Hide keyboard shortcuts

Hot-keys on this page

r m x p   toggle line displays

j k   next/prev highlighted chunk

0   (zero) top of page

1   (one) first highlighted chunk

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

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

"""Indexing mixin for sparse matrix classes. 

""" 

from __future__ import division, print_function, absolute_import 

 

import numpy as np 

from .sputils import isintlike 

 

try: 

INT_TYPES = (int, long, np.integer) 

except NameError: 

# long is not defined in Python3 

INT_TYPES = (int, np.integer) 

 

 

class IndexMixin(object): 

""" 

This class provides common dispatching and validation logic for indexing. 

""" 

def __getitem__(self, key): 

row, col = self._validate_indices(key) 

# Dispatch to specialized methods. 

if isinstance(row, INT_TYPES): 

if isinstance(col, INT_TYPES): 

return self._get_intXint(row, col) 

elif isinstance(col, slice): 

return self._get_intXslice(row, col) 

elif col.ndim == 1: 

return self._get_intXarray(row, col) 

raise IndexError('index results in >2 dimensions') 

elif isinstance(row, slice): 

if isinstance(col, INT_TYPES): 

return self._get_sliceXint(row, col) 

elif isinstance(col, slice): 

if row == slice(None) and row == col: 

return self.copy() 

return self._get_sliceXslice(row, col) 

elif col.ndim == 1: 

return self._get_sliceXarray(row, col) 

raise IndexError('index results in >2 dimensions') 

elif row.ndim == 1: 

if isinstance(col, INT_TYPES): 

return self._get_arrayXint(row, col) 

elif isinstance(col, slice): 

return self._get_arrayXslice(row, col) 

else: # row.ndim == 2 

if isinstance(col, INT_TYPES): 

return self._get_arrayXint(row, col) 

elif isinstance(col, slice): 

raise IndexError('index results in >2 dimensions') 

elif row.shape[1] == 1 and col.ndim == 1: 

# special case for outer indexing 

return self._get_columnXarray(row[:,0], col) 

 

# The only remaining case is inner (fancy) indexing 

row, col = np.broadcast_arrays(row, col) 

if row.shape != col.shape: 

raise IndexError('number of row and column indices differ') 

if row.size == 0: 

return self.__class__(np.atleast_2d(row).shape, dtype=self.dtype) 

return self._get_arrayXarray(row, col) 

 

def __setitem__(self, key, x): 

row, col = self._validate_indices(key) 

 

if isinstance(row, INT_TYPES) and isinstance(col, INT_TYPES): 

x = np.asarray(x, dtype=self.dtype) 

if x.size != 1: 

raise ValueError('Trying to assign a sequence to an item') 

self._set_intXint(row, col, x.flat[0]) 

return 

 

if isinstance(row, slice): 

row = np.arange(*row.indices(self.shape[0]))[:, None] 

else: 

row = np.atleast_1d(row) 

 

if isinstance(col, slice): 

col = np.arange(*col.indices(self.shape[1]))[None, :] 

if row.ndim == 1: 

row = row[:, None] 

else: 

col = np.atleast_1d(col) 

 

i, j = np.broadcast_arrays(row, col) 

if i.shape != j.shape: 

raise IndexError('number of row and column indices differ') 

 

from .base import isspmatrix 

if isspmatrix(x): 

if i.ndim == 1: 

# Inner indexing, so treat them like row vectors. 

i = i[None] 

j = j[None] 

broadcast_row = x.shape[0] == 1 and i.shape[0] != 1 

broadcast_col = x.shape[1] == 1 and i.shape[1] != 1 

if not ((broadcast_row or x.shape[0] == i.shape[0]) and 

(broadcast_col or x.shape[1] == i.shape[1])): 

raise ValueError('shape mismatch in assignment') 

if x.size == 0: 

return 

x = x.tocoo(copy=True) 

x.sum_duplicates() 

self._set_arrayXarray_sparse(i, j, x) 

else: 

# Make x and i into the same shape 

x = np.asarray(x, dtype=self.dtype) 

x, _ = np.broadcast_arrays(x, i) 

if x.shape != i.shape: 

raise ValueError("shape mismatch in assignment") 

if x.size == 0: 

return 

self._set_arrayXarray(i, j, x) 

 

def _validate_indices(self, key): 

M, N = self.shape 

row, col = _unpack_index(key) 

 

if isintlike(row): 

row = int(row) 

if row < -M or row >= M: 

raise IndexError('row index (%d) out of range' % row) 

if row < 0: 

row += M 

elif not isinstance(row, slice): 

row = self._asindices(row, M) 

 

if isintlike(col): 

col = int(col) 

if col < -N or col >= N: 

raise IndexError('column index (%d) out of range' % col) 

if col < 0: 

col += N 

elif not isinstance(col, slice): 

col = self._asindices(col, N) 

 

return row, col 

 

def _asindices(self, idx, length): 

"""Convert `idx` to a valid index for an axis with a given length. 

 

Subclasses that need special validation can override this method. 

""" 

try: 

x = np.asarray(idx) 

except (ValueError, TypeError, MemoryError): 

raise IndexError('invalid index') 

 

if x.ndim not in (1, 2): 

raise IndexError('Index dimension must be <= 2') 

 

if x.size == 0: 

return x 

 

# Check bounds 

max_indx = x.max() 

if max_indx >= length: 

raise IndexError('index (%d) out of range' % max_indx) 

 

min_indx = x.min() 

if min_indx < 0: 

if min_indx < -length: 

raise IndexError('index (%d) out of range' % min_indx) 

if x is idx or not x.flags.owndata: 

x = x.copy() 

x[x < 0] += length 

return x 

 

def getrow(self, i): 

"""Return a copy of row i of the matrix, as a (1 x n) row vector. 

""" 

M, N = self.shape 

i = int(i) 

if i < -M or i >= M: 

raise IndexError('index (%d) out of range' % i) 

if i < 0: 

i += M 

return self._get_intXslice(i, slice(None)) 

 

def getcol(self, i): 

"""Return a copy of column i of the matrix, as a (m x 1) column vector. 

""" 

M, N = self.shape 

i = int(i) 

if i < -N or i >= N: 

raise IndexError('index (%d) out of range' % i) 

if i < 0: 

i += N 

return self._get_sliceXint(slice(None), i) 

 

def _get_intXint(self, row, col): 

raise NotImplementedError() 

 

def _get_intXarray(self, row, col): 

raise NotImplementedError() 

 

def _get_intXslice(self, row, col): 

raise NotImplementedError() 

 

def _get_sliceXint(self, row, col): 

raise NotImplementedError() 

 

def _get_sliceXslice(self, row, col): 

raise NotImplementedError() 

 

def _get_sliceXarray(self, row, col): 

raise NotImplementedError() 

 

def _get_arrayXint(self, row, col): 

raise NotImplementedError() 

 

def _get_arrayXslice(self, row, col): 

raise NotImplementedError() 

 

def _get_columnXarray(self, row, col): 

raise NotImplementedError() 

 

def _get_arrayXarray(self, row, col): 

raise NotImplementedError() 

 

def _set_intXint(self, row, col, x): 

raise NotImplementedError() 

 

def _set_arrayXarray(self, row, col, x): 

raise NotImplementedError() 

 

def _set_arrayXarray_sparse(self, row, col, x): 

# Fall back to densifying x 

x = np.asarray(x.toarray(), dtype=self.dtype) 

x, _ = np.broadcast_arrays(x, row) 

self._set_arrayXarray(row, col, x) 

 

 

def _unpack_index(index): 

""" Parse index. Always return a tuple of the form (row, col). 

Valid type for row/col is integer, slice, or array of integers. 

""" 

# First, check if indexing with single boolean matrix. 

from .base import spmatrix, isspmatrix 

if (isinstance(index, (spmatrix, np.ndarray)) and 

index.ndim == 2 and index.dtype.kind == 'b'): 

return index.nonzero() 

 

# Parse any ellipses. 

index = _check_ellipsis(index) 

 

# Next, parse the tuple or object 

if isinstance(index, tuple): 

if len(index) == 2: 

row, col = index 

elif len(index) == 1: 

row, col = index[0], slice(None) 

else: 

raise IndexError('invalid number of indices') 

else: 

row, col = index, slice(None) 

 

# Next, check for validity and transform the index as needed. 

if isspmatrix(row) or isspmatrix(col): 

# Supporting sparse boolean indexing with both row and col does 

# not work because spmatrix.ndim is always 2. 

raise IndexError( 

'Indexing with sparse matrices is not supported ' 

'except boolean indexing where matrix and index ' 

'are equal shapes.') 

if isinstance(row, np.ndarray) and row.dtype.kind == 'b': 

row = _boolean_index_to_array(row) 

if isinstance(col, np.ndarray) and col.dtype.kind == 'b': 

col = _boolean_index_to_array(col) 

return row, col 

 

 

def _check_ellipsis(index): 

"""Process indices with Ellipsis. Returns modified index.""" 

if index is Ellipsis: 

return (slice(None), slice(None)) 

 

if not isinstance(index, tuple): 

return index 

 

# TODO: Deprecate this multiple-ellipsis handling, 

# as numpy no longer supports it. 

 

# Find first ellipsis. 

for j, v in enumerate(index): 

if v is Ellipsis: 

first_ellipsis = j 

break 

else: 

return index 

 

# Try to expand it using shortcuts for common cases 

if len(index) == 1: 

return (slice(None), slice(None)) 

if len(index) == 2: 

if first_ellipsis == 0: 

if index[1] is Ellipsis: 

return (slice(None), slice(None)) 

return (slice(None), index[1]) 

return (index[0], slice(None)) 

 

# Expand it using a general-purpose algorithm 

tail = [] 

for v in index[first_ellipsis+1:]: 

if v is not Ellipsis: 

tail.append(v) 

nd = first_ellipsis + len(tail) 

nslice = max(0, 2 - nd) 

return index[:first_ellipsis] + (slice(None),)*nslice + tuple(tail) 

 

 

def _boolean_index_to_array(idx): 

if idx.ndim > 1: 

raise IndexError('invalid index shape') 

return idx.nonzero()[0]