def _nddata_to_glue_data(ndd, data_label):
if ndd.data.ndim != 2:
raise ValueError(f'Imviz cannot load this NDData with ndim={ndd.data.ndim}')
for attrib in ['data', 'mask', 'uncertainty']:
arr = getattr(ndd, attrib)
if arr is None:
continue
comp_label = attrib.upper()
cur_label = f'{data_label}[{comp_label}]'
cur_data = Data(label=cur_label)
cur_data.meta.update(ndd.meta)
if ndd.wcs is not None:
cur_data.coords = ndd.wcs
raw_arr = arr
if attrib == 'data':
bunit = ndd.unit or ''
elif attrib == 'uncertainty':
raw_arr = arr.array
bunit = arr.unit or ''
else:
bunit = ''
component = Component.autotyped(raw_arr, units=bunit)
cur_data.add_component(component=component, label=comp_label)
yield cur_data, cur_label
def test_image():
data = Data(label="Test Image")
comp_a = Component(np.ones((25, 25)))
data.add_component(comp_a, 'test_1')
comp_b = Component(np.zeros((25, 25)))
data.add_component(comp_b, 'test_2')
return data
def astropy_tabular_data(*args, **kwargs):
"""
Build a data set from a table. We restrict ourselves to tables
with 1D columns.
All arguments are passed to
astropy.table.Table.read(...).
"""
result = Data()
table = astropy_table_read(*args, **kwargs)
result.meta = table.meta
# Loop through columns and make component list
for column_name in table.columns:
c = table[column_name]
u = c.unit if hasattr(c, 'unit') else c.units
if table.masked:
# fill array for now
try:
c = c.filled(fill_value=np.nan)
except (ValueError, TypeError): # assigning nan to integer dtype
c = c.filled(fill_value=-1)
nc = Component.autotyped(c, units=u)
result.add_component(nc, column_name)
return result
def astropy_tabular_data(*args, **kwargs):
"""
Build a data set from a table. We restrict ourselves to tables
with 1D columns.
All arguments are passed to
astropy.table.Table.read(...).
"""
result = Data()
table = astropy_table_read(*args, **kwargs)
result.meta = table.meta
# Loop through columns and make component list
for column_name in table.columns:
c = table[column_name]
u = c.unit if hasattr(c, 'unit') else c.units
if table.masked:
# fill array for now
try:
c = c.filled(fill_value=np.nan)
except (ValueError, TypeError): # assigning nan to integer dtype
c = c.filled(fill_value=-1)
nc = Component.autotyped(c, units=u)
result.add_component(nc, column_name)
return result
def merge(self, *data, **kwargs):
"""
Merge two or more datasets into a single dataset.
This has the following effects:
All components from all datasets are added to the first argument
All datasets except the first argument are removed from the collection
Any component name conflicts are disambiguated
The pixel and world components apart from the first argument are discarded
:note: All arguments must have the same shape
:param data: One or more :class:`~glue.core.data.Data` instances.
:returns: self
"""
if len(data) < 2:
raise ValueError("merge requires 2 or more arguments")
shp = data[0].shape
for d in data:
if d.shape != shp:
raise ValueError("All arguments must have the same shape")
label = kwargs.get('label', data[0].label)
master = Data(label=label)
self.append(master)
master.coords = data[0].coords
for i, d in enumerate(data):
if isinstance(d.coords, WCSCoordinates):
master.coords = d.coords
break
# Find ambiguous components (ones which have labels in more than one
# dataset
from collections import Counter
clabel_count = Counter(
[c.label for d in data for c in d.visible_components])
for d in data:
for c in d.components:
if c in master.components: # already present (via a link)
continue
lbl = c.label
if clabel_count[lbl] > 1:
lbl = lbl + " [{0}]".format(d.label)
c._label = lbl
c.parent = master
master.add_component(d.get_component(c), c)
self.remove(d)
return self
def test_histogram_data():
data = Data(label="Test Data")
comp_a = Component(np.random.uniform(size=500))
comp_b = Component(np.random.normal(size=500))
data.add_component(comp_a, 'uniform')
data.add_component(comp_b, 'normal')
return data
def panda_process(indf):
"""
Build a data set from a table using pandas. This attempts to respect
categorical data input by letting pandas.read_csv infer the type
"""
result = Data()
for name, column in indf.iteritems():
if (column.dtype == np.object) | (column.dtype == np.bool):
# try to salvage numerical data
coerced = column.convert_objects(convert_numeric=True)
if (coerced.dtype != column.dtype) and coerced.isnull().mean() < 0.4:
c = Component(coerced.values)
else:
# pandas has a 'special' nan implementation and this doesn't
# play well with np.unique
c = CategoricalComponent(column.fillna(''))
else:
c = Component(column.values)
# convert header to string - in some cases if the first row contains
# numbers, these are cast to numerical types, so we want to change that
# here.
if not isinstance(name, six.string_types):
name = str(name)
# strip off leading #
name = name.strip()
if name.startswith('#'):
name = name[1:].strip()
result.add_component(c, name)
return result
def npz_reader(filename, format='auto', auto_merge=False, **kwargs):
"""
Read in a Numpy structured array saved to a .npy or .npz file.
Parameters
----------
source: str
The pathname to the Numpy save file.
"""
import numpy as np
npy_data = np.load(filename)
groups = []
for groupname in sorted(npy_data.files):
d = Data(label=groupname)
arr = npy_data[groupname]
if not hasattr(arr.dtype, 'names'):
raise ValueError("Numpy save file loading currently only supports structured"
" arrays, e.g., with specified names.")
for name in arr.dtype.names:
comp = Component.autotyped(arr[name])
d.add_component(comp, label=name)
groups.append(d)
return groups
def npy_npz_reader(filename, format='auto', auto_merge=False, **kwargs):
"""
Read in a Numpy structured array saved to a .npy or .npz file.
Parameters
----------
source: str
The pathname to the Numpy save file.
"""
import numpy as np
data = np.load(filename)
if isinstance(data, np.ndarray):
data = {None: data}
groups = []
for groupname in sorted(data):
d = Data(label=groupname)
arr = data[groupname]
if arr.dtype.names is None:
comp = Component.autotyped(arr)
d.add_component(comp, label='array')
else:
for name in arr.dtype.names:
comp = Component.autotyped(arr[name])
d.add_component(comp, label=name)
groups.append(d)
return groups
def _load_data(rec, context):
label = rec['label']
result = Data(label=label)
result.coords = context.object(rec['coords'])
# we manually rebuild pixel/world components, so
# we override this function. This is pretty ugly
result._create_pixel_and_world_components = lambda: None
comps = [list(map(context.object, [cid, comp]))
for cid, comp in rec['components']]
comps = sorted(comps,
key=lambda x: isinstance(x[1], (DerivedComponent,
CoordinateComponent)))
for cid, comp in comps:
if isinstance(comp, CoordinateComponent):
comp._data = result
result.add_component(comp, cid)
assert result._world_component_ids == []
coord = [c for c in comps if isinstance(c[1], CoordinateComponent)]
coord = [x[0] for x in sorted(coord, key=lambda x: x[1])]
assert len(coord) == result.ndim * 2
result._world_component_ids = coord[:len(coord) // 2]
result._pixel_component_ids = coord[len(coord) // 2:]
for s in rec['subsets']:
result.add_subset(context.object(s))
return result
def npy_npz_reader(filename, format='auto', auto_merge=False, **kwargs):
"""
Read in a Numpy structured array saved to a .npy or .npz file.
Parameters
----------
source: str
The pathname to the Numpy save file.
"""
import numpy as np
data = np.load(filename)
if isinstance(data, np.ndarray):
data = {None: data}
groups = []
for groupname in sorted(data):
d = Data(label=groupname)
arr = data[groupname]
if arr.dtype.names is None:
comp = Component.autotyped(arr)
d.add_component(comp, label='array')
else:
for name in arr.dtype.names:
comp = Component.autotyped(arr[name])
d.add_component(comp, label=name)
groups.append(d)
return groups
def panda_process(indf):
"""
Build a data set from a table using pandas. This attempts to respect
categorical data input by letting pandas.read_csv infer the type
"""
result = Data()
for name, column in indf.iteritems():
if (column.dtype == np.object) | (column.dtype == np.bool):
# try to salvage numerical data
coerced = column.convert_objects(convert_numeric=True)
if (coerced.dtype !=
column.dtype) and coerced.isnull().mean() < 0.4:
c = Component(coerced.values)
else:
# pandas has a 'special' nan implementation and this doesn't
# play well with np.unique
c = CategoricalComponent(column.fillna(''))
else:
c = Component(column.values)
# convert header to string - in some cases if the first row contains
# numbers, these are cast to numerical types, so we want to change that
# here.
if not isinstance(name, six.string_types):
name = str(name)
# strip off leading #
name = name.strip()
if name.startswith('#'):
name = name[1:].strip()
result.add_component(c, name)
return result
def merge(self, *data, **kwargs):
"""
Merge two or more datasets into a single dataset.
This has the following effects:
All components from all datasets are added to the first argument
All datasets except the first argument are removed from the collection
Any component name conflicts are disambiguated
The pixel and world components apart from the first argument are discarded
:note: All arguments must have the same shape
:param data: One or more :class:`~glue.core.data.Data` instances.
:returns: self
"""
if len(data) < 2:
raise ValueError("merge requires 2 or more arguments")
shp = data[0].shape
for d in data:
if d.shape != shp:
raise ValueError("All arguments must have the same shape")
label = kwargs.get('label', data[0].label)
master = Data(label=label)
self.append(master)
master.coords = data[0].coords
for d in data:
skip = d.pixel_component_ids + d.world_component_ids
for c in d.components:
if c in skip:
continue
if c in master.components: # already present (via a link)
continue
taken = [_.label for _ in master.components]
lbl = c.label
# Special-case 'PRIMARY', rename to data label
if lbl == 'PRIMARY':
lbl = d.label
# First-pass disambiguation, try component_data
if lbl in taken:
lbl = '%s_%s' % (lbl, d.label)
lbl = disambiguate(lbl, taken)
c._label = lbl
master.add_component(d.get_component(c), c)
self.remove(d)
return self
def merge(self, *data, **kwargs):
"""
Merge two or more datasets into a single dataset.
This has the following effects:
All components from all datasets are added to the first argument
All datasets except the first argument are removed from the collection
Any component name conflicts are disambiguated
The pixel and world components apart from the first argument are discarded
:note: All arguments must have the same shape
:param data: One or more :class:`~glue.core.data.Data` instances.
:returns: self
"""
if len(data) < 2:
raise ValueError("merge requires 2 or more arguments")
shp = data[0].shape
for d in data:
if d.shape != shp:
raise ValueError("All arguments must have the same shape")
label = kwargs.get('label', data[0].label)
master = Data(label=label)
self.append(master)
master.coords = data[0].coords
for i, d in enumerate(data):
if isinstance(d.coords, WCSCoordinates):
master.coords = d.coords
break
# Find ambiguous components (ones which have labels in more than one
# dataset
from collections import Counter
clabel_count = Counter([c.label for d in data for c in d.visible_components])
for d in data:
for c in d.components:
if c in master.components: # already present (via a link)
continue
lbl = c.label
if clabel_count[lbl] > 1:
lbl = lbl + " [{0}]".format(d.label)
c._label = lbl
c.parent = master
master.add_component(d.get_component(c), c)
self.remove(d)
return self
def load_stacked_sequence(self, raster_data):
for window, window_data in raster_data.items():
w_data = Data(label=f"{window.replace(' ', '_')}")
w_data.coords = window_data.wcs
w_data.add_component(Component(window_data.data),
f"{window.replace(' ', '_')}")
w_data.style = VisualAttributes(color='#7A617C')
self.datasets.append(w_data)
def _ndarray_to_glue_data(arr, data_label):
if arr.ndim != 2:
raise ValueError(f'Imviz cannot load this array with ndim={arr.ndim}')
data = Data(label=data_label)
component = Component.autotyped(arr)
data.add_component(component=component, label='DATA')
yield data, data_label
def hdf5_reader(filename, format='auto', auto_merge=False, **kwargs):
"""
Read in all datasets from an HDF5 file
Parameters
----------
source: str or HDUList
The pathname to the FITS file.
If an HDUList is passed in, simply use that.
"""
import h5py
from astropy.table import Table
# Open file
file_handle = h5py.File(filename, 'r')
# Define function to read
# Read in all datasets
datasets = extract_hdf5_datasets(file_handle)
label_base = os.path.basename(filename).rpartition('.')[0]
if not label_base:
label_base = os.path.basename(filename)
data_by_shape = {}
groups = OrderedDict()
for key in datasets:
label = '{0}[{1}]'.format(label_base, key)
if datasets[key].dtype.kind in ('f', 'i'):
if auto_merge and datasets[key].value.shape in data_by_shape:
data = data_by_shape[datasets[key].value.shape]
else:
data = Data(label=label)
data_by_shape[datasets[key].value.shape] = data
groups[label] = data
data.add_component(component=datasets[key].value, label=key)
else:
table = Table.read(datasets[key], format='hdf5')
data = Data(label=label)
groups[label] = data
for column_name in table.columns:
column = table[column_name]
if column.ndim == 1:
component = Component(column, units=column.unit)
data.add_component(component=component, label=column_name)
else:
warnings.warn(
"HDF5: Ignoring vector column {0}".format(column_name))
# Close HDF5 file
file_handle.close()
return [groups[idx] for idx in groups]
def read_header(header):
out = []
for stream in header.stream_names:
result = Data(label="{stream}_{uid}".format(stream=stream,
uid=header.start['uid']))
tbl = header.table(stream, fill=True)
for col in tbl.columns:
result.add_component(tbl[col], str(col))
out.append(result)
return out
def merge(self, *data, **kwargs):
"""
Merge two or more datasets into a single dataset.
This has the following effects:
All components from all datasets are added to the first argument
All datasets except the first argument are removed from the collection
Any component name conflicts are disambiguated
The pixel and world components apart from the first argument are discarded
:note: All arguments must have the same shape
:param data: One or more :class:`~glue.core.data.Data` instances.
:returns: self
"""
if len(data) < 2:
raise ValueError("merge requires 2 or more arguments")
shp = data[0].shape
for d in data:
if d.shape != shp:
raise ValueError("All arguments must have the same shape")
label = kwargs.get('label', data[0].label)
master = Data(label=label)
self.append(master)
for d in data:
skip = d.pixel_component_ids + d.world_component_ids
for c in d.components:
if c in skip:
continue
if c in master.components: # already present (via a link)
continue
taken = [_.label for _ in master.components]
lbl = c.label
# Special-case 'PRIMARY', rename to data label
if lbl == 'PRIMARY':
lbl = d.label
# First-pass disambiguation, try component_data
if lbl in taken:
lbl = '%s_%s' % (lbl, d.label)
lbl = disambiguate(lbl, taken)
c._label = lbl
master.add_component(d.get_component(c), c)
self.remove(d)
return self
def load_sunpy_map(self, sunpy_map):
sunpy_map_loaded = sunpy.map.Map(sunpy_map)
label = 'sunpy-map-' + sunpy_map_loaded.name
data = Data(label=label)
data.coords = sunpy_map_loaded.wcs # preferred way, preserves more info in some cases
data.meta = sunpy_map_loaded.meta
data.add_component(Component(sunpy_map_loaded.data),
sunpy_map_loaded.name)
data.style = VisualAttributes(color='#FDB813',
preferred_cmap=sunpy_map.cmap)
self.datasets.append(data)
def load_sji(self, sji):
with fits.open(sji) as hdul:
hdul.verify("fix")
label = hdul[0].header['TDESC1'] + hdul[0].header['OBSID']
data = Data(label=label)
data.coords = WCSCoordinates(hdul[0].header)
data.meta = hdul[0].header
preferred_cmap_name = 'IRIS ' + hdul[0].header['TDESC1'].replace(
'_', ' ')
data.style = VisualAttributes(preferred_cmap=preferred_cmap_name)
data.add_component(Component(hdul[0].data), label)
self.datasets.append(data)
def make_test_data():
data = Data(label="Test Cat Data 1")
comp_x1 = Component(np.array([4, 5, 6, 3]))
comp_y1 = Component(np.array([1, 2, 3, 2]))
comp_z1 = Component(np.array([2, 3, 4, 1]))
data.add_component(comp_x1, 'x_gal')
data.add_component(comp_y1, 'y_gal')
data.add_component(comp_z1, 'z_gal')
return data
def make_test_data():
data = Data(label="Test Cat Data 1")
comp_x1 = Component(np.array([4, 5, 6, 3]))
comp_y1 = Component(np.array([1, 2, 3, 2]))
comp_z1 = Component(np.array([2, 3, 4, 1]))
data.add_component(comp_x1, 'x_gal')
data.add_component(comp_y1, 'y_gal')
data.add_component(comp_z1, 'z_gal')
return data
def load_sequence(self, raster_data):
for window, window_data in raster_data.items():
for i, scan_data in enumerate(window_data):
w_data = Data(
label=
f"{window.replace(' ', '_')}-{scan_data.meta['OBSID']}-scan-{i}"
)
w_data.coords = scan_data.wcs
w_data.add_component(Component(scan_data.data),
f"{window.replace(' ', '_')}-scan-{i}")
w_data.meta = scan_data.meta
w_data.style = VisualAttributes(color='#5A4FCF')
self.datasets.append(w_data)
def test_data():
data = Data(label="Test Data 1")
data2 = Data(label="Teset Data 2")
comp_a = Component(np.array([1, 2, 3]))
comp_b = Component(np.array([1, 2, 3]))
comp_c = Component(np.array([2, 4, 6]))
comp_d = Component(np.array([1, 3, 5]))
data.add_component(comp_a, 'a')
data.add_component(comp_b, 'b')
data2.add_component(comp_c, 'c')
data2.add_component(comp_d, 'd')
return data, data2
def _load_data(rec, context):
label = rec['label']
result = Data(label=label)
if 'coords' in rec:
result.coords = context.object(rec['coords'])
# we manually rebuild pixel/world components, so
# we override this function. This is pretty ugly
result._create_pixel_and_world_components = lambda ndim: None
comps = [
list(map(context.object, [cid, comp]))
for cid, comp in rec['components']
]
for icomp, (cid, comp) in enumerate(comps):
if isinstance(comp, CoordinateComponent):
comp._data = result
# For backward compatibility, we need to check for cases where
# the component ID for the pixel components was not a PixelComponentID
# and upgrade it to one. This can be removed once we no longer
# support pre-v0.8 session files.
if not comp.world and not isinstance(cid, PixelComponentID):
cid = PixelComponentID(comp.axis, cid.label, parent=cid.parent)
comps[icomp] = (cid, comp)
result.add_component(comp, cid)
assert result._world_component_ids == []
coord = [c for c in comps if isinstance(c[1], CoordinateComponent)]
coord = [x[0] for x in sorted(coord, key=lambda x: x[1])]
if getattr(result, 'coords') is not None:
assert len(coord) == result.ndim * 2
result._world_component_ids = coord[:len(coord) // 2]
result._pixel_component_ids = coord[len(coord) // 2:]
else:
assert len(coord) == result.ndim
result._pixel_component_ids = coord
# We can now re-generate the coordinate links
result._set_up_coordinate_component_links(result.ndim)
for s in rec['subsets']:
result.add_subset(context.object(s))
return result
def test_categorical_data():
data = Data(label="Test Cat Data 1")
data2 = Data(label="Teset Cat Data 2")
comp_x1 = CategoricalComponent(np.array(['a', 'a', 'b']))
comp_y1 = Component(np.array([1, 2, 3]))
comp_x2 = CategoricalComponent(np.array(['c', 'a', 'b']))
comp_y2 = Component(np.array([1, 3, 5]))
data.add_component(comp_x1, 'x1')
data.add_component(comp_y1, 'y1')
data2.add_component(comp_x2, 'x2')
data2.add_component(comp_y2, 'y2')
return data, data2
def test_high_cardinatility_timing(self):
card = 50000
data = Data()
card_data = [str(num) for num in range(card)]
data.add_component(Component(np.arange(card * 5)), 'y')
data.add_component(
CategoricalComponent(np.repeat([card_data], 5)), 'xcat')
self.add_data(data)
comp = data.find_component_id('xcat')
timer_func = partial(self.client._set_xydata, 'x', comp)
timer = timeit(timer_func, number=1)
assert timer < 3 # this is set for Travis speed
def hdf5_reader(filename, auto_merge=False, memmap=True, **kwargs):
"""
Read in all datasets from an HDF5 file
Parameters
----------
filename : str
The filename of the HDF5 file
memmap : bool, optional
Whether to use memory mapping
"""
from astropy.table import Table
# Read in all datasets
datasets = extract_hdf5_datasets(filename, memmap=memmap)
label_base = os.path.basename(filename).rpartition('.')[0]
if not label_base:
label_base = os.path.basename(filename)
data_by_shape = {}
groups = OrderedDict()
for key in datasets:
label = '{0}[{1}]'.format(label_base, key)
array = datasets[key]
if isinstance(array, Table):
data = Data(label=label)
groups[label] = data
for column_name in array.columns:
column = array[column_name]
if column.ndim == 1:
component = Component.autotyped(column, units=column.unit)
data.add_component(component=component,
label=column_name)
else:
warnings.warn("HDF5: Ignoring vector column {0}".format(column_name))
else:
if auto_merge and array.shape in data_by_shape:
data = data_by_shape[datasets[key].shape]
else:
data = Data(label=label)
data_by_shape[array.shape] = data
groups[label] = data
data.add_component(component=datasets[key], label=key[1:])
return [groups[idx] for idx in groups]
def test_high_cardinatility_timing(self):
card = 50000
data = Data()
card_data = [str(num) for num in range(card)]
data.add_component(Component(np.arange(card * 5)), 'y')
data.add_component(CategoricalComponent(np.repeat([card_data], 5)),
'xcat')
self.add_data(data)
comp = data.find_component_id('xcat')
timer_func = partial(self.client._set_xydata, 'x', comp)
timer = timeit(timer_func, number=1)
assert timer < 3 # this is set for Travis speed
def hdf5_reader(filename, auto_merge=False, memmap=True, **kwargs):
"""
Read in all datasets from an HDF5 file
Parameters
----------
filename : str
The filename of the HDF5 file
memmap : bool, optional
Whether to use memory mapping
"""
from astropy.table import Table
# Read in all datasets
datasets = extract_hdf5_datasets(filename, memmap=memmap)
label_base = os.path.basename(filename).rpartition('.')[0]
if not label_base:
label_base = os.path.basename(filename)
data_by_shape = {}
groups = OrderedDict()
for key in datasets:
label = '{0}[{1}]'.format(label_base, key)
array = datasets[key]
if isinstance(array, Table):
data = Data(label=label)
groups[label] = data
for column_name in array.columns:
column = array[column_name]
if column.ndim == 1:
component = Component.autotyped(column, units=column.unit)
data.add_component(component=component, label=column_name)
else:
warnings.warn(
"HDF5: Ignoring vector column {0}".format(column_name))
else:
if auto_merge and array.shape in data_by_shape:
data = data_by_shape[datasets[key].shape]
else:
data = Data(label=label)
data_by_shape[array.shape] = data
groups[label] = data
data.add_component(component=datasets[key], label=key[1:])
return [groups[idx] for idx in groups]
def _parse_iris_raster(data, label):
"""
Parse IRIS Level 2 raster files so that it can be loaded by glue.
"""
w_dataset = []
for window, window_data in data.items():
for i, scan_data in enumerate(window_data):
w_data = Data(label=f"{window.replace(' ', '_')}-{scan_data.meta['OBSID']}-scan-{i}")
w_data.coords = WCSCoordinates(scan_data.wcs.to_header())
w_data.add_component(Component(scan_data.data), f"{window.replace(' ', '_')}-{scan_data.meta['OBSID']}-scan-{i}")
w_data.meta = scan_data.meta
w_data.style = VisualAttributes(color='#5A4FCF')
w_dataset.append(w_data)
return w_dataset
def newView(self,
type="scatter",
components=[],
title="New View",
**kwargs):
only_subsets = kwargs.get('only_subsets', False)
only_view = kwargs.get('only_view', False)
if (self.parent is not None):
kwargs.setdefault('modal', self.parent.modal)
if (self.debug is not None):
kwargs.setdefault('debug', self.debug)
gp = None
mode = "tab-after"
if (len(self.active_views.values()) == 0):
mode = "split-bottom"
kwargs.setdefault('mode', mode)
if only_view is False:
gp = self.factory.createGluePlot(type, self.data, components,
title, **kwargs)
else:
data = Data(label=self.data.label)
for c in components:
data.add_component(self.data[c, self.selection], label=c)
data.get_component(c).color = self.data.get_component(c).color
if (data.size > 0):
gp = self.factory.createGluePlot(type, data, components, title,
**kwargs)
if gp is not None:
if only_view is False:
gp.setParent(self)
key = id(gp.window)
self.active_views[key] = gp
self.views[key] = {
'type': type,
'components': components,
'title': title,
'kwargs': kwargs
}
if isinstance(gp.window, Floatview):
gp.window.observe(
lambda changes: GlueManager.removeViewIfDisposed(
self, gp.window), 'uid')
self.parent.updateHistory()
return gp
def _hdu2data(hdu, data_label, hdulist, include_wcs=True):
if 'BUNIT' in hdu.header and _validate_bunit(hdu.header['BUNIT'], raise_error=False):
bunit = hdu.header['BUNIT']
else:
bunit = ''
comp_label = f'{hdu.name.upper()},{hdu.ver}'
new_data_label = f'{data_label}[{comp_label}]'
data = Data(label=new_data_label)
if include_wcs:
data.coords = WCS(hdu.header, hdulist)
component = Component.autotyped(hdu.data, units=bunit)
data.add_component(component=component, label=comp_label)
return data, new_data_label
def casalike_cube(filename, **kwargs):
"""
This provides special support for 4D CASA FITS - like cubes,
which have 2 spatial axes, a spectral axis, and a stokes axis
in that order.
Each stokes cube is split out as a separate component
"""
from astropy.io import fits
result = Data()
with fits.open(filename, **kwargs) as hdulist:
array = hdulist[0].data
header = hdulist[0].header
result.coords = coordinates_from_header(header)
for i in range(array.shape[0]):
result.add_component(array[[i]], label='STOKES %i' % i)
return result
def casalike_cube(filename, **kwargs):
"""
This provides special support for 4D CASA FITS - like cubes,
which have 2 spatial axes, a spectral axis, and a stokes axis
in that order.
Each stokes cube is split out as a separate component
"""
from astropy.io import fits
result = Data()
with fits.open(filename, **kwargs) as hdulist:
array = hdulist[0].data
header = hdulist[0].header
result.coords = coordinates_from_header(header)
for i in range(array.shape[0]):
result.add_component(array[[i]], label='STOKES %i' % i)
return result
def _load_data(rec, context):
label = rec['label']
result = Data(label=label)
result.coords = context.object(rec['coords'])
# we manually rebuild pixel/world components, so
# we override this function. This is pretty ugly
result._create_pixel_and_world_components = lambda ndim: None
comps = [list(map(context.object, [cid, comp]))
for cid, comp in rec['components']]
for icomp, (cid, comp) in enumerate(comps):
if isinstance(comp, CoordinateComponent):
comp._data = result
# For backward compatibility, we need to check for cases where
# the component ID for the pixel components was not a PixelComponentID
# and upgrade it to one. This can be removed once we no longer
# support pre-v0.8 session files.
if not comp.world and not isinstance(cid, PixelComponentID):
cid = PixelComponentID(comp.axis, cid.label, parent=cid.parent)
comps[icomp] = (cid, comp)
result.add_component(comp, cid)
assert result._world_component_ids == []
coord = [c for c in comps if isinstance(c[1], CoordinateComponent)]
coord = [x[0] for x in sorted(coord, key=lambda x: x[1])]
assert len(coord) == result.ndim * 2
result._world_component_ids = coord[:len(coord) // 2]
result._pixel_component_ids = coord[len(coord) // 2:]
# We can now re-generate the coordinate links
result._set_up_coordinate_component_links(result.ndim)
for s in rec['subsets']:
result.add_subset(context.object(s))
return result
def _hdu2data(hdu, data_label, hdulist, include_wcs=True):
if 'BUNIT' in hdu.header:
bunit = _validate_bunit(hdu.header['BUNIT'], raise_error=False)
else:
bunit = ''
comp_label = f'{hdu.name.upper()},{hdu.ver}'
new_data_label = f'{data_label}[{comp_label}]'
data = Data(label=new_data_label)
if hdulist is not None and hdu.name != 'PRIMARY' and 'PRIMARY' in hdulist:
data.meta.update(dict(hdulist['PRIMARY'].header))
data.meta.update(dict(hdu.header))
if include_wcs:
data.coords = WCS(hdu.header, hdulist)
component = Component.autotyped(hdu.data, units=bunit)
data.add_component(component=component, label=comp_label)
return data, new_data_label
def pandas_to_glue(df, label='data', name_map=default_name_map):
"""Convert dataframe to glue.core.data.Data. Glue categorical variables require hashing,
store array of unhashable components in ComponentID._unhashable. Override column names
in name_map with dictionary values.
"""
data = Data(label=label)
for c in df.columns:
c_name = map_column_names(c)
try:
data.add_component(df[c], c_name)
except TypeError:
# pd.factorize error with int list input to CategoricalComponent
r = ['%09d' % i for i in range(len(df[c]))]
cc = CategoricalComponent(r)
c_id = ComponentID(c_name)
c_id._unhashable = np.array(df[c])
data.add_component(cc, c_id)
return data
def _jwst2data(file_obj, ext, data_label):
comp_label = ext.upper()
new_data_label = f'{data_label}[{comp_label}]'
data = Data(label=new_data_label)
unit_attr = f'bunit_{ext}'
try:
# This is very specific to JWST pipeline image output.
with AsdfInFits.open(file_obj) as af:
dm = af.tree
dm_meta = af.tree["meta"]
if (unit_attr in dm_meta and
_validate_bunit(dm_meta[unit_attr], raise_error=False)):
bunit = dm_meta[unit_attr]
else:
bunit = ''
# This is instance of gwcs.WCS, not astropy.wcs.WCS
if 'wcs' in dm_meta:
data.coords = dm_meta['wcs']
imdata = dm[ext]
component = Component.autotyped(imdata, units=bunit)
# Might have bad GWCS. If so, we exclude it.
try:
data.add_component(component=component, label=comp_label)
except Exception: # pragma: no cover
data.coords = None
data.add_component(component=component, label=comp_label)
# TODO: Do not need this when jwst.datamodels finally its own package.
# This might happen for grism image; fall back to FITS loader without WCS.
except Exception:
if ext == 'data':
ext = 'sci'
hdu = file_obj[ext]
return _hdu2data(hdu, data_label, file_obj, include_wcs=False)
return data, new_data_label
def test_indexed(self):
# Here we slice two of the dimensions and then compare the results to a
# manually sliced dataset.
derived = IndexedData(self.data, (None, 2, None, 4, None))
manual = Data()
manual.add_component(self.data[self.x_id][:, 2, :, 4, :], label=self.x_id)
manual.add_component(self.data[self.y_id][:, 2, :, 4, :], label=self.y_id)
assert derived.label == 'Test data[:,2,:,4,:]'
assert derived.shape == manual.shape
assert [str(c) for c in derived.main_components] == [str(c) for c in manual.main_components]
assert derived.get_kind(self.x_id) == manual.get_kind(self.x_id)
for view in [None, (1, slice(None), slice(1, 4))]:
assert_equal(derived.get_data(self.x_id, view=view),
manual.get_data(self.x_id, view=view))
assert_equal(derived.get_mask(self.subset_state, view=view),
manual.get_mask(self.subset_state, view=view))
bounds = [2, (-5, 5, 10), (-3, 3, 10)]
assert_equal(derived.compute_fixed_resolution_buffer(bounds=bounds, target_cid=self.x_id),
manual.compute_fixed_resolution_buffer(bounds=bounds, target_cid=self.x_id))
assert_equal(derived.compute_statistic('mean', self.x_id),
manual.compute_statistic('mean', self.x_id))
assert_equal(derived.compute_statistic('mean', self.x_id, axis=2),
manual.compute_statistic('mean', self.x_id, axis=2))
assert_equal(derived.compute_statistic('mean', self.x_id, subset_state=self.subset_state),
manual.compute_statistic('mean', self.x_id, subset_state=self.subset_state))
assert_equal(derived.compute_histogram([self.x_id], range=[(0, 1000)], bins=[30]),
manual.compute_histogram([self.x_id], range=[(0, 1000)], bins=[30]))
assert_equal(derived.compute_histogram([self.x_id], range=[(0, 1000)], bins=[30], subset_state=self.subset_state),
manual.compute_histogram([self.x_id], range=[(0, 1000)], bins=[30], subset_state=self.subset_state))
def test_limits_inf(self):
d = Data()
x = Component(np.array([[1, 2], [np.infty, 4]]))
y = Component(np.array([[2, 4], [-np.infty, 8]]))
xid = d.add_component(x, 'x')
yid = d.add_component(y, 'y')
self.collect.append(d)
self.client.add_layer(d)
self.client.xatt = xid
self.client.yatt = yid
assert self.client._visible_limits(0) == (1, 4)
assert self.client._visible_limits(1) == (2, 8)
def test_limits_inf(self):
d = Data()
x = Component(np.array([[1, 2], [np.infty, 4]]))
y = Component(np.array([[2, 4], [-np.infty, 8]]))
xid = d.add_component(x, 'x')
yid = d.add_component(y, 'y')
self.collect.append(d)
self.client.add_layer(d)
self.client.xatt = xid
self.client.yatt = yid
assert self.client._visible_limits(0) == (1, 4)
assert self.client._visible_limits(1) == (2, 8)
def npy_reader(filename, format='auto', auto_merge=False, **kwargs):
"""
Read in a Numpy structured array saved to a .npy or .npz file.
Parameters
----------
source: str
The pathname to the Numpy save file.
"""
import numpy as np
npy_data = np.load(filename)
if not hasattr(npy_data.dtype, 'names'):
raise ValueError("Numpy save file loading currently only supports structured"
" arrays, e.g., with specified names.")
d = Data()
for name in npy_data.dtype.names:
comp = Component(npy_data[name])
d.add_component(comp, label=name)
return d
def gridded_data(filename, format='auto', **kwargs):
result = Data()
# Try and automatically find the format if not specified
if format == 'auto':
format = file_format(filename)
# Read in the data
if is_fits(filename):
from astropy.io import fits
arrays = extract_data_fits(filename, **kwargs)
header = fits.getheader(filename)
result.coords = coordinates_from_header(header)
elif is_hdf5(filename):
arrays = extract_data_hdf5(filename, **kwargs)
else:
raise Exception("Unkonwn format: %s" % format)
for component_name in arrays:
comp = Component.autotyped(arrays[component_name])
result.add_component(comp, component_name)
return result
def img_data(file_name):
"""Load common image files into a Glue data object"""
result = Data()
data = img_loader(file_name)
data = np.flipud(data)
shp = data.shape
comps = []
labels = []
# split 3 color images into each color plane
if len(shp) == 3 and shp[2] in [3, 4]:
comps.extend([data[:, :, 0], data[:, :, 1], data[:, :, 2]])
labels.extend(['red', 'green', 'blue'])
if shp[2] == 4:
comps.append(data[:, :, 3])
labels.append('alpha')
else:
comps = [data]
labels = ['PRIMARY']
# look for AVM coordinate metadata
try:
from pyavm import AVM
avm = AVM(str(file_name)) # avoid unicode
wcs = avm.to_wcs()
except:
pass
else:
result.coords = coordinates_from_wcs(wcs)
for c, l in zip(comps, labels):
result.add_component(c, l)
return result
def casalike_cube(filename, **kwargs):
"""
This provides special support for 4D CASA FITS - like cubes,
which have 2 spatial axes, a spectral axis, and a stokes axis
in that order.
Each stokes cube is split out as a separate component
"""
from astropy.io import fits
result = Data()
if 'ignore_missing_end' not in kwargs:
kwargs['ignore_missing_end'] = True
with fits.open(filename, **kwargs) as hdulist:
array = hdulist[0].data
header = hdulist[0].header
result.coords = coordinates_from_header(header)
for i in range(array.shape[0]):
units = header.get('BUNIT')
component = Component.autotyped(array[[i]], units=units)
result.add_component(component, label='STOKES %i' % i)
return result
def test_ticks_go_back_after_changing(self):
""" If you change to a categorical axis and then change back
to a numeric, the axis ticks should fix themselves properly.
"""
data = Data()
data.add_component(Component(np.arange(100)), 'y')
data.add_component(
CategoricalComponent(['a'] * 50 + ['b'] * 50), 'xcat')
data.add_component(Component(2 * np.arange(100)), 'xcont')
self.add_data(data=data)
self.client.yatt = data.find_component_id('y')
self.client.xatt = data.find_component_id('xcat')
self.check_ticks(self.client.axes.xaxis, False, True)
self.check_ticks(self.client.axes.yaxis, False, False)
self.client.xatt = data.find_component_id('xcont')
self.check_ticks(self.client.axes.yaxis, False, False)
self.check_ticks(self.client.axes.xaxis, False, False)
def hdf5_reader(filename, format='auto', auto_merge=False, **kwargs):
"""
Read in all datasets from an HDF5 file
Parameters
----------
source: str or HDUList
The pathname to the FITS file.
If an HDUList is passed in, simply use that.
"""
import h5py
from astropy.table import Table
# Open file
file_handle = h5py.File(filename, 'r')
# Define function to read
# Read in all datasets
datasets = extract_hdf5_datasets(file_handle)
label_base = os.path.basename(filename).rpartition('.')[0]
if not label_base:
label_base = os.path.basename(filename)
data_by_shape = {}
groups = OrderedDict()
for key in datasets:
label = '{0}[{1}]'.format(
label_base,
key
)
if datasets[key].dtype.kind in ('f', 'i'):
if auto_merge and datasets[key].value.shape in data_by_shape:
data = data_by_shape[datasets[key].value.shape]
else:
data = Data(label=label)
data_by_shape[datasets[key].value.shape] = data
groups[label] = data
data.add_component(component=datasets[key].value, label=key)
else:
table = Table.read(datasets[key], format='hdf5')
data = Data(label=label)
groups[label] = data
for column_name in table.columns:
column = table[column_name]
if column.ndim == 1:
component = Component(column, units=column.unit)
data.add_component(component=component,
label=column_name)
else:
warnings.warn("HDF5: Ignoring vector column {0}".format(column_name))
# Close HDF5 file
file_handle.close()
return [groups[idx] for idx in groups]
def _parse_data_dataframe(data, label):
label = label or 'Data'
result = Data(label=label)
for c in data.columns:
result.add_component(data[c], str(c))
return [result]
def fits_reader(source, auto_merge=False, exclude_exts=None, label=None):
"""
Read in all extensions from a FITS file.
Parameters
----------
source: str or HDUList
The pathname to the FITS file.
If an HDUList is passed in, simply use that.
auto_merge: bool
Merge extensions that have the same shape
and only one has a defined WCS.
exclude_exts: [hdu, ] or [index, ]
List of HDU's to exclude from reading.
This can be a list of HDU's or a list
of HDU indexes.
"""
from astropy.io import fits
from astropy.table import Table
exclude_exts = exclude_exts or []
if isinstance(source, fits.hdu.hdulist.HDUList):
hdulist = source
close_hdulist = False
else:
hdulist = fits.open(source, ignore_missing_end=True)
hdulist.verify('fix')
close_hdulist = True
groups = OrderedDict()
extension_by_shape = OrderedDict()
if label is not None:
label_base = label
else:
hdulist_name = hdulist.filename()
if hdulist_name is None:
hdulist_name = "HDUList"
label_base = basename(hdulist_name).rpartition('.')[0]
if not label_base:
label_base = basename(hdulist_name)
# Create a new image Data.
def new_data(suffix=True):
if suffix:
label = '{0}[{1}]'.format(label_base, hdu_name)
else:
label = label_base
data = Data(label=label)
data.coords = coords
# We need to be careful here because some header values are special
# objects that we should convert to strings
for key, value in hdu.header.items():
if (key == 'COMMENT' or key == 'HISTORY'):
if key not in data.meta:
data.meta[key] = [str(value)]
else:
data.meta[key].append(str(value))
elif isinstance(value, string_types) or isinstance(value, (int, float, bool)):
data.meta[key] = value
else:
data.meta[key] = str(value)
groups[hdu_name] = data
extension_by_shape[shape] = hdu_name
return data
for extnum, hdu in enumerate(hdulist):
hdu_name = hdu.name if hdu.name else "HDU{0}".format(extnum)
if (hdu.data is not None and
hdu.data.size > 0 and
hdu_name not in exclude_exts and
extnum not in exclude_exts):
if is_image_hdu(hdu):
shape = hdu.data.shape
coords = coordinates_from_header(hdu.header)
units = hdu.header.get('BUNIT')
if not auto_merge or has_wcs(coords):
data = new_data(suffix=len(hdulist) > 1)
else:
try:
data = groups[extension_by_shape[shape]]
except KeyError:
data = new_data(suffix=len(hdulist) > 1)
component = Component.autotyped(hdu.data, units=units)
data.add_component(component=component,
label=hdu_name)
elif is_table_hdu(hdu):
# Loop through columns and make component list
table = Table.read(hdu, format='fits')
label = '{0}[{1}]'.format(label_base, hdu_name)
data = Data(label=label)
groups[hdu_name] = data
for column_name in table.columns:
column = table[column_name]
if column.ndim != 1:
warnings.warn("Dropping column '{0}' since it is not 1-dimensional".format(column_name))
continue
component = Component.autotyped(column, units=column.unit)
data.add_component(component=component,
label=column_name)
if close_hdulist:
hdulist.close()
return [groups[idx] for idx in groups]
def fits_reader(source, auto_merge=False, exclude_exts=None, label=None):
"""
Read in all extensions from a FITS file.
Parameters
----------
source: str or HDUList
The pathname to the FITS file.
If an HDUList is passed in, simply use that.
auto_merge: bool
Merge extensions that have the same shape
and only one has a defined WCS.
exclude_exts: [hdu, ] or [index, ]
List of HDU's to exclude from reading.
This can be a list of HDU's or a list
of HDU indexes.
"""
from astropy.io import fits
from astropy.table import Table
exclude_exts = exclude_exts or []
if not isinstance(source, fits.hdu.hdulist.HDUList):
hdulist = fits.open(source, ignore_missing_end=True)
hdulist.verify('fix')
else:
hdulist = source
groups = OrderedDict()
extension_by_shape = OrderedDict()
if label is not None:
label_base = label
else:
hdulist_name = hdulist.filename()
if hdulist_name is None:
hdulist_name = "HDUList"
label_base = basename(hdulist_name).rpartition('.')[0]
if not label_base:
label_base = basename(hdulist_name)
# Create a new image Data.
def new_data():
label = '{0}[{1}]'.format(
label_base,
hdu_name
)
data = Data(label=label)
data.coords = coords
groups[hdu_name] = data
extension_by_shape[shape] = hdu_name
return data
for extnum, hdu in enumerate(hdulist):
hdu_name = hdu.name if hdu.name else "HDU{0}".format(extnum)
if (hdu.data is not None and
hdu.data.size > 0 and
hdu_name not in exclude_exts and
extnum not in exclude_exts):
if is_image_hdu(hdu):
shape = hdu.data.shape
coords = coordinates_from_header(hdu.header)
if not auto_merge or has_wcs(coords):
data = new_data()
else:
try:
data = groups[extension_by_shape[shape]]
except KeyError:
data = new_data()
data.add_component(component=hdu.data,
label=hdu_name)
elif is_table_hdu(hdu):
# Loop through columns and make component list
table = Table.read(hdu, format='fits')
label = '{0}[{1}]'.format(
label_base,
hdu_name
)
data = Data(label=label)
groups[hdu_name] = data
for column_name in table.columns:
column = table[column_name]
if column.ndim != 1:
warnings.warn("Dropping column '{0}' since it is not 1-dimensional".format(column_name))
continue
component = Component.autotyped(column, units=column.unit)
data.add_component(component=component,
label=column_name)
return [groups[idx] for idx in groups]
class TestCategoricalHistogram(TestHistogramClient):
def setup_method(self, method):
self.data = Data(y=[-1, -1, -1, -2, -2, -2, -3, -5, -7])
self.data.add_component(CategoricalComponent(['a', 'a', 'a', 'b', 'c', 'd', 'd', 'e', 'f']), 'x')
self.subset = self.data.new_subset()
self.collect = DataCollection(self.data)
self.client = HistogramClient(self.collect, FIGURE)
self.axes = self.client.axes
FIGURE.canvas.draw = MagicMock()
assert FIGURE.canvas.draw.call_count == 0
def test_xlimit_single_set(self):
self.client.add_layer(self.data)
self.client.set_component(self.data.id['x'])
self.client.xlimits = (None, 5)
assert self.client.xlimits == (-0.5, 5)
self.client.xlimits = (3, None)
assert self.client.xlimits == (3, 5)
def test_default_xlimits(self):
self.client.add_layer(self.data)
self.client.set_component(self.data.id['x'])
assert self.client.xlimits == (-0.5, 5.5)
self.client.set_component(self.data.id['y'])
assert self.client.xlimits == (-7, -1)
def test_change_default_bins(self):
self.client.add_layer(self.data)
self.client.set_component(self.data.id['x'])
assert self.client.nbins == 6
def test_tick_labels(self):
self.client.add_layer(self.data)
self.client.set_component(self.data.id['x'])
correct_labels = ['a', 'b', 'c', 'd', 'e', 'f']
formatter = self.client.axes.xaxis.get_major_formatter()
xlabels = [formatter.format_data(pos) for pos in range(6)]
assert correct_labels == xlabels
def test_apply_roi(self):
self.client.add_layer(self.data)
self.client.set_component(self.data.id['x'])
# bins are 1...4
self.data.edit_subset = [self.data.subsets[0]]
roi = MagicMock()
roi.to_polygon.return_value = [1.2, 2, 4], [2, 3, 4]
self.client.apply_roi(roi)
state = self.data.subsets[0].subset_state
assert isinstance(state, CategoricalROISubsetState)
np.testing.assert_equal(self.data.subsets[0].subset_state.roi.categories,
np.array(['b', 'c', 'd', 'e']))
# REMOVED TESTS
def test_xlog_axes_labels(self):
""" log-scale doesn't make sense for categorical data"""
pass
def test_xlog_snaps_limits(self):
""" log-scale doesn't make sense for categorical data"""
pass
def test_apply_roi_xlog(self):
""" log-scale doesn't make sense for categorical data"""
pass
def test_nbin_override_persists_over_attribute_change(self):
# regression test for #398
self.collect.append(self.data)
self.client.add_layer(self.data)
self.client.set_component(self.data.id['x'])
self.client.nbins = 7
self.client.set_component(self.data.id['y'])
assert self.client.nbins == 7
def test_cube():
data = Data(label="Test Cube")
comp_a = Component(np.ones((16, 16, 16)))
data.add_component(comp_a, 'test_3')
return data
