False)
|
---|
[1846] | 1527 |
|
---|
[513] | 1528 | """
|
---|
| 1529 | if insitu is None: insitu = rcParams['insitu']
|
---|
[876] | 1530 | self._math._setinsitu(insitu)
|
---|
[513] | 1531 | varlist = vars()
|
---|
[1593] | 1532 | jyperk = jyperk or -1.0
|
---|
| 1533 | d = d or -1.0
|
---|
| 1534 | eta = eta or -1.0
|
---|
[876] | 1535 | s = scantable(self._math._convertflux(self, d, eta, jyperk))
|
---|
| 1536 | s._add_history("convert_flux", varlist)
|
---|
| 1537 | if insitu: self._assign(s)
|
---|
| 1538 | else: return s
|
---|
[513] | 1539 |
|
---|
[1862] | 1540 | @asaplog_post_dec
|
---|
[876] | 1541 | def gain_el(self, poly=None, filename="", method="linear", insitu=None):
|
---|
[1846] | 1542 | """\
|
---|
[513] | 1543 | Return a scan after applying a gain-elevation correction.
|
---|
| 1544 | The correction can be made via either a polynomial or a
|
---|
| 1545 | table-based interpolation (and extrapolation if necessary).
|
---|
| 1546 | You specify polynomial coefficients, an ascii table or neither.
|
---|
| 1547 | If you specify neither, then a polynomial correction will be made
|
---|
| 1548 | with built in coefficients known for certain telescopes (an error
|
---|
| 1549 | will occur if the instrument is not known).
|
---|
| 1550 | The data and Tsys are *divided* by the scaling factors.
|
---|
[1846] | 1551 |
|
---|
[513] | 1552 | Parameters:
|
---|
[1846] | 1553 |
|
---|
[513] | 1554 | poly: Polynomial coefficients (default None) to compute a
|
---|
| 1555 | gain-elevation correction as a function of
|
---|
| 1556 | elevation (in degrees).
|
---|
[1855] | 1557 |
|
---|
[513] | 1558 | filename: The name of an ascii file holding correction factors.
|
---|
| 1559 | The first row of the ascii file must give the column
|
---|
| 1560 | names and these MUST include columns
|
---|
| 1561 | "ELEVATION" (degrees) and "FACTOR" (multiply data
|
---|
| 1562 | by this) somewhere.
|
---|
| 1563 | The second row must give the data type of the
|
---|
| 1564 | column. Use 'R' for Real and 'I' for Integer.
|
---|
| 1565 | An example file would be
|
---|
| 1566 | (actual factors are arbitrary) :
|
---|
| 1567 |
|
---|
| 1568 | TIME ELEVATION FACTOR
|
---|
| 1569 | R R R
|
---|
| 1570 | 0.1 0 0.8
|
---|
| 1571 | 0.2 20 0.85
|
---|
| 1572 | 0.3 40 0.9
|
---|
| 1573 | 0.4 60 0.85
|
---|
| 1574 | 0.5 80 0.8
|
---|
| 1575 | 0.6 90 0.75
|
---|
[1855] | 1576 |
|
---|
[513] | 1577 | method: Interpolation method when correcting from a table.
|
---|
| 1578 | Values are "nearest", "linear" (default), "cubic"
|
---|
| 1579 | and "spline"
|
---|
[1855] | 1580 |
|
---|
[513] | 1581 | insitu: if False a new scantable is returned.
|
---|
| 1582 | Otherwise, the scaling is done in-situ
|
---|
| 1583 | The default is taken from .asaprc (False)
|
---|
[1846] | 1584 |
|
---|
[513] | 1585 | """
|
---|
| 1586 |
|
---|
| 1587 | if insitu is None: insitu = rcParams['insitu']
|
---|
[876] | 1588 | self._math._setinsitu(insitu)
|
---|
[513] | 1589 | varlist = vars()
|
---|
[1593] | 1590 | poly = poly or ()
|
---|
[513] | 1591 | from os.path import expandvars
|
---|
| 1592 | filename = expandvars(filename)
|
---|
[876] | 1593 | s = scantable(self._math._gainel(self, poly, filename, method))
|
---|
| 1594 | s._add_history("gain_el", varlist)
|
---|
[1593] | 1595 | if insitu:
|
---|
| 1596 | self._assign(s)
|
---|
| 1597 | else:
|
---|
| 1598 | return s
|
---|
[710] | 1599 |
|
---|
[1862] | 1600 | @asaplog_post_dec
|
---|
[931] | 1601 | def freq_align(self, reftime=None, method='cubic', insitu=None):
|
---|
[1846] | 1602 | """\
|
---|
[513] | 1603 | Return a scan where all rows have been aligned in frequency/velocity.
|
---|
| 1604 | The alignment frequency frame (e.g. LSRK) is that set by function
|
---|
| 1605 | set_freqframe.
|
---|
[1846] | 1606 |
|
---|
[513] | 1607 | Parameters:
|
---|
[1855] | 1608 |
|
---|
[513] | 1609 | reftime: reference time to align at. By default, the time of
|
---|
| 1610 | the first row of data is used.
|
---|
[1855] | 1611 |
|
---|
[513] | 1612 | method: Interpolation method for regridding the spectra.
|
---|
| 1613 | Choose from "nearest", "linear", "cubic" (default)
|
---|
| 1614 | and "spline"
|
---|
[1855] | 1615 |
|
---|
[513] | 1616 | insitu: if False a new scantable is returned.
|
---|
| 1617 | Otherwise, the scaling is done in-situ
|
---|
| 1618 | The default is taken from .asaprc (False)
|
---|
[1846] | 1619 |
|
---|
[513] | 1620 | """
|
---|
[931] | 1621 | if insitu is None: insitu = rcParams["insitu"]
|
---|
[876] | 1622 | self._math._setinsitu(insitu)
|
---|
[513] | 1623 | varlist = vars()
|
---|
[1593] | 1624 | reftime = reftime or ""
|
---|
[931] | 1625 | s = scantable(self._math._freq_align(self, reftime, method))
|
---|
[876] | 1626 | s._add_history("freq_align", varlist)
|
---|
| 1627 | if insitu: self._assign(s)
|
---|
| 1628 | else: return s
|
---|
[513] | 1629 |
|
---|
[1862] | 1630 | @asaplog_post_dec
|
---|
[1725] | 1631 | def opacity(self, tau=None, insitu=None):
|
---|
[1846] | 1632 | """\
|
---|
[513] | 1633 | Apply an opacity correction. The data
|
---|
| 1634 | and Tsys are multiplied by the correction factor.
|
---|
[1846] | 1635 |
|
---|
[513] | 1636 | Parameters:
|
---|
[1855] | 1637 |
|
---|
[1689] | 1638 | tau: (list of) opacity from which the correction factor is
|
---|
[513] | 1639 | exp(tau*ZD)
|
---|
[1689] | 1640 | where ZD is the zenith-distance.
|
---|
| 1641 | If a list is provided, it has to be of length nIF,
|
---|
| 1642 | nIF*nPol or 1 and in order of IF/POL, e.g.
|
---|
| 1643 | [opif0pol0, opif0pol1, opif1pol0 ...]
|
---|
[1725] | 1644 | if tau is `None` the opacities are determined from a
|
---|
| 1645 | model.
|
---|
[1855] | 1646 |
|
---|
[513] | 1647 | insitu: if False a new scantable is returned.
|
---|
| 1648 | Otherwise, the scaling is done in-situ
|
---|
| 1649 | The default is taken from .asaprc (False)
|
---|
[1846] | 1650 |
|
---|
[513] | 1651 | """
|
---|
| 1652 | if insitu is None: insitu = rcParams['insitu']
|
---|
[876] | 1653 | self._math._setinsitu(insitu)
|
---|
[513] | 1654 | varlist = vars()
|
---|
[1689] | 1655 | if not hasattr(tau, "__len__"):
|
---|
| 1656 | tau = [tau]
|
---|
[876] | 1657 | s = scantable(self._math._opacity(self, tau))
|
---|
| 1658 | s._add_history("opacity", varlist)
|
---|
| 1659 | if insitu: self._assign(s)
|
---|
| 1660 | else: return s
|
---|
[513] | 1661 |
|
---|
[1862] | 1662 | @asaplog_post_dec
|
---|
[513] | 1663 | def bin(self, width=5, insitu=None):
|
---|
[1846] | 1664 | """\
|
---|
[513] | 1665 | Return a scan where all spectra have been binned up.
|
---|
[1846] | 1666 |
|
---|
[1348] | 1667 | Parameters:
|
---|
[1846] | 1668 |
|
---|
[513] | 1669 | width: The bin width (default=5) in pixels
|
---|
[1855] | 1670 |
|
---|
[513] | 1671 | insitu: if False a new scantable is returned.
|
---|
| 1672 | Otherwise, the scaling is done in-situ
|
---|
| 1673 | The default is taken from .asaprc (False)
|
---|
[1846] | 1674 |
|
---|
[513] | 1675 | """
|
---|
| 1676 | if insitu is None: insitu = rcParams['insitu']
|
---|
[876] | 1677 | self._math._setinsitu(insitu)
|
---|
[513] | 1678 | varlist = vars()
|
---|
[876] | 1679 | s = scantable(self._math._bin(self, width))
|
---|
[1118] | 1680 | s._add_history("bin", varlist)
|
---|
[1589] | 1681 | if insitu:
|
---|
| 1682 | self._assign(s)
|
---|
| 1683 | else:
|
---|
| 1684 | return s
|
---|
[513] | 1685 |
|
---|
[1862] | 1686 | @asaplog_post_dec
|
---|
[513] | 1687 | def resample(self, width=5, method='cubic', insitu=None):
|
---|
[1846] | 1688 | """\
|
---|
[1348] | 1689 | Return a scan where all spectra have been binned up.
|
---|
[1573] | 1690 |
|
---|
[1348] | 1691 | Parameters:
|
---|
[1846] | 1692 |
|
---|
[513] | 1693 | width: The bin width (default=5) in pixels
|
---|
[1855] | 1694 |
|
---|
[513] | 1695 | method: Interpolation method when correcting from a table.
|
---|
| 1696 | Values are "nearest", "linear", "cubic" (default)
|
---|
| 1697 | and "spline"
|
---|
[1855] | 1698 |
|
---|
[513] | 1699 | insitu: if False a new scantable is returned.
|
---|
| 1700 | Otherwise, the scaling is done in-situ
|
---|
| 1701 | The default is taken from .asaprc (False)
|
---|
[1846] | 1702 |
|
---|
[513] | 1703 | """
|
---|
| 1704 | if insitu is None: insitu = rcParams['insitu']
|
---|
[876] | 1705 | self._math._setinsitu(insitu)
|
---|
[513] | 1706 | varlist = vars()
|
---|
[876] | 1707 | s = scantable(self._math._resample(self, method, width))
|
---|
[1118] | 1708 | s._add_history("resample", varlist)
|
---|
[876] | 1709 | if insitu: self._assign(s)
|
---|
| 1710 | else: return s
|
---|
[513] | 1711 |
|
---|
[1862] | 1712 | @asaplog_post_dec
|
---|
[946] | 1713 | def average_pol(self, mask=None, weight='none'):
|
---|
[1846] | 1714 | """\
|
---|
[946] | 1715 | Average the Polarisations together.
|
---|
[1846] | 1716 |
|
---|
[946] | 1717 | Parameters:
|
---|
[1846] | 1718 |
|
---|
[946] | 1719 | mask: An optional mask defining the region, where the
|
---|
| 1720 | averaging will be applied. The output will have all
|
---|
| 1721 | specified points masked.
|
---|
[1855] | 1722 |
|
---|
[946] | 1723 | weight: Weighting scheme. 'none' (default), 'var' (1/var(spec)
|
---|
| 1724 | weighted), or 'tsys' (1/Tsys**2 weighted)
|
---|
[1846] | 1725 |
|
---|
[946] | 1726 | """
|
---|
| 1727 | varlist = vars()
|
---|
[1593] | 1728 | mask = mask or ()
|
---|
[1010] | 1729 | s = scantable(self._math._averagepol(self, mask, weight.upper()))
|
---|
[1118] | 1730 | s._add_history("average_pol", varlist)
|
---|
[992] | 1731 | return s
|
---|
[513] | 1732 |
|
---|
[1862] | 1733 | @asaplog_post_dec
|
---|
[1145] | 1734 | def average_beam(self, mask=None, weight='none'):
|
---|
[1846] | 1735 | """\
|
---|
[1145] | 1736 | Average the Beams together.
|
---|
[1846] | 1737 |
|
---|
[1145] | 1738 | Parameters:
|
---|
| 1739 | mask: An optional mask defining the region, where the
|
---|
| 1740 | averaging will be applied. The output will have all
|
---|
| 1741 | specified points masked.
|
---|
[1855] | 1742 |
|
---|
[1145] | 1743 | weight: Weighting scheme. 'none' (default), 'var' (1/var(spec)
|
---|
| 1744 | weighted), or 'tsys' (1/Tsys**2 weighted)
|
---|
[1846] | 1745 |
|
---|
[1145] | 1746 | """
|
---|
| 1747 | varlist = vars()
|
---|
[1593] | 1748 | mask = mask or ()
|
---|
[1145] | 1749 | s = scantable(self._math._averagebeams(self, mask, weight.upper()))
|
---|
| 1750 | s._add_history("average_beam", varlist)
|
---|
| 1751 | return s
|
---|
| 1752 |
|
---|
[1586] | 1753 | def parallactify(self, pflag):
|
---|
[1846] | 1754 | """\
|
---|
[1843] | 1755 | Set a flag to indicate whether this data should be treated as having
|
---|
[1617] | 1756 | been 'parallactified' (total phase == 0.0)
|
---|
[1846] | 1757 |
|
---|
[1617] | 1758 | Parameters:
|
---|
[1855] | 1759 |
|
---|
[1843] | 1760 | pflag: Bool indicating whether to turn this on (True) or
|
---|
[1617] | 1761 | off (False)
|
---|
[1846] | 1762 |
|
---|
[1617] | 1763 | """
|
---|
[1586] | 1764 | varlist = vars()
|
---|
| 1765 | self._parallactify(pflag)
|
---|
| 1766 | self._add_history("parallactify", varlist)
|
---|
| 1767 |
|
---|
[1862] | 1768 | @asaplog_post_dec
|
---|
[992] | 1769 | def convert_pol(self, poltype=None):
|
---|
[1846] | 1770 | """\
|
---|
[992] | 1771 | Convert the data to a different polarisation type.
|
---|
[1565] | 1772 | Note that you will need cross-polarisation terms for most conversions.
|
---|
[1846] | 1773 |
|
---|
[992] | 1774 | Parameters:
|
---|
[1855] | 1775 |
|
---|
[992] | 1776 | poltype: The new polarisation type. Valid types are:
|
---|
[1565] | 1777 | "linear", "circular", "stokes" and "linpol"
|
---|
[1846] | 1778 |
|
---|
[992] | 1779 | """
|
---|
| 1780 | varlist = vars()
|
---|
[1859] | 1781 | s = scantable(self._math._convertpol(self, poltype))
|
---|
[1118] | 1782 | s._add_history("convert_pol", varlist)
|
---|
[992] | 1783 | return s
|
---|
| 1784 |
|
---|
[1862] | 1785 | @asaplog_post_dec
|
---|
[1819] | 1786 | def smooth(self, kernel="hanning", width=5.0, order=2, plot=False, insitu=None):
|
---|
[1846] | 1787 | """\
|
---|
[513] | 1788 | Smooth the spectrum by the specified kernel (conserving flux).
|
---|
[1846] | 1789 |
|
---|
[513] | 1790 | Parameters:
|
---|
[1846] | 1791 |
|
---|
[513] | 1792 | kernel: The type of smoothing kernel. Select from
|
---|
[1574] | 1793 | 'hanning' (default), 'gaussian', 'boxcar', 'rmedian'
|
---|
| 1794 | or 'poly'
|
---|
[1855] | 1795 |
|
---|
[513] | 1796 | width: The width of the kernel in pixels. For hanning this is
|
---|
| 1797 | ignored otherwise it defauls to 5 pixels.
|
---|
| 1798 | For 'gaussian' it is the Full Width Half
|
---|
| 1799 | Maximum. For 'boxcar' it is the full width.
|
---|
[1574] | 1800 | For 'rmedian' and 'poly' it is the half width.
|
---|
[1855] | 1801 |
|
---|
[1574] | 1802 | order: Optional parameter for 'poly' kernel (default is 2), to
|
---|
| 1803 | specify the order of the polnomial. Ignored by all other
|
---|
| 1804 | kernels.
|
---|
[1855] | 1805 |
|
---|
[1819] | 1806 | plot: plot the original and the smoothed spectra.
|
---|
| 1807 | In this each indivual fit has to be approved, by
|
---|
| 1808 | typing 'y' or 'n'
|
---|
[1855] | 1809 |
|
---|
[513] | 1810 | insitu: if False a new scantable is returned.
|
---|
| 1811 | Otherwise, the scaling is done in-situ
|
---|
| 1812 | The default is taken from .asaprc (False)
|
---|
[1846] | 1813 |
|
---|
[513] | 1814 | """
|
---|
| 1815 | if insitu is None: insitu = rcParams['insitu']
|
---|
[876] | 1816 | self._math._setinsitu(insitu)
|
---|
[513] | 1817 | varlist = vars()
|
---|
[1819] | 1818 |
|
---|
| 1819 | if plot: orgscan = self.copy()
|
---|
| 1820 |
|
---|
[1574] | 1821 | s = scantable(self._math._smooth(self, kernel.lower(), width, order))
|
---|
[876] | 1822 | s._add_history("smooth", varlist)
|
---|
[1819] | 1823 |
|
---|
| 1824 | if plot:
|
---|
| 1825 | if rcParams['plotter.gui']:
|
---|
| 1826 | from asap.asaplotgui import asaplotgui as asaplot
|
---|
| 1827 | else:
|
---|
| 1828 | from asap.asaplot import asaplot
|
---|
| 1829 | self._p=asaplot()
|
---|
| 1830 | self._p.set_panels()
|
---|
| 1831 | ylab=s._get_ordinate_label()
|
---|
| 1832 | #self._p.palette(0,["#777777","red"])
|
---|
| 1833 | for r in xrange(s.nrow()):
|
---|
| 1834 | xsm=s._getabcissa(r)
|
---|
| 1835 | ysm=s._getspectrum(r)
|
---|
| 1836 | xorg=orgscan._getabcissa(r)
|
---|
| 1837 | yorg=orgscan._getspectrum(r)
|
---|
| 1838 | self._p.clear()
|
---|
| 1839 | self._p.hold()
|
---|
| 1840 | self._p.set_axes('ylabel',ylab)
|
---|
| 1841 | self._p.set_axes('xlabel',s._getabcissalabel(r))
|
---|
| 1842 | self._p.set_axes('title',s._getsourcename(r))
|
---|
| 1843 | self._p.set_line(label='Original',color="#777777")
|
---|
| 1844 | self._p.plot(xorg,yorg)
|
---|
| 1845 | self._p.set_line(label='Smoothed',color="red")
|
---|
| 1846 | self._p.plot(xsm,ysm)
|
---|
| 1847 | ### Ugly part for legend
|
---|
| 1848 | for i in [0,1]:
|
---|
| 1849 | self._p.subplots[0]['lines'].append([self._p.subplots[0]['axes'].lines[i]])
|
---|
| 1850 | self._p.release()
|
---|
| 1851 | ### Ugly part for legend
|
---|
| 1852 | self._p.subplots[0]['lines']=[]
|
---|
| 1853 | res = raw_input("Accept smoothing ([y]/n): ")
|
---|
| 1854 | if res.upper() == 'N':
|
---|
| 1855 | s._setspectrum(yorg, r)
|
---|
| 1856 | self._p.unmap()
|
---|
| 1857 | self._p = None
|
---|
| 1858 | del orgscan
|
---|
| 1859 |
|
---|
[876] | 1860 | if insitu: self._assign(s)
|
---|
| 1861 | else: return s
|
---|
[513] | 1862 |
|
---|
[1862] | 1863 | @asaplog_post_dec
|
---|
[1907] | 1864 | def old_poly_baseline(self, mask=None, order=0, plot=False, uselin=False, insitu=None, rows=None):
|
---|
[1846] | 1865 | """\
|
---|
[513] | 1866 | Return a scan which has been baselined (all rows) by a polynomial.
|
---|
[1907] | 1867 |
|
---|
[513] | 1868 | Parameters:
|
---|
[1846] | 1869 |
|
---|
[794] | 1870 | mask: an optional mask
|
---|
[1855] | 1871 |
|
---|
[794] | 1872 | order: the order of the polynomial (default is 0)
|
---|
[1855] | 1873 |
|
---|
[1061] | 1874 | plot: plot the fit and the residual. In this each
|
---|
| 1875 | indivual fit has to be approved, by typing 'y'
|
---|
| 1876 | or 'n'
|
---|
[1855] | 1877 |
|
---|
[1391] | 1878 | uselin: use linear polynomial fit
|
---|
[1855] | 1879 |
|
---|
[794] | 1880 | insitu: if False a new scantable is returned.
|
---|
| 1881 | Otherwise, the scaling is done in-situ
|
---|
| 1882 | The default is taken from .asaprc (False)
|
---|
[1846] | 1883 |
|
---|
[1907] | 1884 | rows: row numbers of spectra to be processed.
|
---|
| 1885 | (default is None: for all rows)
|
---|
| 1886 |
|
---|
| 1887 | Example:
|
---|
[513] | 1888 | # return a scan baselined by a third order polynomial,
|
---|
| 1889 | # not using a mask
|
---|
| 1890 | bscan = scan.poly_baseline(order=3)
|
---|
[1846] | 1891 |
|
---|
[579] | 1892 | """
|
---|
[513] | 1893 | if insitu is None: insitu = rcParams['insitu']
|
---|
[1819] | 1894 | if not insitu:
|
---|
| 1895 | workscan = self.copy()
|
---|
| 1896 | else:
|
---|
| 1897 | workscan = self
|
---|
[513] | 1898 | varlist = vars()
|
---|
| 1899 | if mask is None:
|
---|
[1907] | 1900 | mask = [True for i in xrange(self.nchan())]
|
---|
[1819] | 1901 |
|
---|
[1217] | 1902 | try:
|
---|
| 1903 | f = fitter()
|
---|
[1391] | 1904 | if uselin:
|
---|
| 1905 | f.set_function(lpoly=order)
|
---|
| 1906 | else:
|
---|
| 1907 | f.set_function(poly=order)
|
---|
[1819] | 1908 |
|
---|
[1907] | 1909 | if rows == None:
|
---|
| 1910 | rows = xrange(workscan.nrow())
|
---|
| 1911 | elif isinstance(rows, int):
|
---|
| 1912 | rows = [ rows ]
|
---|
| 1913 |
|
---|
[1819] | 1914 | if len(rows) > 0:
|
---|
| 1915 | self.blpars = []
|
---|
[1907] | 1916 | self.masklists = []
|
---|
| 1917 | self.actualmask = []
|
---|
| 1918 |
|
---|
[1819] | 1919 | for r in rows:
|
---|
| 1920 | f.x = workscan._getabcissa(r)
|
---|
| 1921 | f.y = workscan._getspectrum(r)
|
---|
[1907] | 1922 | f.mask = mask_and(mask, workscan._getmask(r)) # (CAS-1434)
|
---|
[1819] | 1923 | f.data = None
|
---|
| 1924 | f.fit()
|
---|
| 1925 | if plot:
|
---|
| 1926 | f.plot(residual=True)
|
---|
| 1927 | x = raw_input("Accept fit ( [y]/n ): ")
|
---|
| 1928 | if x.upper() == 'N':
|
---|
| 1929 | self.blpars.append(None)
|
---|
[1907] | 1930 | self.masklists.append(None)
|
---|
| 1931 | self.actualmask.append(None)
|
---|
[1819] | 1932 | continue
|
---|
| 1933 | workscan._setspectrum(f.fitter.getresidual(), r)
|
---|
| 1934 | self.blpars.append(f.get_parameters())
|
---|
[1907] | 1935 | self.masklists.append(workscan.get_masklist(f.mask, row=r))
|
---|
| 1936 | self.actualmask.append(f.mask)
|
---|
[1819] | 1937 |
|
---|
| 1938 | if plot:
|
---|
| 1939 | f._p.unmap()
|
---|
| 1940 | f._p = None
|
---|
| 1941 | workscan._add_history("poly_baseline", varlist)
|
---|
[1856] | 1942 | if insitu:
|
---|
| 1943 | self._assign(workscan)
|
---|
| 1944 | else:
|
---|
| 1945 | return workscan
|
---|
[1217] | 1946 | except RuntimeError:
|
---|
| 1947 | msg = "The fit failed, possibly because it didn't converge."
|
---|
[1859] | 1948 | raise RuntimeError(msg)
|
---|
[513] | 1949 |
|
---|
[1819] | 1950 |
|
---|
[1907] | 1951 | def poly_baseline(self, mask=None, order=0, plot=False, batch=False, insitu=None, rows=None):
|
---|
| 1952 | """\
|
---|
| 1953 | Return a scan which has been baselined (all rows) by a polynomial.
|
---|
| 1954 | Parameters:
|
---|
| 1955 | mask: an optional mask
|
---|
| 1956 | order: the order of the polynomial (default is 0)
|
---|
| 1957 | plot: plot the fit and the residual. In this each
|
---|
| 1958 | indivual fit has to be approved, by typing 'y'
|
---|
| 1959 | or 'n'. Ignored if batch = True.
|
---|
| 1960 | batch: if True a faster algorithm is used and logs
|
---|
| 1961 | including the fit results are not output
|
---|
| 1962 | (default is False)
|
---|
| 1963 | insitu: if False a new scantable is returned.
|
---|
| 1964 | Otherwise, the scaling is done in-situ
|
---|
| 1965 | The default is taken from .asaprc (False)
|
---|
| 1966 | rows: row numbers of spectra to be processed.
|
---|
| 1967 | (default is None: for all rows)
|
---|
| 1968 | Example:
|
---|
| 1969 | # return a scan baselined by a third order polynomial,
|
---|
| 1970 | # not using a mask
|
---|
| 1971 | bscan = scan.poly_baseline(order=3)
|
---|
| 1972 | """
|
---|
| 1973 | if insitu is None: insitu = rcParams["insitu"]
|
---|
| 1974 | if insitu:
|
---|
| 1975 | workscan = self
|
---|
| 1976 | else:
|
---|
| 1977 | workscan = self.copy()
|
---|
| 1978 |
|
---|
| 1979 | varlist = vars()
|
---|
| 1980 | nchan = workscan.nchan()
|
---|
| 1981 |
|
---|
| 1982 | if mask is None:
|
---|
| 1983 | mask = [True for i in xrange(nchan)]
|
---|
| 1984 |
|
---|
| 1985 | try:
|
---|
| 1986 | if rows == None:
|
---|
| 1987 | rows = xrange(workscan.nrow())
|
---|
| 1988 | elif isinstance(rows, int):
|
---|
| 1989 | rows = [ rows ]
|
---|
| 1990 |
|
---|
| 1991 | if len(rows) > 0:
|
---|
| 1992 | self.blpars = []
|
---|
| 1993 | self.masklists = []
|
---|
| 1994 | self.actualmask = []
|
---|
| 1995 |
|
---|
| 1996 | if batch:
|
---|
| 1997 | for r in rows:
|
---|
[1908] | 1998 | workscan._poly_baseline_batch(mask, order, r)
|
---|
[1907] | 1999 | elif plot:
|
---|
| 2000 | f = fitter()
|
---|
| 2001 | f.set_function(lpoly=order)
|
---|
| 2002 | for r in rows:
|
---|
| 2003 | f.x = workscan._getabcissa(r)
|
---|
| 2004 | f.y = workscan._getspectrum(r)
|
---|
| 2005 | f.mask = mask_and(mask, workscan._getmask(r)) # (CAS-1434)
|
---|
| 2006 | f.data = None
|
---|
| 2007 | f.fit()
|
---|
| 2008 |
|
---|
| 2009 | f.plot(residual=True)
|
---|
| 2010 | accept_fit = raw_input("Accept fit ( [y]/n ): ")
|
---|
| 2011 | if accept_fit.upper() == "N":
|
---|
| 2012 | self.blpars.append(None)
|
---|
| 2013 | self.masklists.append(None)
|
---|
| 2014 | self.actualmask.append(None)
|
---|
| 2015 | continue
|
---|
| 2016 | workscan._setspectrum(f.fitter.getresidual(), r)
|
---|
| 2017 | self.blpars.append(f.get_parameters())
|
---|
| 2018 | self.masklists.append(workscan.get_masklist(f.mask, row=r))
|
---|
| 2019 | self.actualmask.append(f.mask)
|
---|
| 2020 |
|
---|
| 2021 | f._p.unmap()
|
---|
| 2022 | f._p = None
|
---|
| 2023 | else:
|
---|
| 2024 | import array
|
---|
| 2025 | for r in rows:
|
---|
| 2026 | pars = array.array("f", [0.0 for i in range(order+1)])
|
---|
| 2027 | pars_adr = pars.buffer_info()[0]
|
---|
| 2028 | pars_len = pars.buffer_info()[1]
|
---|
| 2029 |
|
---|
| 2030 | errs = array.array("f", [0.0 for i in range(order+1)])
|
---|
| 2031 | errs_adr = errs.buffer_info()[0]
|
---|
| 2032 | errs_len = errs.buffer_info()[1]
|
---|
| 2033 |
|
---|
| 2034 | fmsk = array.array("i", [1 for i in range(nchan)])
|
---|
| 2035 | fmsk_adr = fmsk.buffer_info()[0]
|
---|
| 2036 | fmsk_len = fmsk.buffer_info()[1]
|
---|
| 2037 |
|
---|
| 2038 | workscan._poly_baseline(mask, order, r, pars_adr, pars_len, errs_adr, errs_len, fmsk_adr, fmsk_len)
|
---|
| 2039 |
|
---|
| 2040 | params = pars.tolist()
|
---|
| 2041 | fmtd = ""
|
---|
| 2042 | for i in xrange(len(params)): fmtd += " p%d= %3.6f," % (i, params[i])
|
---|
| 2043 | fmtd = fmtd[:-1] # remove trailing ","
|
---|
| 2044 | errors = errs.tolist()
|
---|
| 2045 | fmask = fmsk.tolist()
|
---|
| 2046 | for i in xrange(len(fmask)): fmask[i] = (fmask[i] > 0) # transform (1/0) -> (True/False)
|
---|
| 2047 |
|
---|
| 2048 | self.blpars.append({"params":params, "fixed":[], "formatted":fmtd, "errors":errors})
|
---|
| 2049 | self.masklists.append(workscan.get_masklist(fmask, r))
|
---|
| 2050 | self.actualmask.append(fmask)
|
---|
| 2051 |
|
---|
| 2052 | asaplog.push(str(fmtd))
|
---|
| 2053 |
|
---|
| 2054 | workscan._add_history("poly_baseline", varlist)
|
---|
| 2055 |
|
---|
| 2056 | if insitu:
|
---|
| 2057 | self._assign(workscan)
|
---|
| 2058 | else:
|
---|
| 2059 | return workscan
|
---|
| 2060 |
|
---|
| 2061 | except RuntimeError:
|
---|
| 2062 | msg = "The fit failed, possibly because it didn't converge."
|
---|
| 2063 | if rcParams["verbose"]:
|
---|
| 2064 | asaplog.push(str(msg))
|
---|
| 2065 | return
|
---|
| 2066 | else:
|
---|
| 2067 | raise RuntimeError(msg)
|
---|
| 2068 |
|
---|
| 2069 |
|
---|
| 2070 | def auto_poly_baseline(self, mask=None, edge=(0, 0), order=0,
|
---|
[1280] | 2071 | threshold=3, chan_avg_limit=1, plot=False,
|
---|
[1907] | 2072 | insitu=None, rows=None):
|
---|
[1846] | 2073 | """\
|
---|
[880] | 2074 | Return a scan which has been baselined (all rows) by a polynomial.
|
---|
| 2075 | Spectral lines are detected first using linefinder and masked out
|
---|
| 2076 | to avoid them affecting the baseline solution.
|
---|
| 2077 |
|
---|
| 2078 | Parameters:
|
---|
[1846] | 2079 |
|
---|
[880] | 2080 | mask: an optional mask retreived from scantable
|
---|
[1846] | 2081 |
|
---|
| 2082 | edge: an optional number of channel to drop at the edge of
|
---|
| 2083 | spectrum. If only one value is
|
---|
[880] | 2084 | specified, the same number will be dropped from
|
---|
| 2085 | both sides of the spectrum. Default is to keep
|
---|
[907] | 2086 | all channels. Nested tuples represent individual
|
---|
[976] | 2087 | edge selection for different IFs (a number of spectral
|
---|
| 2088 | channels can be different)
|
---|
[1846] | 2089 |
|
---|
[880] | 2090 | order: the order of the polynomial (default is 0)
|
---|
[1846] | 2091 |
|
---|
[880] | 2092 | threshold: the threshold used by line finder. It is better to
|
---|
| 2093 | keep it large as only strong lines affect the
|
---|
| 2094 | baseline solution.
|
---|
[1846] | 2095 |
|
---|
[1280] | 2096 | chan_avg_limit:
|
---|
| 2097 | a maximum number of consequtive spectral channels to
|
---|
| 2098 | average during the search of weak and broad lines.
|
---|
| 2099 | The default is no averaging (and no search for weak
|
---|
| 2100 | lines). If such lines can affect the fitted baseline
|
---|
| 2101 | (e.g. a high order polynomial is fitted), increase this
|
---|
| 2102 | parameter (usually values up to 8 are reasonable). Most
|
---|
| 2103 | users of this method should find the default value
|
---|
| 2104 | sufficient.
|
---|
[1846] | 2105 |
|
---|
[1061] | 2106 | plot: plot the fit and the residual. In this each
|
---|
| 2107 | indivual fit has to be approved, by typing 'y'
|
---|
| 2108 | or 'n'
|
---|
[1846] | 2109 |
|
---|
[880] | 2110 | insitu: if False a new scantable is returned.
|
---|
| 2111 | Otherwise, the scaling is done in-situ
|
---|
| 2112 | The default is taken from .asaprc (False)
|
---|
[1907] | 2113 | rows: row numbers of spectra to be processed.
|
---|
| 2114 | (default is None: for all rows)
|
---|
[880] | 2115 |
|
---|
[1846] | 2116 |
|
---|
| 2117 | Example::
|
---|
| 2118 |
|
---|
| 2119 | scan2 = scan.auto_poly_baseline(order=7, insitu=False)
|
---|
| 2120 |
|
---|
[880] | 2121 | """
|
---|
| 2122 | if insitu is None: insitu = rcParams['insitu']
|
---|
| 2123 | varlist = vars()
|
---|
| 2124 | from asap.asaplinefind import linefinder
|
---|
| 2125 | from asap import _is_sequence_or_number as _is_valid
|
---|
| 2126 |
|
---|
[976] | 2127 | # check whether edge is set up for each IF individually
|
---|
[1118] | 2128 | individualedge = False;
|
---|
| 2129 | if len(edge) > 1:
|
---|
| 2130 | if isinstance(edge[0], list) or isinstance(edge[0], tuple):
|
---|
| 2131 | individualedge = True;
|
---|
[907] | 2132 |
|
---|
[1118] | 2133 | if not _is_valid(edge, int) and not individualedge:
|
---|
[909] | 2134 | raise ValueError, "Parameter 'edge' has to be an integer or a \
|
---|
[907] | 2135 | pair of integers specified as a tuple. Nested tuples are allowed \
|
---|
| 2136 | to make individual selection for different IFs."
|
---|
[919] | 2137 |
|
---|
[1118] | 2138 | curedge = (0, 0)
|
---|
| 2139 | if individualedge:
|
---|
| 2140 | for edgepar in edge:
|
---|
| 2141 | if not _is_valid(edgepar, int):
|
---|
| 2142 | raise ValueError, "Each element of the 'edge' tuple has \
|
---|
| 2143 | to be a pair of integers or an integer."
|
---|
[907] | 2144 | else:
|
---|
[1118] | 2145 | curedge = edge;
|
---|
[880] | 2146 |
|
---|
[1907] | 2147 | if not insitu:
|
---|
| 2148 | workscan = self.copy()
|
---|
| 2149 | else:
|
---|
| 2150 | workscan = self
|
---|
| 2151 |
|
---|
[880] | 2152 | # setup fitter
|
---|
| 2153 | f = fitter()
|
---|
[1907] | 2154 | f.set_function(lpoly=order)
|
---|
[880] | 2155 |
|
---|
| 2156 | # setup line finder
|
---|
[1118] | 2157 | fl = linefinder()
|
---|
[1268] | 2158 | fl.set_options(threshold=threshold,avg_limit=chan_avg_limit)
|
---|
[880] | 2159 |
|
---|
[907] | 2160 | fl.set_scan(workscan)
|
---|
| 2161 |
|
---|
[1907] | 2162 | if mask is None:
|
---|
| 2163 | mask = _n_bools(workscan.nchan(), True)
|
---|
| 2164 |
|
---|
| 2165 | if rows is None:
|
---|
| 2166 | rows = xrange(workscan.nrow())
|
---|
| 2167 | elif isinstance(rows, int):
|
---|
| 2168 | rows = [ rows ]
|
---|
| 2169 |
|
---|
[1819] | 2170 | # Save parameters of baseline fits & masklists as a class attribute.
|
---|
| 2171 | # NOTICE: It does not reflect changes in scantable!
|
---|
| 2172 | if len(rows) > 0:
|
---|
| 2173 | self.blpars=[]
|
---|
| 2174 | self.masklists=[]
|
---|
[1907] | 2175 | self.actualmask=[]
|
---|
[880] | 2176 | asaplog.push("Processing:")
|
---|
| 2177 | for r in rows:
|
---|
[1118] | 2178 | msg = " Scan[%d] Beam[%d] IF[%d] Pol[%d] Cycle[%d]" % \
|
---|
| 2179 | (workscan.getscan(r), workscan.getbeam(r), workscan.getif(r), \
|
---|
| 2180 | workscan.getpol(r), workscan.getcycle(r))
|
---|
[880] | 2181 | asaplog.push(msg, False)
|
---|
[907] | 2182 |
|
---|
[976] | 2183 | # figure out edge parameter
|
---|
[1118] | 2184 | if individualedge:
|
---|
| 2185 | if len(edge) >= workscan.getif(r):
|
---|
| 2186 | raise RuntimeError, "Number of edge elements appear to " \
|
---|
| 2187 | "be less than the number of IFs"
|
---|
| 2188 | curedge = edge[workscan.getif(r)]
|
---|
[919] | 2189 |
|
---|
[1907] | 2190 | actualmask = mask_and(mask, workscan._getmask(r)) # (CAS-1434)
|
---|
[1819] | 2191 |
|
---|
[976] | 2192 | # setup line finder
|
---|
[1819] | 2193 | fl.find_lines(r, actualmask, curedge)
|
---|
[1907] | 2194 |
|
---|
[1819] | 2195 | f.x = workscan._getabcissa(r)
|
---|
| 2196 | f.y = workscan._getspectrum(r)
|
---|
[1907] | 2197 | f.mask = fl.get_mask()
|
---|
[1819] | 2198 | f.data = None
|
---|
[880] | 2199 | f.fit()
|
---|
[1819] | 2200 |
|
---|
| 2201 | # Show mask list
|
---|
[1907] | 2202 | masklist=workscan.get_masklist(f.mask, row=r)
|
---|
[1819] | 2203 | msg = "mask range: "+str(masklist)
|
---|
| 2204 | asaplog.push(msg, False)
|
---|
| 2205 |
|
---|
[1061] | 2206 | if plot:
|
---|
| 2207 | f.plot(residual=True)
|
---|
| 2208 | x = raw_input("Accept fit ( [y]/n ): ")
|
---|
| 2209 | if x.upper() == 'N':
|
---|
[1819] | 2210 | self.blpars.append(None)
|
---|
| 2211 | self.masklists.append(None)
|
---|
[1907] | 2212 | self.actualmask.append(None)
|
---|
[1061] | 2213 | continue
|
---|
[1819] | 2214 |
|
---|
[880] | 2215 | workscan._setspectrum(f.fitter.getresidual(), r)
|
---|
[1819] | 2216 | self.blpars.append(f.get_parameters())
|
---|
| 2217 | self.masklists.append(masklist)
|
---|
[1907] | 2218 | self.actualmask.append(f.mask)
|
---|
[1061] | 2219 | if plot:
|
---|
| 2220 | f._p.unmap()
|
---|
| 2221 | f._p = None
|
---|
| 2222 | workscan._add_history("auto_poly_baseline", varlist)
|
---|
[880] | 2223 | if insitu:
|
---|
| 2224 | self._assign(workscan)
|
---|
| 2225 | else:
|
---|
| 2226 | return workscan
|
---|
| 2227 |
|
---|
[1862] | 2228 | @asaplog_post_dec
|
---|
[914] | 2229 | def rotate_linpolphase(self, angle):
|
---|
[1846] | 2230 | """\
|
---|
[914] | 2231 | Rotate the phase of the complex polarization O=Q+iU correlation.
|
---|
| 2232 | This is always done in situ in the raw data. So if you call this
|
---|
| 2233 | function more than once then each call rotates the phase further.
|
---|
[1846] | 2234 |
|
---|
[914] | 2235 | Parameters:
|
---|
[1846] | 2236 |
|
---|
[914] | 2237 | angle: The angle (degrees) to rotate (add) by.
|
---|
[1846] | 2238 |
|
---|
| 2239 | Example::
|
---|
| 2240 |
|
---|
[914] | 2241 | scan.rotate_linpolphase(2.3)
|
---|
[1846] | 2242 |
|
---|
[914] | 2243 | """
|
---|
| 2244 | varlist = vars()
|
---|
[936] | 2245 | self._math._rotate_linpolphase(self, angle)
|
---|
[914] | 2246 | self._add_history("rotate_linpolphase", varlist)
|
---|
| 2247 | return
|
---|
[710] | 2248 |
|
---|
[1862] | 2249 | @asaplog_post_dec
|
---|
[914] | 2250 | def rotate_xyphase(self, angle):
|
---|
[1846] | 2251 | """\
|
---|
[914] | 2252 | Rotate the phase of the XY correlation. This is always done in situ
|
---|
| 2253 | in the data. So if you call this function more than once
|
---|
| 2254 | then each call rotates the phase further.
|
---|
[1846] | 2255 |
|
---|
[914] | 2256 | Parameters:
|
---|
[1846] | 2257 |
|
---|
[914] | 2258 | angle: The angle (degrees) to rotate (add) by.
|
---|
[1846] | 2259 |
|
---|
| 2260 | Example::
|
---|
| 2261 |
|
---|
[914] | 2262 | scan.rotate_xyphase(2.3)
|
---|
[1846] | 2263 |
|
---|
[914] | 2264 | """
|
---|
| 2265 | varlist = vars()
|
---|
[936] | 2266 | self._math._rotate_xyphase(self, angle)
|
---|
[914] | 2267 | self._add_history("rotate_xyphase", varlist)
|
---|
| 2268 | return
|
---|
| 2269 |
|
---|
[1862] | 2270 | @asaplog_post_dec
|
---|
[914] | 2271 | def swap_linears(self):
|
---|
[1846] | 2272 | """\
|
---|
[1573] | 2273 | Swap the linear polarisations XX and YY, or better the first two
|
---|
[1348] | 2274 | polarisations as this also works for ciculars.
|
---|
[914] | 2275 | """
|
---|
| 2276 | varlist = vars()
|
---|
[936] | 2277 | self._math._swap_linears(self)
|
---|
[914] | 2278 | self._add_history("swap_linears", varlist)
|
---|
| 2279 | return
|
---|
| 2280 |
|
---|
[1862] | 2281 | @asaplog_post_dec
|
---|
[914] | 2282 | def invert_phase(self):
|
---|
[1846] | 2283 | """\
|
---|
[914] | 2284 | Invert the phase of the complex polarisation
|
---|
| 2285 | """
|
---|
| 2286 | varlist = vars()
|
---|
[936] | 2287 | self._math._invert_phase(self)
|
---|
[914] | 2288 | self._add_history("invert_phase", varlist)
|
---|
| 2289 | return
|
---|
| 2290 |
|
---|
[1862] | 2291 | @asaplog_post_dec
|
---|
[876] | 2292 | def add(self, offset, insitu=None):
|
---|
[1846] | 2293 | """\
|
---|
[513] | 2294 | Return a scan where all spectra have the offset added
|
---|
[1846] | 2295 |
|
---|
[513] | 2296 | Parameters:
|
---|
[1846] | 2297 |
|
---|
[513] | 2298 | offset: the offset
|
---|
[1855] | 2299 |
|
---|
[513] | 2300 | insitu: if False a new scantable is returned.
|
---|
| 2301 | Otherwise, the scaling is done in-situ
|
---|
| 2302 | The default is taken from .asaprc (False)
|
---|
[1846] | 2303 |
|
---|
[513] | 2304 | """
|
---|
| 2305 | if insitu is None: insitu = rcParams['insitu']
|
---|
[876] | 2306 | self._math._setinsitu(insitu)
|
---|
[513] | 2307 | varlist = vars()
|
---|
[876] | 2308 | s = scantable(self._math._unaryop(self, offset, "ADD", False))
|
---|
[1118] | 2309 | s._add_history("add", varlist)
|
---|
[876] | 2310 | if insitu:
|
---|
| 2311 | self._assign(s)
|
---|
| 2312 | else:
|
---|
[513] | 2313 | return s
|
---|
| 2314 |
|
---|
[1862] | 2315 | @asaplog_post_dec
|
---|
[1308] | 2316 | def scale(self, factor, tsys=True, insitu=None):
|
---|
[1846] | 2317 | """\
|
---|
| 2318 |
|
---|
[513] | 2319 | Return a scan where all spectra are scaled by the give 'factor'
|
---|
[1846] | 2320 |
|
---|
[513] | 2321 | Parameters:
|
---|
[1846] | 2322 |
|
---|
[1819] | 2323 | factor: the scaling factor (float or 1D float list)
|
---|
[1855] | 2324 |
|
---|
[513] | 2325 | insitu: if False a new scantable is returned.
|
---|
| 2326 | Otherwise, the scaling is done in-situ
|
---|
| 2327 | The default is taken from .asaprc (False)
|
---|
[1855] | 2328 |
|
---|
[513] | 2329 | tsys: if True (default) then apply the operation to Tsys
|
---|
| 2330 | as well as the data
|
---|
[1846] | 2331 |
|
---|
[513] | 2332 | """
|
---|
| 2333 | if insitu is None: insitu = rcParams['insitu']
|
---|
[876] | 2334 | self._math._setinsitu(insitu)
|
---|
[513] | 2335 | varlist = vars()
|
---|
[1819] | 2336 | s = None
|
---|
| 2337 | import numpy
|
---|
| 2338 | if isinstance(factor, list) or isinstance(factor, numpy.ndarray):
|
---|
| 2339 | if isinstance(factor[0], list) or isinstance(factor[0], numpy.ndarray):
|
---|
| 2340 | from asapmath import _array2dOp
|
---|
| 2341 | s = _array2dOp( self.copy(), factor, "MUL", tsys )
|
---|
| 2342 | else:
|
---|
| 2343 | s = scantable( self._math._arrayop( self.copy(), factor, "MUL", tsys ) )
|
---|
| 2344 | else:
|
---|
| 2345 | s = scantable(self._math._unaryop(self.copy(), factor, "MUL", tsys))
|
---|
[1118] | 2346 | s._add_history("scale", varlist)
|
---|
[876] | 2347 | if insitu:
|
---|
| 2348 | self._assign(s)
|
---|
| 2349 | else:
|
---|
[513] | 2350 | return s
|
---|
| 2351 |
|
---|
[1504] | 2352 | def set_sourcetype(self, match, matchtype="pattern",
|
---|
| 2353 | sourcetype="reference"):
|
---|
[1846] | 2354 | """\
|
---|
[1502] | 2355 | Set the type of the source to be an source or reference scan
|
---|
[1846] | 2356 | using the provided pattern.
|
---|
| 2357 |
|
---|
[1502] | 2358 | Parameters:
|
---|
[1846] | 2359 |
|
---|
[1504] | 2360 | match: a Unix style pattern, regular expression or selector
|
---|
[1855] | 2361 |
|
---|
[1504] | 2362 | matchtype: 'pattern' (default) UNIX style pattern or
|
---|
| 2363 | 'regex' regular expression
|
---|
[1855] | 2364 |
|
---|
[1502] | 2365 | sourcetype: the type of the source to use (source/reference)
|
---|
[1846] | 2366 |
|
---|
[1502] | 2367 | """
|
---|
| 2368 | varlist = vars()
|
---|
| 2369 | basesel = self.get_selection()
|
---|
| 2370 | stype = -1
|
---|
| 2371 | if sourcetype.lower().startswith("r"):
|
---|
| 2372 | stype = 1
|
---|
| 2373 | elif sourcetype.lower().startswith("s"):
|
---|
| 2374 | stype = 0
|
---|
[1504] | 2375 | else:
|
---|
[1502] | 2376 | raise ValueError("Illegal sourcetype use s(ource) or r(eference)")
|
---|
[1504] | 2377 | if matchtype.lower().startswith("p"):
|
---|
| 2378 | matchtype = "pattern"
|
---|
| 2379 | elif matchtype.lower().startswith("r"):
|
---|
| 2380 | matchtype = "regex"
|
---|
| 2381 | else:
|
---|
| 2382 | raise ValueError("Illegal matchtype, use p(attern) or r(egex)")
|
---|
[1502] | 2383 | sel = selector()
|
---|
| 2384 | if isinstance(match, selector):
|
---|
| 2385 | sel = match
|
---|
| 2386 | else:
|
---|
[1504] | 2387 | sel.set_query("SRCNAME == %s('%s')" % (matchtype, match))
|
---|
[1502] | 2388 | self.set_selection(basesel+sel)
|
---|
| 2389 | self._setsourcetype(stype)
|
---|
| 2390 | self.set_selection(basesel)
|
---|
[1573] | 2391 | self._add_history("set_sourcetype", varlist)
|
---|
[1502] | 2392 |
|
---|
[1862] | 2393 | @asaplog_post_dec
|
---|
[1857] | 2394 | @preserve_selection
|
---|
[1819] | 2395 | def auto_quotient(self, preserve=True, mode='paired', verify=False):
|
---|
[1846] | 2396 | """\
|
---|
[670] | 2397 | This function allows to build quotients automatically.
|
---|
[1819] | 2398 | It assumes the observation to have the same number of
|
---|
[670] | 2399 | "ons" and "offs"
|
---|
[1846] | 2400 |
|
---|
[670] | 2401 | Parameters:
|
---|
[1846] | 2402 |
|
---|
[710] | 2403 | preserve: you can preserve (default) the continuum or
|
---|
| 2404 | remove it. The equations used are
|
---|
[1857] | 2405 |
|
---|
[670] | 2406 | preserve: Output = Toff * (on/off) - Toff
|
---|
[1857] | 2407 |
|
---|
[1070] | 2408 | remove: Output = Toff * (on/off) - Ton
|
---|
[1855] | 2409 |
|
---|
[1573] | 2410 | mode: the on/off detection mode
|
---|
[1348] | 2411 | 'paired' (default)
|
---|
| 2412 | identifies 'off' scans by the
|
---|
| 2413 | trailing '_R' (Mopra/Parkes) or
|
---|
| 2414 | '_e'/'_w' (Tid) and matches
|
---|
| 2415 | on/off pairs from the observing pattern
|
---|
[1502] | 2416 | 'time'
|
---|
| 2417 | finds the closest off in time
|
---|
[1348] | 2418 |
|
---|
[1857] | 2419 | .. todo:: verify argument is not implemented
|
---|
| 2420 |
|
---|
[670] | 2421 | """
|
---|
[1857] | 2422 | varlist = vars()
|
---|
[1348] | 2423 | modes = ["time", "paired"]
|
---|
[670] | 2424 | if not mode in modes:
|
---|
[876] | 2425 | msg = "please provide valid mode. Valid modes are %s" % (modes)
|
---|
| 2426 | raise ValueError(msg)
|
---|
[1348] | 2427 | s = None
|
---|
| 2428 | if mode.lower() == "paired":
|
---|
[1857] | 2429 | sel = self.get_selection()
|
---|
[1875] | 2430 | sel.set_query("SRCTYPE==psoff")
|
---|
[1356] | 2431 | self.set_selection(sel)
|
---|
[1348] | 2432 | offs = self.copy()
|
---|
[1875] | 2433 | sel.set_query("SRCTYPE==pson")
|
---|
[1356] | 2434 | self.set_selection(sel)
|
---|
[1348] | 2435 | ons = self.copy()
|
---|
| 2436 | s = scantable(self._math._quotient(ons, offs, preserve))
|
---|
| 2437 | elif mode.lower() == "time":
|
---|
| 2438 | s = scantable(self._math._auto_quotient(self, mode, preserve))
|
---|
[1118] | 2439 | s._add_history("auto_quotient", varlist)
|
---|
[876] | 2440 | return s
|
---|
[710] | 2441 |
|
---|
[1862] | 2442 | @asaplog_post_dec
|
---|
[1145] | 2443 | def mx_quotient(self, mask = None, weight='median', preserve=True):
|
---|
[1846] | 2444 | """\
|
---|
[1143] | 2445 | Form a quotient using "off" beams when observing in "MX" mode.
|
---|
[1846] | 2446 |
|
---|
[1143] | 2447 | Parameters:
|
---|
[1846] | 2448 |
|
---|
[1145] | 2449 | mask: an optional mask to be used when weight == 'stddev'
|
---|
[1855] | 2450 |
|
---|
[1143] | 2451 | weight: How to average the off beams. Default is 'median'.
|
---|
[1855] | 2452 |
|
---|
[1145] | 2453 | preserve: you can preserve (default) the continuum or
|
---|
[1855] | 2454 | remove it. The equations used are:
|
---|
[1846] | 2455 |
|
---|
[1855] | 2456 | preserve: Output = Toff * (on/off) - Toff
|
---|
| 2457 |
|
---|
| 2458 | remove: Output = Toff * (on/off) - Ton
|
---|
| 2459 |
|
---|
[1217] | 2460 | """
|
---|
[1593] | 2461 | mask = mask or ()
|
---|
[1141] | 2462 | varlist = vars()
|
---|
| 2463 | on = scantable(self._math._mx_extract(self, 'on'))
|
---|
[1143] | 2464 | preoff = scantable(self._math._mx_extract(self, 'off'))
|
---|
| 2465 | off = preoff.average_time(mask=mask, weight=weight, scanav=False)
|
---|
[1217] | 2466 | from asapmath import quotient
|
---|
[1145] | 2467 | q = quotient(on, off, preserve)
|
---|
[1143] | 2468 | q._add_history("mx_quotient", varlist)
|
---|
[1217] | 2469 | return q
|
---|
[513] | 2470 |
|
---|
[1862] | 2471 | @asaplog_post_dec
|
---|
[718] | 2472 | def freq_switch(self, insitu=None):
|
---|
[1846] | 2473 | """\
|
---|
[718] | 2474 | Apply frequency switching to the data.
|
---|
[1846] | 2475 |
|
---|
[718] | 2476 | Parameters:
|
---|
[1846] | 2477 |
|
---|
[718] | 2478 | insitu: if False a new scantable is returned.
|
---|
| 2479 | Otherwise, the swictching is done in-situ
|
---|
| 2480 | The default is taken from .asaprc (False)
|
---|
[1846] | 2481 |
|
---|
[718] | 2482 | """
|
---|
| 2483 | if insitu is None: insitu = rcParams['insitu']
|
---|
[876] | 2484 | self._math._setinsitu(insitu)
|
---|
[718] | 2485 | varlist = vars()
|
---|
[876] | 2486 | s = scantable(self._math._freqswitch(self))
|
---|
[1118] | 2487 | s._add_history("freq_switch", varlist)
|
---|
[1856] | 2488 | if insitu:
|
---|
| 2489 | self._assign(s)
|
---|
| 2490 | else:
|
---|
| 2491 | return s
|
---|
[718] | 2492 |
|
---|
[1862] | 2493 | @asaplog_post_dec
|
---|
[780] | 2494 | def recalc_azel(self):
|
---|
[1846] | 2495 | """Recalculate the azimuth and elevation for each position."""
|
---|
[780] | 2496 | varlist = vars()
|
---|
[876] | 2497 | self._recalcazel()
|
---|
[780] | 2498 | self._add_history("recalc_azel", varlist)
|
---|
| 2499 | return
|
---|
| 2500 |
|
---|
[1862] | 2501 | @asaplog_post_dec
|
---|
[513] | 2502 | def __add__(self, other):
|
---|
| 2503 | varlist = vars()
|
---|
| 2504 | s = None
|
---|
| 2505 | if isinstance(other, scantable):
|
---|
[1573] | 2506 | s = scantable(self._math._binaryop(self, other, "ADD"))
|
---|
[513] | 2507 | elif isinstance(other, float):
|
---|
[876] | 2508 | s = scantable(self._math._unaryop(self, other, "ADD", False))
|
---|
[513] | 2509 | else:
|
---|
[718] | 2510 | raise TypeError("Other input is not a scantable or float value")
|
---|
[513] | 2511 | s._add_history("operator +", varlist)
|
---|
| 2512 | return s
|
---|
| 2513 |
|
---|
[1862] | 2514 | @asaplog_post_dec
|
---|
[513] | 2515 | def __sub__(self, other):
|
---|
| 2516 | """
|
---|
| 2517 | implicit on all axes and on Tsys
|
---|
| 2518 | """
|
---|
| 2519 | varlist = vars()
|
---|
| 2520 | s = None
|
---|
| 2521 | if isinstance(other, scantable):
|
---|
[1588] | 2522 | s = scantable(self._math._binaryop(self, other, "SUB"))
|
---|
[513] | 2523 | elif isinstance(other, float):
|
---|
[876] | 2524 | s = scantable(self._math._unaryop(self, other, "SUB", False))
|
---|
[513] | 2525 | else:
|
---|
[718] | 2526 | raise TypeError("Other input is not a scantable or float value")
|
---|
[513] | 2527 | s._add_history("operator -", varlist)
|
---|
| 2528 | return s
|
---|
[710] | 2529 |
|
---|
[1862] | 2530 | @asaplog_post_dec
|
---|
[513] | 2531 | def __mul__(self, other):
|
---|
| 2532 | """
|
---|
| 2533 | implicit on all axes and on Tsys
|
---|
| 2534 | """
|
---|
| 2535 | varlist = vars()
|
---|
| 2536 | s = None
|
---|
| 2537 | if isinstance(other, scantable):
|
---|
[1588] | 2538 | s = scantable(self._math._binaryop(self, other, "MUL"))
|
---|
[513] | 2539 | elif isinstance(other, float):
|
---|
[876] | 2540 | s = scantable(self._math._unaryop(self, other, "MUL", False))
|
---|
[513] | 2541 | else:
|
---|
[718] | 2542 | raise TypeError("Other input is not a scantable or float value")
|
---|
[513] | 2543 | s._add_history("operator *", varlist)
|
---|
| 2544 | return s
|
---|
| 2545 |
|
---|
[710] | 2546 |
|
---|
[1862] | 2547 | @asaplog_post_dec
|
---|
[513] | 2548 | def __div__(self, other):
|
---|
| 2549 | """
|
---|
| 2550 | implicit on all axes and on Tsys
|
---|
| 2551 | """
|
---|
| 2552 | varlist = vars()
|
---|
| 2553 | s = None
|
---|
| 2554 | if isinstance(other, scantable):
|
---|
[1589] | 2555 | s = scantable(self._math._binaryop(self, other, "DIV"))
|
---|
[513] | 2556 | elif isinstance(other, float):
|
---|
| 2557 | if other == 0.0:
|
---|
[718] | 2558 | raise ZeroDivisionError("Dividing by zero is not recommended")
|
---|
[876] | 2559 | s = scantable(self._math._unaryop(self, other, "DIV", False))
|
---|
[513] | 2560 | else:
|
---|
[718] | 2561 | raise TypeError("Other input is not a scantable or float value")
|
---|
[513] | 2562 | s._add_history("operator /", varlist)
|
---|
| 2563 | return s
|
---|
| 2564 |
|
---|
[1862] | 2565 | @asaplog_post_dec
|
---|
[530] | 2566 | def get_fit(self, row=0):
|
---|
[1846] | 2567 | """\
|
---|
[530] | 2568 | Print or return the stored fits for a row in the scantable
|
---|
[1846] | 2569 |
|
---|
[530] | 2570 | Parameters:
|
---|
[1846] | 2571 |
|
---|
[530] | 2572 | row: the row which the fit has been applied to.
|
---|
[1846] | 2573 |
|
---|
[530] | 2574 | """
|
---|
| 2575 | if row > self.nrow():
|
---|
| 2576 | return
|
---|
[976] | 2577 | from asap.asapfit import asapfit
|
---|
[530] | 2578 | fit = asapfit(self._getfit(row))
|
---|
[1859] | 2579 | asaplog.push( '%s' %(fit) )
|
---|
| 2580 | return fit.as_dict()
|
---|
[530] | 2581 |
|
---|
[1483] | 2582 | def flag_nans(self):
|
---|
[1846] | 2583 | """\
|
---|
[1483] | 2584 | Utility function to flag NaN values in the scantable.
|
---|
| 2585 | """
|
---|
| 2586 | import numpy
|
---|
| 2587 | basesel = self.get_selection()
|
---|
| 2588 | for i in range(self.nrow()):
|
---|
[1589] | 2589 | sel = self.get_row_selector(i)
|
---|
| 2590 | self.set_selection(basesel+sel)
|
---|
[1483] | 2591 | nans = numpy.isnan(self._getspectrum(0))
|
---|
| 2592 | if numpy.any(nans):
|
---|
| 2593 | bnans = [ bool(v) for v in nans]
|
---|
| 2594 | self.flag(bnans)
|
---|
| 2595 | self.set_selection(basesel)
|
---|
| 2596 |
|
---|
[1588] | 2597 | def get_row_selector(self, rowno):
|
---|
| 2598 | return selector(beams=self.getbeam(rowno),
|
---|
| 2599 | ifs=self.getif(rowno),
|
---|
| 2600 | pols=self.getpol(rowno),
|
---|
| 2601 | scans=self.getscan(rowno),
|
---|
| 2602 | cycles=self.getcycle(rowno))
|
---|
[1573] | 2603 |
|
---|
[484] | 2604 | def _add_history(self, funcname, parameters):
|
---|
[1435] | 2605 | if not rcParams['scantable.history']:
|
---|
| 2606 | return
|
---|
[484] | 2607 | # create date
|
---|
| 2608 | sep = "##"
|
---|
| 2609 | from datetime import datetime
|
---|
| 2610 | dstr = datetime.now().strftime('%Y/%m/%d %H:%M:%S')
|
---|
| 2611 | hist = dstr+sep
|
---|
| 2612 | hist += funcname+sep#cdate+sep
|
---|
| 2613 | if parameters.has_key('self'): del parameters['self']
|
---|
[1118] | 2614 | for k, v in parameters.iteritems():
|
---|
[484] | 2615 | if type(v) is dict:
|
---|
[1118] | 2616 | for k2, v2 in v.iteritems():
|
---|
[484] | 2617 | hist += k2
|
---|
| 2618 | hist += "="
|
---|
[1118] | 2619 | if isinstance(v2, scantable):
|
---|
[484] | 2620 | hist += 'scantable'
|
---|
| 2621 | elif k2 == 'mask':
|
---|
[1118] | 2622 | if isinstance(v2, list) or isinstance(v2, tuple):
|
---|
[513] | 2623 | hist += str(self._zip_mask(v2))
|
---|
| 2624 | else:
|
---|
| 2625 | hist += str(v2)
|
---|
[484] | 2626 | else:
|
---|
[513] | 2627 | hist += str(v2)
|
---|
[484] | 2628 | else:
|
---|
| 2629 | hist += k
|
---|
| 2630 | hist += "="
|
---|
[1118] | 2631 | if isinstance(v, scantable):
|
---|
[484] | 2632 | hist += 'scantable'
|
---|
| 2633 | elif k == 'mask':
|
---|
[1118] | 2634 | if isinstance(v, list) or isinstance(v, tuple):
|
---|
[513] | 2635 | hist += str(self._zip_mask(v))
|
---|
| 2636 | else:
|
---|
| 2637 | hist += str(v)
|
---|
[484] | 2638 | else:
|
---|
| 2639 | hist += str(v)
|
---|
| 2640 | hist += sep
|
---|
| 2641 | hist = hist[:-2] # remove trailing '##'
|
---|
| 2642 | self._addhistory(hist)
|
---|
| 2643 |
|
---|
[710] | 2644 |
|
---|
[484] | 2645 | def _zip_mask(self, mask):
|
---|
| 2646 | mask = list(mask)
|
---|
| 2647 | i = 0
|
---|
| 2648 | segments = []
|
---|
| 2649 | while mask[i:].count(1):
|
---|
| 2650 | i += mask[i:].index(1)
|
---|
| 2651 | if mask[i:].count(0):
|
---|
| 2652 | j = i + mask[i:].index(0)
|
---|
| 2653 | else:
|
---|
[710] | 2654 | j = len(mask)
|
---|
[1118] | 2655 | segments.append([i, j])
|
---|
[710] | 2656 | i = j
|
---|
[484] | 2657 | return segments
|
---|
[714] | 2658 |
|
---|
[626] | 2659 | def _get_ordinate_label(self):
|
---|
| 2660 | fu = "("+self.get_fluxunit()+")"
|
---|
| 2661 | import re
|
---|
| 2662 | lbl = "Intensity"
|
---|
[1118] | 2663 | if re.match(".K.", fu):
|
---|
[626] | 2664 | lbl = "Brightness Temperature "+ fu
|
---|
[1118] | 2665 | elif re.match(".Jy.", fu):
|
---|
[626] | 2666 | lbl = "Flux density "+ fu
|
---|
| 2667 | return lbl
|
---|
[710] | 2668 |
|
---|
[876] | 2669 | def _check_ifs(self):
|
---|
| 2670 | nchans = [self.nchan(i) for i in range(self.nif(-1))]
|
---|
[889] | 2671 | nchans = filter(lambda t: t > 0, nchans)
|
---|
[876] | 2672 | return (sum(nchans)/len(nchans) == nchans[0])
|
---|
[976] | 2673 |
|
---|
[1862] | 2674 | @asaplog_post_dec
|
---|
[1916] | 2675 | #def _fill(self, names, unit, average, getpt, antenna):
|
---|
| 2676 | def _fill(self, names, unit, average, opts={}):
|
---|
[976] | 2677 | first = True
|
---|
| 2678 | fullnames = []
|
---|
| 2679 | for name in names:
|
---|
| 2680 | name = os.path.expandvars(name)
|
---|
| 2681 | name = os.path.expanduser(name)
|
---|
| 2682 | if not os.path.exists(name):
|
---|
| 2683 | msg = "File '%s' does not exists" % (name)
|
---|
| 2684 | raise IOError(msg)
|
---|
| 2685 | fullnames.append(name)
|
---|
| 2686 | if average:
|
---|
| 2687 | asaplog.push('Auto averaging integrations')
|
---|
[1079] | 2688 | stype = int(rcParams['scantable.storage'].lower() == 'disk')
|
---|
[976] | 2689 | for name in fullnames:
|
---|
[1073] | 2690 | tbl = Scantable(stype)
|
---|
[1843] | 2691 | r = filler(tbl)
|
---|
[1504] | 2692 | rx = rcParams['scantable.reference']
|
---|
[1843] | 2693 | r.setreferenceexpr(rx)
|
---|
[976] | 2694 | msg = "Importing %s..." % (name)
|
---|
[1118] | 2695 | asaplog.push(msg, False)
|
---|
[1916] | 2696 | #opts = {'ms': {'antenna' : antenna, 'getpt': getpt} }
|
---|
[1904] | 2697 | r.open(name, opts)# antenna, -1, -1, getpt)
|
---|
[1843] | 2698 | r.fill()
|
---|
[976] | 2699 | if average:
|
---|
[1118] | 2700 | tbl = self._math._average((tbl, ), (), 'NONE', 'SCAN')
|
---|
[976] | 2701 | if not first:
|
---|
| 2702 | tbl = self._math._merge([self, tbl])
|
---|
| 2703 | Scantable.__init__(self, tbl)
|
---|
[1843] | 2704 | r.close()
|
---|
[1118] | 2705 | del r, tbl
|
---|
[976] | 2706 | first = False
|
---|
[1861] | 2707 | #flush log
|
---|
| 2708 | asaplog.post()
|
---|
[976] | 2709 | if unit is not None:
|
---|
| 2710 | self.set_fluxunit(unit)
|
---|
[1824] | 2711 | if not is_casapy():
|
---|
| 2712 | self.set_freqframe(rcParams['scantable.freqframe'])
|
---|
[976] | 2713 |
|
---|
[1402] | 2714 | def __getitem__(self, key):
|
---|
| 2715 | if key < 0:
|
---|
| 2716 | key += self.nrow()
|
---|
| 2717 | if key >= self.nrow():
|
---|
| 2718 | raise IndexError("Row index out of range.")
|
---|
| 2719 | return self._getspectrum(key)
|
---|
| 2720 |
|
---|
| 2721 | def __setitem__(self, key, value):
|
---|
| 2722 | if key < 0:
|
---|
| 2723 | key += self.nrow()
|
---|
| 2724 | if key >= self.nrow():
|
---|
| 2725 | raise IndexError("Row index out of range.")
|
---|
| 2726 | if not hasattr(value, "__len__") or \
|
---|
| 2727 | len(value) > self.nchan(self.getif(key)):
|
---|
| 2728 | raise ValueError("Spectrum length doesn't match.")
|
---|
| 2729 | return self._setspectrum(value, key)
|
---|
| 2730 |
|
---|
| 2731 | def __len__(self):
|
---|
| 2732 | return self.nrow()
|
---|
| 2733 |
|
---|
| 2734 | def __iter__(self):
|
---|
| 2735 | for i in range(len(self)):
|
---|
| 2736 | yield self[i]
|
---|