def _parse_html_result(self, response, verbose=False):
# parse the HTML return...
root = BeautifulSoup(response.content, 'html5lib')
htmltable = root.findAll('table')
# if len(htmltable) != 1:
# raise ValueError("Found the wrong number of tables: {0}"
# .format(len(htmltable)))
string_to_parse = htmltable[-1].encode('ascii')
if six.PY2:
from astropy.io.ascii import html
from astropy.io.ascii.core import convert_numpy
htmlreader = html.HTML({'parser': 'html5lib'})
htmlreader.outputter.default_converters.append(convert_numpy(np.unicode))
table = htmlreader.read(string_to_parse)
else:
table = Table.read(string_to_parse.decode('utf-8'), format='ascii.html')
return table
def _convert_vals(self, cols):
"""READ: Convert str_vals in `cols` to final arrays with correct dtypes.
This is adapted from ``BaseOutputter._convert_vals``. In the case of ECSV
there is no guessing and all types are known in advance. A big change
is handling the possibility of JSON-encoded values, both unstructured
object data and structured values that may contain masked data.
"""
for col in cols:
try:
# 1-d or N-d object columns are serialized as JSON.
if col.subtype == 'object':
_check_dtype_is_str(col)
col_vals = [json.loads(val) for val in col.str_vals]
col.data = np.empty([len(col_vals)] + col.shape,
dtype=object)
col.data[...] = col_vals
# Variable length arrays with shape (n, m, ..., *) for fixed
# n, m, .. and variable in last axis. Masked values here are
# not currently supported.
elif col.shape and col.shape[-1] is None:
_check_dtype_is_str(col)
# Remake as a 1-d object column of numpy ndarrays using the
# datatype specified in the ECSV file.
col_vals = [
np.array(json.loads(val), dtype=col.subtype)
for val in col.str_vals
]
col.shape = ()
col.dtype = np.dtype(object)
# np.array(col_vals_arr, dtype=object) fails ?? so this workaround:
col.data = np.empty(len(col_vals), dtype=object)
col.data[:] = col_vals
# Multidim columns with consistent shape (n, m, ...). These
# might be masked.
elif col.shape:
_check_dtype_is_str(col)
col_vals = [json.loads(val) for val in col.str_vals]
# Make a numpy object array of col_vals to look for None
# (masked values)
data = np.array(col_vals, dtype=object)
mask = (data == None) # noqa: E711
if not np.any(mask):
# No None's, just convert to required dtype
col.data = data.astype(col.subtype)
else:
# Replace all the None with an appropriate fill value
kind = np.dtype(col.subtype).kind
data[mask] = {'U': '', 'S': b''}.get(kind, 0)
# Finally make a MaskedArray with the filled data + mask
col.data = np.ma.array(data.astype(col.subtype),
mask=mask)
# Regular scalar value column
else:
if col.subtype:
warnings.warn(
f'unexpected subtype {col.subtype!r} set for column '
f'{col.name!r}, using dtype={col.dtype!r} instead.',
category=AstropyUserWarning)
converter_func, _ = convert_numpy(col.dtype)
col.data = converter_func(col.str_vals)
if col.data.shape[1:] != tuple(col.shape):
raise ValueError(
'shape mismatch between value and column specifier')
except json.JSONDecodeError:
raise ValueError(f'column {col.name!r} failed to convert: '
'column value is not valid JSON')
except Exception as exc:
raise ValueError(
f'column {col.name!r} failed to convert: {exc}')
def _convert_vals(self, cols):
"""READ: Convert str_vals in `cols` to final arrays with correct dtypes.
This is adapted from ``BaseOutputter._convert_vals``. In the case of ECSV
there is no guessing and all types are known in advance. A big change
is handling the possibility of JSON-encoded values, both unstructured
object data and structured values that may contain masked data.
"""
for col in cols:
try:
# 1-d or N-d object columns are serialized as JSON.
if col.subtype == 'object':
_check_dtype_is_str(col)
col_vals = [json.loads(val) for val in col.str_vals]
col.data = np.empty([len(col_vals)] + col.shape,
dtype=object)
col.data[...] = col_vals
# Variable length arrays with shape (n, m, ..., *) for fixed
# n, m, .. and variable in last axis. Masked values here are
# not currently supported.
elif col.shape and col.shape[-1] is None:
_check_dtype_is_str(col)
# Empty (blank) values in original ECSV are changed to "0"
# in str_vals with corresponding col.mask being created and
# set accordingly. Instead use an empty list here.
if hasattr(col, 'mask'):
for idx in np.nonzero(col.mask)[0]:
col.str_vals[idx] = '[]'
# Remake as a 1-d object column of numpy ndarrays or
# MaskedArray using the datatype specified in the ECSV file.
col_vals = []
for str_val in col.str_vals:
obj_val = json.loads(str_val) # list or nested lists
try:
arr_val = np.array(obj_val, dtype=col.subtype)
except TypeError:
# obj_val has entries that are inconsistent with
# dtype. For a valid ECSV file the only possibility
# is None values (indicating missing values).
data = np.array(obj_val, dtype=object)
# Replace all the None with an appropriate fill value
mask = (data == None) # noqa: E711
kind = np.dtype(col.subtype).kind
data[mask] = {'U': '', 'S': b''}.get(kind, 0)
arr_val = np.ma.array(data.astype(col.subtype),
mask=mask)
col_vals.append(arr_val)
col.shape = ()
col.dtype = np.dtype(object)
# np.array(col_vals_arr, dtype=object) fails ?? so this workaround:
col.data = np.empty(len(col_vals), dtype=object)
col.data[:] = col_vals
# Multidim columns with consistent shape (n, m, ...). These
# might be masked.
elif col.shape:
_check_dtype_is_str(col)
# Change empty (blank) values in original ECSV to something
# like "[[null, null],[null,null]]" so subsequent JSON
# decoding works. Delete `col.mask` so that later code in
# core TableOutputter.__call__() that deals with col.mask
# does not run (since handling is done here already).
if hasattr(col, 'mask'):
all_none_arr = np.full(shape=col.shape,
fill_value=None,
dtype=object)
all_none_json = json.dumps(all_none_arr.tolist())
for idx in np.nonzero(col.mask)[0]:
col.str_vals[idx] = all_none_json
del col.mask
col_vals = [json.loads(val) for val in col.str_vals]
# Make a numpy object array of col_vals to look for None
# (masked values)
data = np.array(col_vals, dtype=object)
mask = (data == None) # noqa: E711
if not np.any(mask):
# No None's, just convert to required dtype
col.data = data.astype(col.subtype)
else:
# Replace all the None with an appropriate fill value
kind = np.dtype(col.subtype).kind
data[mask] = {'U': '', 'S': b''}.get(kind, 0)
# Finally make a MaskedArray with the filled data + mask
col.data = np.ma.array(data.astype(col.subtype),
mask=mask)
# Regular scalar value column
else:
if col.subtype:
warnings.warn(
f'unexpected subtype {col.subtype!r} set for column '
f'{col.name!r}, using dtype={col.dtype!r} instead.',
category=InvalidEcsvDatatypeWarning)
converter_func, _ = convert_numpy(col.dtype)
col.data = converter_func(col.str_vals)
if col.data.shape[1:] != tuple(col.shape):
raise ValueError(
'shape mismatch between value and column specifier')
except json.JSONDecodeError:
raise ValueError(f'column {col.name!r} failed to convert: '
'column value is not valid JSON')
except Exception as exc:
raise ValueError(
f'column {col.name!r} failed to convert: {exc}')
