def columns(self):
if self._has_data and hasattr(self.data, '_coldefs'):
return self.data._coldefs
format = self._bitpix2tform[self._header['BITPIX']]
pcount = self._header['PCOUNT']
parnames = []
bscales = []
bzeros = []
for idx in range(pcount):
bscales.append(self._header.get('PSCAL' + str(idx + 1), None))
bzeros.append(self._header.get('PZERO' + str(idx + 1), None))
parnames.append(self._header['PTYPE' + str(idx + 1)])
formats = [format] * len(parnames)
dim = [None] * len(parnames)
# Now create columns from collected parameters, but first add the DATA
# column too, to contain the group data.
parnames.append('DATA')
bscales.append(self._header.get('BSCALE'))
bzeros.append(self._header.get('BZEROS'))
data_shape = self.shape[:-1]
formats.append(str(int(np.prod(data_shape))) + format)
dim.append(data_shape)
parnames = _unique_parnames(parnames)
self._data_field = parnames[-1]
cols = [
Column(name=name, format=fmt, bscale=bscale, bzero=bzero,
dim=dim) for name, fmt, bscale, bzero, dim in zip(
parnames, formats, bscales, bzeros, dim)
]
coldefs = ColDefs(cols)
return coldefs
def __new__(cls, input=None, bitpix=None, pardata=None, parnames=[],
bscale=None, bzero=None, parbscales=None, parbzeros=None):
"""
Parameters
----------
input : array or FITS_rec instance
input data, either the group data itself (a
`numpy.ndarray`) or a record array (`FITS_rec`) which will
contain both group parameter info and the data. The rest
of the arguments are used only for the first case.
bitpix : int
data type as expressed in FITS ``BITPIX`` value (8, 16, 32,
64, -32, or -64)
pardata : sequence of arrays
parameter data, as a list of (numeric) arrays.
parnames : sequence of str
list of parameter names.
bscale : int
``BSCALE`` of the data
bzero : int
``BZERO`` of the data
parbscales : sequence of int
list of bscales for the parameters
parbzeros : sequence of int
list of bzeros for the parameters
"""
if not isinstance(input, FITS_rec):
if pardata is None:
npars = 0
else:
npars = len(pardata)
if parbscales is None:
parbscales = [None] * npars
if parbzeros is None:
parbzeros = [None] * npars
if parnames is None:
parnames = ['PAR{}'.format(idx + 1) for idx in range(npars)]
if len(parnames) != npars:
raise ValueError('The number of parameter data arrays does '
'not match the number of parameters.')
unique_parnames = _unique_parnames(parnames + ['DATA'])
if bitpix is None:
bitpix = DTYPE2BITPIX[input.dtype.name]
fits_fmt = GroupsHDU._bitpix2tform[bitpix] # -32 -> 'E'
format = FITS2NUMPY[fits_fmt] # 'E' -> 'f4'
data_fmt = '{}{}'.format(str(input.shape[1:]), format)
formats = ','.join(([format] * npars) + [data_fmt])
gcount = input.shape[0]
cols = [Column(name=unique_parnames[idx], format=fits_fmt,
bscale=parbscales[idx], bzero=parbzeros[idx])
for idx in range(npars)]
cols.append(Column(name=unique_parnames[-1], format=fits_fmt,
bscale=bscale, bzero=bzero))
coldefs = ColDefs(cols)
self = FITS_rec.__new__(cls,
np.rec.array(None,
formats=formats,
names=coldefs.names,
shape=gcount))
# By default the data field will just be 'DATA', but it may be
# uniquified if 'DATA' is already used by one of the group names
self._data_field = unique_parnames[-1]
self._coldefs = coldefs
self.parnames = parnames
for idx, name in enumerate(unique_parnames[:-1]):
column = coldefs[idx]
# Note: _get_scale_factors is used here and in other cases
# below to determine whether the column has non-default
# scale/zero factors.
# TODO: Find a better way to do this than using this interface
scale, zero = self._get_scale_factors(column)[3:5]
if scale or zero:
self._cache_field(name, pardata[idx])
else:
np.rec.recarray.field(self, idx)[:] = pardata[idx]
column = coldefs[self._data_field]
scale, zero = self._get_scale_factors(column)[3:5]
if scale or zero:
self._cache_field(self._data_field, input)
else:
np.rec.recarray.field(self, npars)[:] = input
else:
self = FITS_rec.__new__(cls, input)
self.parnames = None
return self