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

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

"""Compressed Sparse Row matrix format""" 

 

from __future__ import division, print_function, absolute_import 

 

__docformat__ = "restructuredtext en" 

 

__all__ = ['csr_matrix', 'isspmatrix_csr'] 

 

import numpy as np 

from scipy._lib.six import xrange 

 

from .base import spmatrix 

from ._sparsetools import (csr_tocsc, csr_tobsr, csr_count_blocks, 

get_csr_submatrix) 

from .sputils import upcast, get_index_dtype 

 

from .compressed import _cs_matrix 

 

 

class csr_matrix(_cs_matrix): 

""" 

Compressed Sparse Row matrix 

 

This can be instantiated in several ways: 

csr_matrix(D) 

with a dense matrix or rank-2 ndarray D 

 

csr_matrix(S) 

with another sparse matrix S (equivalent to S.tocsr()) 

 

csr_matrix((M, N), [dtype]) 

to construct an empty matrix with shape (M, N) 

dtype is optional, defaulting to dtype='d'. 

 

csr_matrix((data, (row_ind, col_ind)), [shape=(M, N)]) 

where ``data``, ``row_ind`` and ``col_ind`` satisfy the 

relationship ``a[row_ind[k], col_ind[k]] = data[k]``. 

 

csr_matrix((data, indices, indptr), [shape=(M, N)]) 

is the standard CSR representation where the column indices for 

row i are stored in ``indices[indptr[i]:indptr[i+1]]`` and their 

corresponding values are stored in ``data[indptr[i]:indptr[i+1]]``. 

If the shape parameter is not supplied, the matrix dimensions 

are inferred from the index arrays. 

 

Attributes 

---------- 

dtype : dtype 

Data type of the matrix 

shape : 2-tuple 

Shape of the matrix 

ndim : int 

Number of dimensions (this is always 2) 

nnz 

Number of nonzero elements 

data 

CSR format data array of the matrix 

indices 

CSR format index array of the matrix 

indptr 

CSR format index pointer array of the matrix 

has_sorted_indices 

Whether indices are sorted 

 

Notes 

----- 

 

Sparse matrices can be used in arithmetic operations: they support 

addition, subtraction, multiplication, division, and matrix power. 

 

Advantages of the CSR format 

- efficient arithmetic operations CSR + CSR, CSR * CSR, etc. 

- efficient row slicing 

- fast matrix vector products 

 

Disadvantages of the CSR format 

- slow column slicing operations (consider CSC) 

- changes to the sparsity structure are expensive (consider LIL or DOK) 

 

Examples 

-------- 

 

>>> import numpy as np 

>>> from scipy.sparse import csr_matrix 

>>> csr_matrix((3, 4), dtype=np.int8).toarray() 

array([[0, 0, 0, 0], 

[0, 0, 0, 0], 

[0, 0, 0, 0]], dtype=int8) 

 

>>> row = np.array([0, 0, 1, 2, 2, 2]) 

>>> col = np.array([0, 2, 2, 0, 1, 2]) 

>>> data = np.array([1, 2, 3, 4, 5, 6]) 

>>> csr_matrix((data, (row, col)), shape=(3, 3)).toarray() 

array([[1, 0, 2], 

[0, 0, 3], 

[4, 5, 6]]) 

 

>>> indptr = np.array([0, 2, 3, 6]) 

>>> indices = np.array([0, 2, 2, 0, 1, 2]) 

>>> data = np.array([1, 2, 3, 4, 5, 6]) 

>>> csr_matrix((data, indices, indptr), shape=(3, 3)).toarray() 

array([[1, 0, 2], 

[0, 0, 3], 

[4, 5, 6]]) 

 

As an example of how to construct a CSR matrix incrementally, 

the following snippet builds a term-document matrix from texts: 

 

>>> docs = [["hello", "world", "hello"], ["goodbye", "cruel", "world"]] 

>>> indptr = [0] 

>>> indices = [] 

>>> data = [] 

>>> vocabulary = {} 

>>> for d in docs: 

... for term in d: 

... index = vocabulary.setdefault(term, len(vocabulary)) 

... indices.append(index) 

... data.append(1) 

... indptr.append(len(indices)) 

... 

>>> csr_matrix((data, indices, indptr), dtype=int).toarray() 

array([[2, 1, 0, 0], 

[0, 1, 1, 1]]) 

 

""" 

format = 'csr' 

 

def transpose(self, axes=None, copy=False): 

if axes is not None: 

raise ValueError(("Sparse matrices do not support " 

"an 'axes' parameter because swapping " 

"dimensions is the only logical permutation.")) 

 

M, N = self.shape 

 

from .csc import csc_matrix 

return csc_matrix((self.data, self.indices, 

self.indptr), shape=(N, M), copy=copy) 

 

transpose.__doc__ = spmatrix.transpose.__doc__ 

 

def tolil(self, copy=False): 

from .lil import lil_matrix 

lil = lil_matrix(self.shape,dtype=self.dtype) 

 

self.sum_duplicates() 

ptr,ind,dat = self.indptr,self.indices,self.data 

rows, data = lil.rows, lil.data 

 

for n in xrange(self.shape[0]): 

start = ptr[n] 

end = ptr[n+1] 

rows[n] = ind[start:end].tolist() 

data[n] = dat[start:end].tolist() 

 

return lil 

 

tolil.__doc__ = spmatrix.tolil.__doc__ 

 

def tocsr(self, copy=False): 

if copy: 

return self.copy() 

else: 

return self 

 

tocsr.__doc__ = spmatrix.tocsr.__doc__ 

 

def tocsc(self, copy=False): 

idx_dtype = get_index_dtype((self.indptr, self.indices), 

maxval=max(self.nnz, self.shape[0])) 

indptr = np.empty(self.shape[1] + 1, dtype=idx_dtype) 

indices = np.empty(self.nnz, dtype=idx_dtype) 

data = np.empty(self.nnz, dtype=upcast(self.dtype)) 

 

csr_tocsc(self.shape[0], self.shape[1], 

self.indptr.astype(idx_dtype), 

self.indices.astype(idx_dtype), 

self.data, 

indptr, 

indices, 

data) 

 

from .csc import csc_matrix 

A = csc_matrix((data, indices, indptr), shape=self.shape) 

A.has_sorted_indices = True 

return A 

 

tocsc.__doc__ = spmatrix.tocsc.__doc__ 

 

def tobsr(self, blocksize=None, copy=True): 

from .bsr import bsr_matrix 

 

if blocksize is None: 

from .spfuncs import estimate_blocksize 

return self.tobsr(blocksize=estimate_blocksize(self)) 

 

elif blocksize == (1,1): 

arg1 = (self.data.reshape(-1,1,1),self.indices,self.indptr) 

return bsr_matrix(arg1, shape=self.shape, copy=copy) 

 

else: 

R,C = blocksize 

M,N = self.shape 

 

if R < 1 or C < 1 or M % R != 0 or N % C != 0: 

raise ValueError('invalid blocksize %s' % blocksize) 

 

blks = csr_count_blocks(M,N,R,C,self.indptr,self.indices) 

 

idx_dtype = get_index_dtype((self.indptr, self.indices), 

maxval=max(N//C, blks)) 

indptr = np.empty(M//R+1, dtype=idx_dtype) 

indices = np.empty(blks, dtype=idx_dtype) 

data = np.zeros((blks,R,C), dtype=self.dtype) 

 

csr_tobsr(M, N, R, C, 

self.indptr.astype(idx_dtype), 

self.indices.astype(idx_dtype), 

self.data, 

indptr, indices, data.ravel()) 

 

return bsr_matrix((data,indices,indptr), shape=self.shape) 

 

tobsr.__doc__ = spmatrix.tobsr.__doc__ 

 

# these functions are used by the parent class (_cs_matrix) 

# to remove redudancy between csc_matrix and csr_matrix 

def _swap(self, x): 

"""swap the members of x if this is a column-oriented matrix 

""" 

return x 

 

def __iter__(self): 

indptr = np.zeros(2, dtype=self.indptr.dtype) 

shape = (1, self.shape[1]) 

i0 = 0 

for i1 in self.indptr[1:]: 

indptr[1] = i1 - i0 

indices = self.indices[i0:i1] 

data = self.data[i0:i1] 

yield csr_matrix((data, indices, indptr), shape=shape, copy=True) 

i0 = i1 

 

def getrow(self, i): 

"""Returns a copy of row i of the matrix, as a (1 x n) 

CSR matrix (row vector). 

""" 

M, N = self.shape 

i = int(i) 

if i < 0: 

i += M 

if i < 0 or i >= M: 

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

indptr, indices, data = get_csr_submatrix( 

M, N, self.indptr, self.indices, self.data, i, i + 1, 0, N) 

return csr_matrix((data, indices, indptr), shape=(1, N), 

dtype=self.dtype, copy=False) 

 

def getcol(self, i): 

"""Returns a copy of column i of the matrix, as a (m x 1) 

CSR matrix (column vector). 

""" 

M, N = self.shape 

i = int(i) 

if i < 0: 

i += N 

if i < 0 or i >= N: 

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

indptr, indices, data = get_csr_submatrix( 

M, N, self.indptr, self.indices, self.data, 0, M, i, i + 1) 

return csr_matrix((data, indices, indptr), shape=(M, 1), 

dtype=self.dtype, copy=False) 

 

def _get_intXarray(self, row, col): 

return self.getrow(row)._minor_index_fancy(col) 

 

def _get_intXslice(self, row, col): 

if col.step in (1, None): 

return self._get_submatrix(row, col, copy=True) 

# TODO: uncomment this once it's faster: 

# return self.getrow(row)._minor_slice(col) 

 

M, N = self.shape 

start, stop, stride = col.indices(N) 

 

ii, jj = self.indptr[row:row+2] 

row_indices = self.indices[ii:jj] 

row_data = self.data[ii:jj] 

 

if stride > 0: 

ind = (row_indices >= start) & (row_indices < stop) 

else: 

ind = (row_indices <= start) & (row_indices > stop) 

 

if abs(stride) > 1: 

ind &= (row_indices - start) % stride == 0 

 

row_indices = (row_indices[ind] - start) // stride 

row_data = row_data[ind] 

row_indptr = np.array([0, len(row_indices)]) 

 

if stride < 0: 

row_data = row_data[::-1] 

row_indices = abs(row_indices[::-1]) 

 

shape = (1, int(np.ceil(float(stop - start) / stride))) 

return csr_matrix((row_data, row_indices, row_indptr), shape=shape, 

dtype=self.dtype, copy=False) 

 

def _get_sliceXint(self, row, col): 

if row.step in (1, None): 

return self._get_submatrix(row, col, copy=True) 

return self._major_slice(row)._get_submatrix(minor=col) 

 

def _get_sliceXarray(self, row, col): 

return self._major_slice(row)._minor_index_fancy(col) 

 

def _get_arrayXint(self, row, col): 

return self._major_index_fancy(row)._get_submatrix(minor=col) 

 

def _get_arrayXslice(self, row, col): 

if col.step not in (1, None): 

col = np.arange(*col.indices(self.shape[1])) 

return self._get_arrayXarray(row, col) 

return self._major_index_fancy(row)._get_submatrix(minor=col) 

 

 

def isspmatrix_csr(x): 

"""Is x of csr_matrix type? 

 

Parameters 

---------- 

x 

object to check for being a csr matrix 

 

Returns 

------- 

bool 

True if x is a csr matrix, False otherwise 

 

Examples 

-------- 

>>> from scipy.sparse import csr_matrix, isspmatrix_csr 

>>> isspmatrix_csr(csr_matrix([[5]])) 

True 

 

>>> from scipy.sparse import csc_matrix, csr_matrix, isspmatrix_csc 

>>> isspmatrix_csr(csc_matrix([[5]])) 

False 

""" 

return isinstance(x, csr_matrix)