def generate_destination(self, *args, **kwargs):
'''Correctly set up a destination data format
depending on the supplied input
Parameters
----------
args, kwargs
'''
if len(args) == 1 and len(kwargs) == 0:
dest = args[0]
if isinstance(dest, dm.GridData):
grid = dest.grid
grid.initialize(self.source)
return dm.GridData(grid)
if isinstance(dest, dm.Grid):
grid = dest
grid.initialize(self.source)
return dm.GridData(grid)
if isinstance(dest, dm.GridArray):
grid = dest.grid
grid.initialize(self.source)
return dm.GridArray(np.empty(grid.shape), grid=grid)
if isinstance(dest, dm.PointData):
# check which vars we need:
if self.source_has_grid:
return dest[self.source.grid.vars]
else:
return dest
# check if source has a grid and if any args are in there
if isinstance(self.source, (dm.GridData, dm.GridArray)):
dims = []
for arg in args:
# in this case the source may have a grid, get those edges
if isinstance(arg, str):
if arg in self.source.grid.vars:
dims.append(self.source.grid[arg])
continue
dims.append(arg)
args = dims
# instantiate
grid = dm.Grid(*args, **kwargs)
grid.initialize(self.source)
if isinstance(self.source, dm.GridArray):
return dm.GridArray(np.empty(grid.shape), grid=grid)
return dm.GridData(grid)
def std_devs(self, sigmas=[
1.,
], fill_value=np.nan, **kwargs):
""" Mean + Sigma values
sigmas : iterable
values to produce bands = mean =/- sigmas * std
"""
mean = self.mean(fill_value=np.nan, **kwargs)
std = self.std(fill_value=np.nan, **kwargs)
output_maps = {}
for var in self.data.vars:
m = np.ma.asarray(mean[var])
s = np.ma.asarray(std[var])
arrays = [m]
for sigma in sigmas:
delta = sigma * s
arrays.append(m + delta)
arrays.insert(0, m - delta)
output_maps[var] = np.swapaxes(np.stack(arrays), 0, -1)
# Pack into GridArray
if self.single:
out_data = dm.GridArray(output_maps['test'], grid=self.grid)
# Pack into GridData
else:
out_data = dm.GridData(self.grid)
for var, output_map in output_maps.items():
out_data[var] = output_map
return out_data
def T(self):
'''transpose'''
new_obj = dm.GridData()
new_obj._grid = self._grid.T
for n, d in self.items():
new_obj[n] = d.T
return new_obj
def __getitem__(self, item):
if isinstance(item, str):
if item in self.vars:
if item in self.data_vars:
data = self._data[item]
if callable(data):
self[item] = data()
data = self._data[item]
else:
data = self.get_array(item)
new_data = dm.GridArray(data, grid=self._grid)
return new_data
else:
raise IndexError('No variable %s in DataSet' % item)
# mask
if isinstance(item, dm.GridArray):
if item.dtype == np.bool:
# in this case it is a mask
# ToDo: masked operations behave strangely, operations are applyed to all elements, even if masked
new_data = dm.GridData(self._grid)
for v in self.data_vars:
new_data[v] = self[v][item]
return new_data
raise NotImplementedError('get item %s' % item)
# create new instance with only those vars
if isinstance(item, Iterable) and all(
[isinstance(v, str) for v in item]):
new_data = dm.GridData(self._grid)
for v in item:
if v in self.data_vars:
new_data[v] = self[v]
return new_data
# slice
new_grid = self._grid[item]
if len(new_grid) == 0:
return {n: d[item] for n, d in self.items()}
new_data = dm.GridData(new_grid)
for n, d in self.items():
new_data[n] = d[item]
return new_data
def run_np_indexed(self, method, fill_value=np.nan, **kwargs):
'''run the numpy indexed methods
Parameters:
-----------
method : str
choice of ['sum', 'mean', 'min', 'max', 'std', 'var', 'argmin', 'argmax', 'median', 'mode', 'prod']
'''
axis = kwargs.pop('axis', None)
assert axis is None, 'Axis kw not supported for BinnedData'
if self.indices is None:
self.compute_indices()
outputs = {}
output_maps = {}
for var in self.data.vars:
source_data = self.data[var]
if source_data.ndim > 1:
output_maps[var] = np.full(self.grid.shape +
source_data.shape[1:],
fill_value=fill_value,
dtype=self.data[var].dtype)
else:
output_maps[var] = np.full(self.grid.shape,
fill_value=fill_value,
dtype=self.data[var].dtype)
source_data = self.data[var]
indices, outputs[var] = self.group.__getattribute__(method)(
source_data)
for i, idx in enumerate(indices):
if idx < 0:
continue
out_idx = np.unravel_index(idx, self.grid.shape)
for var in self.data.vars:
output_maps[var][out_idx] = result = outputs[var][i]
# Pack into GridArray
if self.single:
out_data = dm.GridArray(output_maps['test'], grid=self.grid)
else:
# Pack into GridData
out_data = dm.GridData(self.grid)
for var, output_map in output_maps.items():
out_data[var] = output_map
return out_data
def __array_ufunc__(self, ufunc, method, *inputs, **kwargs):
'''interface to numpy unversal functions'''
scalar_results = OrderedDict()
array_result = dm.GridData()
for var in inputs[0].data_vars:
converted_inputs = [inp[var] for inp in inputs]
result = converted_inputs[0].__array_ufunc__(
ufunc, method, *converted_inputs, **kwargs)
if isinstance(result, dm.GridArray):
array_result[var] = result
else:
scalar_results[var] = result
if len(array_result.data_vars) == 0:
return scalar_results
if len(scalar_results) == 0:
return array_result
return scalar_results, array_result
def apply_function(self,
function,
*args,
fill_value=np.nan,
return_len=None,
**kwargs):
'''apply function per bin
function : callable
return_len : int (optional)
the shape (array length) of the `function` output. If None is passed,
`return_len` is (tried to be) determined on the fly
if weights are passed as kwarg, they are interpreted as being a variable in the dataset
'''
if self.indices is None:
self.compute_indices()
outputs = {}
output_maps = {}
weights = kwargs.pop('weights', None)
for var in self.data.vars:
source_data = self.data[var]
if return_len is None:
# try to figure out return length of function
if source_data.ndim > 1:
if weights is not None:
kwargs['weights'] = self.data[weights][:3, [0] * (
source_data.ndim - 1)]
test_value = function(
source_data[:3, [0] * (source_data.ndim - 1)], *args,
**kwargs)
else:
if weights is not None:
kwargs['weights'] = self.data[weights][:3]
test_value = function(source_data[:3], *args, **kwargs)
if np.isscalar(test_value):
return_len = 1
else:
return_len = len(test_value)
if source_data.ndim > 1:
if return_len > 1:
output_maps[var] = np.full(self.grid.shape +
source_data.shape[1:] +
(return_len, ),
fill_value=fill_value,
dtype=source_data.dtype)
else:
output_maps[var] = np.full(self.grid.shape +
source_data.shape[1:],
fill_value=fill_value,
dtype=source_data.dtype)
else:
if return_len > 1:
output_maps[var] = np.full(self.grid.shape +
(return_len, ),
fill_value=fill_value,
dtype=source_data.dtype)
else:
output_maps[var] = np.full(self.grid.shape,
fill_value=fill_value,
dtype=source_data.dtype)
for i in range(self.grid.size):
mask = self.indices == i
if np.any(mask):
out_idx = np.unravel_index(i, self.grid.shape)
for var in self.data.vars:
source_data = self.data[var]
if source_data.ndim > 1:
for idx in np.ndindex(*source_data.shape[1:]):
if weights is not None:
kwargs['weights'] = self.data[weights][:, idx][
mask]
output_maps[var][out_idx + (idx, )] = function(
source_data[:, idx][mask], *args, **kwargs)
else:
if weights is not None:
kwargs['weights'] = self.data[weights][mask]
output_maps[var][out_idx] = function(
source_data[mask], *args, **kwargs)
# Pack into GridArray
if self.single:
out_data = dm.GridArray(output_maps['test'], grid=self.grid)
# Pack into GridData
else:
out_data = dm.GridData(self.grid)
for var, output_map in output_maps.items():
out_data[var] = output_map
return out_data
def apply_function(self,
function,
*args,
fill_value=np.nan,
return_len=None,
**kwargs):
'''apply function per bin'''
if self.indices is None:
self.compute_indices()
outputs = {}
output_maps = {}
weights = kwargs.pop('weights', None)
for var in self.data.vars:
source_data = self.data[var]
if return_len is None:
# try to figure out return length of function
if source_data.ndim > 1:
test_value = function(
source_data[:3, [0] * (source_data.ndim - 1)], *args,
**kwargs)
else:
test_value = function(source_data[:3], *args, **kwargs)
if np.isscalar(test_value):
return_len = 1
else:
return_len = len(test_value)
if source_data.ndim > 1:
if return_len > 1:
output_maps[var] = np.full(self.grid.shape +
source_data.shape[1:] +
(return_len, ),
fill_value=fill_value,
dtype=source_data.dtype)
else:
output_maps[var] = np.full(self.grid.shape +
source_data.shape[1:],
fill_value=fill_value,
dtype=source_data.dtype)
else:
if return_len > 1:
output_maps[var] = np.full(self.grid.shape +
(return_len, ),
fill_value=fill_value,
dtype=source_data.dtype)
else:
output_maps[var] = np.full(self.grid.shape,
fill_value=fill_value,
dtype=source_data.dtype)
for i in range(self.grid.size):
mask = self.indices == i
if np.any(mask):
out_idx = np.unravel_index(i, self.grid.shape)
for var in self.data.vars:
source_data = self.data[var]
if source_data.ndim > 1:
for idx in np.ndindex(*source_data.shape[1:]):
if weights is not None:
kwargs['weights'] = self.data[weights][:, idx][
mask]
output_maps[var][out_idx + (idx, )] = function(
source_data[:, idx][mask], *args, **kwargs)
else:
if weights is not None:
kwargs['weights'] = self.data[weights][mask]
output_maps[var][out_idx] = function(
source_data[mask], *args, **kwargs)
# Pack into GridArray
if self.single:
out_data = dm.GridArray(output_maps['test'], grid=self.grid)
# Pack into GridData
else:
out_data = dm.GridData(self.grid)
for var, output_map in output_maps.items():
out_data[var] = output_map
return out_data