def __setitem__(self, idx, value):
try:
if idx.dtype == numpy.dtype('bool') or (idx[0].dtype == 'bool' and len(idx) == 1):
# Special case for boolean setitem
self_flat = self.flat
idx = afnumpy.where(idx.flat)
self_flat[idx] = value
return
except AttributeError:
pass
except RuntimeError:
# idx is all False
return
idx, idx_shape = indexing.__convert_dim__(self.shape, idx)
if any(x is None for x in idx):
# one of the indices is empty
return
idx = tuple(idx)
if len(idx) == 0:
idx = tuple([0])
if(isinstance(value, ndarray)):
if(value.dtype != self.dtype):
raise TypeError('left hand side must have same dtype as right hand side')
value = indexing.__expand_dim__(self.shape, value, idx).d_array
elif(isinstance(value, numbers.Number)):
pass
else:
raise NotImplementedError('values must be a afnumpy.ndarray')
self.d_array[idx] = value
# This is a hack to be able to make it look like arrays with stride[0] > 1 can still be views
# In practise right now it only applies to ndarray.real and ndarray.imag
try:
self._base[self._base_index] = self
except AttributeError:
pass
def __getitem__(self, args):
if not isinstance(args, tuple):
args = (args,)
if len(args) == 1 and isinstance(args[0], afnumpy.ndarray) and args[0].dtype == numpy.dtype('bool'):
# Special case for boolean getitem
return self.flat[afnumpy.where(args[0].flat)]
idx, new_shape = indexing.__convert_dim__(self.shape, args)
if any(x is None for x in idx):
# one of the indices is empty
return ndarray(indexing.__index_shape__(self.shape, idx), dtype=self.dtype)
idx = tuple(idx)
if len(idx) == 0:
idx = tuple([0])
s = self.d_array[idx]
shape = pu.af_shape(s)
array = ndarray(shape, dtype=self.dtype, af_array=s)
if(shape != new_shape):
array = array.reshape(new_shape)
if new_shape == () and Ellipsis not in args:
# Return the actual scalar
return numpy.array(array)[()]
return array
