1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

61

62

63

64

65

66

67

68

69

70

71

72

73

74

75

76

77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

""" 

Utilities that manipulate strides to achieve desirable effects. 

 

An explanation of strides can be found in the "ndarray.rst" file in the 

NumPy reference guide. 

 

""" 

from __future__ import division, absolute_import, print_function 

 

import numpy as np 

from numpy.core.overrides import array_function_dispatch 

 

__all__ = ['broadcast_to', 'broadcast_arrays'] 

 

 

class DummyArray(object): 

"""Dummy object that just exists to hang __array_interface__ dictionaries 

and possibly keep alive a reference to a base array. 

""" 

 

def __init__(self, interface, base=None): 

self.__array_interface__ = interface 

self.base = base 

 

 

def _maybe_view_as_subclass(original_array, new_array): 

if type(original_array) is not type(new_array): 

# if input was an ndarray subclass and subclasses were OK, 

# then view the result as that subclass. 

new_array = new_array.view(type=type(original_array)) 

# Since we have done something akin to a view from original_array, we 

# should let the subclass finalize (if it has it implemented, i.e., is 

# not None). 

if new_array.__array_finalize__: 

new_array.__array_finalize__(original_array) 

return new_array 

 

 

def as_strided(x, shape=None, strides=None, subok=False, writeable=True): 

""" 

Create a view into the array with the given shape and strides. 

 

.. warning:: This function has to be used with extreme care, see notes. 

 

Parameters 

---------- 

x : ndarray 

Array to create a new. 

shape : sequence of int, optional 

The shape of the new array. Defaults to ``x.shape``. 

strides : sequence of int, optional 

The strides of the new array. Defaults to ``x.strides``. 

subok : bool, optional 

.. versionadded:: 1.10 

 

If True, subclasses are preserved. 

writeable : bool, optional 

.. versionadded:: 1.12 

 

If set to False, the returned array will always be readonly. 

Otherwise it will be writable if the original array was. It 

is advisable to set this to False if possible (see Notes). 

 

Returns 

------- 

view : ndarray 

 

See also 

-------- 

broadcast_to: broadcast an array to a given shape. 

reshape : reshape an array. 

 

Notes 

----- 

``as_strided`` creates a view into the array given the exact strides 

and shape. This means it manipulates the internal data structure of 

ndarray and, if done incorrectly, the array elements can point to 

invalid memory and can corrupt results or crash your program. 

It is advisable to always use the original ``x.strides`` when 

calculating new strides to avoid reliance on a contiguous memory 

layout. 

 

Furthermore, arrays created with this function often contain self 

overlapping memory, so that two elements are identical. 

Vectorized write operations on such arrays will typically be 

unpredictable. They may even give different results for small, large, 

or transposed arrays. 

Since writing to these arrays has to be tested and done with great 

care, you may want to use ``writeable=False`` to avoid accidental write 

operations. 

 

For these reasons it is advisable to avoid ``as_strided`` when 

possible. 

""" 

# first convert input to array, possibly keeping subclass 

x = np.array(x, copy=False, subok=subok) 

interface = dict(x.__array_interface__) 

if shape is not None: 

interface['shape'] = tuple(shape) 

if strides is not None: 

interface['strides'] = tuple(strides) 

 

array = np.asarray(DummyArray(interface, base=x)) 

# The route via `__interface__` does not preserve structured 

# dtypes. Since dtype should remain unchanged, we set it explicitly. 

array.dtype = x.dtype 

 

view = _maybe_view_as_subclass(x, array) 

 

if view.flags.writeable and not writeable: 

view.flags.writeable = False 

 

return view 

 

 

def _broadcast_to(array, shape, subok, readonly): 

shape = tuple(shape) if np.iterable(shape) else (shape,) 

array = np.array(array, copy=False, subok=subok) 

if not shape and array.shape: 

raise ValueError('cannot broadcast a non-scalar to a scalar array') 

if any(size < 0 for size in shape): 

raise ValueError('all elements of broadcast shape must be non-' 

'negative') 

needs_writeable = not readonly and array.flags.writeable 

extras = ['reduce_ok'] if needs_writeable else [] 

op_flag = 'readwrite' if needs_writeable else 'readonly' 

it = np.nditer( 

(array,), flags=['multi_index', 'refs_ok', 'zerosize_ok'] + extras, 

op_flags=[op_flag], itershape=shape, order='C') 

with it: 

# never really has writebackifcopy semantics 

broadcast = it.itviews[0] 

result = _maybe_view_as_subclass(array, broadcast) 

if needs_writeable and not result.flags.writeable: 

result.flags.writeable = True 

return result 

 

 

def _broadcast_to_dispatcher(array, shape, subok=None): 

return (array,) 

 

 

@array_function_dispatch(_broadcast_to_dispatcher, module='numpy') 

def broadcast_to(array, shape, subok=False): 

"""Broadcast an array to a new shape. 

 

Parameters 

---------- 

array : array_like 

The array to broadcast. 

shape : tuple 

The shape of the desired array. 

subok : bool, optional 

If True, then sub-classes will be passed-through, otherwise 

the returned array will be forced to be a base-class array (default). 

 

Returns 

------- 

broadcast : array 

A readonly view on the original array with the given shape. It is 

typically not contiguous. Furthermore, more than one element of a 

broadcasted array may refer to a single memory location. 

 

Raises 

------ 

ValueError 

If the array is not compatible with the new shape according to NumPy's 

broadcasting rules. 

 

Notes 

----- 

.. versionadded:: 1.10.0 

 

Examples 

-------- 

>>> x = np.array([1, 2, 3]) 

>>> np.broadcast_to(x, (3, 3)) 

array([[1, 2, 3], 

[1, 2, 3], 

[1, 2, 3]]) 

""" 

return _broadcast_to(array, shape, subok=subok, readonly=True) 

 

 

def _broadcast_shape(*args): 

"""Returns the shape of the arrays that would result from broadcasting the 

supplied arrays against each other. 

""" 

if not args: 

return () 

# use the old-iterator because np.nditer does not handle size 0 arrays 

# consistently 

b = np.broadcast(*args[:32]) 

# unfortunately, it cannot handle 32 or more arguments directly 

for pos in range(32, len(args), 31): 

# ironically, np.broadcast does not properly handle np.broadcast 

# objects (it treats them as scalars) 

# use broadcasting to avoid allocating the full array 

b = broadcast_to(0, b.shape) 

b = np.broadcast(b, *args[pos:(pos + 31)]) 

return b.shape 

 

 

def _broadcast_arrays_dispatcher(*args, **kwargs): 

return args 

 

 

@array_function_dispatch(_broadcast_arrays_dispatcher, module='numpy') 

def broadcast_arrays(*args, **kwargs): 

""" 

Broadcast any number of arrays against each other. 

 

Parameters 

---------- 

`*args` : array_likes 

The arrays to broadcast. 

 

subok : bool, optional 

If True, then sub-classes will be passed-through, otherwise 

the returned arrays will be forced to be a base-class array (default). 

 

Returns 

------- 

broadcasted : list of arrays 

These arrays are views on the original arrays. They are typically 

not contiguous. Furthermore, more than one element of a 

broadcasted array may refer to a single memory location. If you 

need to write to the arrays, make copies first. 

 

Examples 

-------- 

>>> x = np.array([[1,2,3]]) 

>>> y = np.array([[4],[5]]) 

>>> np.broadcast_arrays(x, y) 

[array([[1, 2, 3], 

[1, 2, 3]]), array([[4, 4, 4], 

[5, 5, 5]])] 

 

Here is a useful idiom for getting contiguous copies instead of 

non-contiguous views. 

 

>>> [np.array(a) for a in np.broadcast_arrays(x, y)] 

[array([[1, 2, 3], 

[1, 2, 3]]), array([[4, 4, 4], 

[5, 5, 5]])] 

 

""" 

# nditer is not used here to avoid the limit of 32 arrays. 

# Otherwise, something like the following one-liner would suffice: 

# return np.nditer(args, flags=['multi_index', 'zerosize_ok'], 

# order='C').itviews 

 

subok = kwargs.pop('subok', False) 

if kwargs: 

raise TypeError('broadcast_arrays() got an unexpected keyword ' 

'argument {!r}'.format(list(kwargs.keys())[0])) 

args = [np.array(_m, copy=False, subok=subok) for _m in args] 

 

shape = _broadcast_shape(*args) 

 

if all(array.shape == shape for array in args): 

# Common case where nothing needs to be broadcasted. 

return args 

 

# TODO: consider making the results of broadcast_arrays readonly to match 

# broadcast_to. This will require a deprecation cycle. 

return [_broadcast_to(array, shape, subok=subok, readonly=False) 

for array in args]