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

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

"""QR decomposition functions.""" 

from __future__ import division, print_function, absolute_import 

 

import numpy 

 

# Local imports 

from .lapack import get_lapack_funcs 

from .misc import _datacopied 

 

__all__ = ['qr', 'qr_multiply', 'rq'] 

 

 

def safecall(f, name, *args, **kwargs): 

"""Call a LAPACK routine, determining lwork automatically and handling 

error return values""" 

lwork = kwargs.get("lwork", None) 

if lwork in (None, -1): 

kwargs['lwork'] = -1 

ret = f(*args, **kwargs) 

kwargs['lwork'] = ret[-2][0].real.astype(numpy.int) 

ret = f(*args, **kwargs) 

if ret[-1] < 0: 

raise ValueError("illegal value in %d-th argument of internal %s" 

% (-ret[-1], name)) 

return ret[:-2] 

 

 

def qr(a, overwrite_a=False, lwork=None, mode='full', pivoting=False, 

check_finite=True): 

""" 

Compute QR decomposition of a matrix. 

 

Calculate the decomposition ``A = Q R`` where Q is unitary/orthogonal 

and R upper triangular. 

 

Parameters 

---------- 

a : (M, N) array_like 

Matrix to be decomposed 

overwrite_a : bool, optional 

Whether data in a is overwritten (may improve performance) 

lwork : int, optional 

Work array size, lwork >= a.shape[1]. If None or -1, an optimal size 

is computed. 

mode : {'full', 'r', 'economic', 'raw'}, optional 

Determines what information is to be returned: either both Q and R 

('full', default), only R ('r') or both Q and R but computed in 

economy-size ('economic', see Notes). The final option 'raw' 

(added in Scipy 0.11) makes the function return two matrices 

(Q, TAU) in the internal format used by LAPACK. 

pivoting : bool, optional 

Whether or not factorization should include pivoting for rank-revealing 

qr decomposition. If pivoting, compute the decomposition 

``A P = Q R`` as above, but where P is chosen such that the diagonal 

of R is non-increasing. 

check_finite : bool, optional 

Whether to check that the input matrix contains only finite numbers. 

Disabling may give a performance gain, but may result in problems 

(crashes, non-termination) if the inputs do contain infinities or NaNs. 

 

Returns 

------- 

Q : float or complex ndarray 

Of shape (M, M), or (M, K) for ``mode='economic'``. Not returned 

if ``mode='r'``. 

R : float or complex ndarray 

Of shape (M, N), or (K, N) for ``mode='economic'``. ``K = min(M, N)``. 

P : int ndarray 

Of shape (N,) for ``pivoting=True``. Not returned if 

``pivoting=False``. 

 

Raises 

------ 

LinAlgError 

Raised if decomposition fails 

 

Notes 

----- 

This is an interface to the LAPACK routines dgeqrf, zgeqrf, 

dorgqr, zungqr, dgeqp3, and zgeqp3. 

 

If ``mode=economic``, the shapes of Q and R are (M, K) and (K, N) instead 

of (M,M) and (M,N), with ``K=min(M,N)``. 

 

Examples 

-------- 

>>> from scipy import random, linalg, dot, diag, all, allclose 

>>> a = random.randn(9, 6) 

 

>>> q, r = linalg.qr(a) 

>>> allclose(a, np.dot(q, r)) 

True 

>>> q.shape, r.shape 

((9, 9), (9, 6)) 

 

>>> r2 = linalg.qr(a, mode='r') 

>>> allclose(r, r2) 

True 

 

>>> q3, r3 = linalg.qr(a, mode='economic') 

>>> q3.shape, r3.shape 

((9, 6), (6, 6)) 

 

>>> q4, r4, p4 = linalg.qr(a, pivoting=True) 

>>> d = abs(diag(r4)) 

>>> all(d[1:] <= d[:-1]) 

True 

>>> allclose(a[:, p4], dot(q4, r4)) 

True 

>>> q4.shape, r4.shape, p4.shape 

((9, 9), (9, 6), (6,)) 

 

>>> q5, r5, p5 = linalg.qr(a, mode='economic', pivoting=True) 

>>> q5.shape, r5.shape, p5.shape 

((9, 6), (6, 6), (6,)) 

 

""" 

# 'qr' was the old default, equivalent to 'full'. Neither 'full' nor 

# 'qr' are used below. 

# 'raw' is used internally by qr_multiply 

if mode not in ['full', 'qr', 'r', 'economic', 'raw']: 

raise ValueError("Mode argument should be one of ['full', 'r'," 

"'economic', 'raw']") 

 

if check_finite: 

a1 = numpy.asarray_chkfinite(a) 

else: 

a1 = numpy.asarray(a) 

if len(a1.shape) != 2: 

raise ValueError("expected 2D array") 

M, N = a1.shape 

overwrite_a = overwrite_a or (_datacopied(a1, a)) 

 

if pivoting: 

geqp3, = get_lapack_funcs(('geqp3',), (a1,)) 

qr, jpvt, tau = safecall(geqp3, "geqp3", a1, overwrite_a=overwrite_a) 

jpvt -= 1 # geqp3 returns a 1-based index array, so subtract 1 

else: 

geqrf, = get_lapack_funcs(('geqrf',), (a1,)) 

qr, tau = safecall(geqrf, "geqrf", a1, lwork=lwork, 

overwrite_a=overwrite_a) 

 

if mode not in ['economic', 'raw'] or M < N: 

R = numpy.triu(qr) 

else: 

R = numpy.triu(qr[:N, :]) 

 

if pivoting: 

Rj = R, jpvt 

else: 

Rj = R, 

 

if mode == 'r': 

return Rj 

elif mode == 'raw': 

return ((qr, tau),) + Rj 

 

gor_un_gqr, = get_lapack_funcs(('orgqr',), (qr,)) 

 

if M < N: 

Q, = safecall(gor_un_gqr, "gorgqr/gungqr", qr[:, :M], tau, 

lwork=lwork, overwrite_a=1) 

elif mode == 'economic': 

Q, = safecall(gor_un_gqr, "gorgqr/gungqr", qr, tau, lwork=lwork, 

overwrite_a=1) 

else: 

t = qr.dtype.char 

qqr = numpy.empty((M, M), dtype=t) 

qqr[:, :N] = qr 

Q, = safecall(gor_un_gqr, "gorgqr/gungqr", qqr, tau, lwork=lwork, 

overwrite_a=1) 

 

return (Q,) + Rj 

 

 

def qr_multiply(a, c, mode='right', pivoting=False, conjugate=False, 

overwrite_a=False, overwrite_c=False): 

""" 

Calculate the QR decomposition and multiply Q with a matrix. 

 

Calculate the decomposition ``A = Q R`` where Q is unitary/orthogonal 

and R upper triangular. Multiply Q with a vector or a matrix c. 

 

Parameters 

---------- 

a : (M, N), array_like 

Input array 

c : array_like 

Input array to be multiplied by ``q``. 

mode : {'left', 'right'}, optional 

``Q @ c`` is returned if mode is 'left', ``c @ Q`` is returned if 

mode is 'right'. 

The shape of c must be appropriate for the matrix multiplications, 

if mode is 'left', ``min(a.shape) == c.shape[0]``, 

if mode is 'right', ``a.shape[0] == c.shape[1]``. 

pivoting : bool, optional 

Whether or not factorization should include pivoting for rank-revealing 

qr decomposition, see the documentation of qr. 

conjugate : bool, optional 

Whether Q should be complex-conjugated. This might be faster 

than explicit conjugation. 

overwrite_a : bool, optional 

Whether data in a is overwritten (may improve performance) 

overwrite_c : bool, optional 

Whether data in c is overwritten (may improve performance). 

If this is used, c must be big enough to keep the result, 

i.e. ``c.shape[0]`` = ``a.shape[0]`` if mode is 'left'. 

 

Returns 

------- 

CQ : ndarray 

The product of ``Q`` and ``c``. 

R : (K, N), ndarray 

R array of the resulting QR factorization where ``K = min(M, N)``. 

P : (N,) ndarray 

Integer pivot array. Only returned when ``pivoting=True``. 

 

Raises 

------ 

LinAlgError 

Raised if QR decomposition fails. 

 

Notes 

----- 

This is an interface to the LAPACK routines ``?GEQRF``, ``?ORMQR``, 

``?UNMQR``, and ``?GEQP3``. 

 

.. versionadded:: 0.11.0 

 

Examples 

-------- 

>>> from scipy.linalg import qr_multiply, qr 

>>> A = np.array([[1, 3, 3], [2, 3, 2], [2, 3, 3], [1, 3, 2]]) 

>>> qc, r1, piv1 = qr_multiply(A, 2*np.eye(4), pivoting=1) 

>>> qc 

array([[-1., 1., -1.], 

[-1., -1., 1.], 

[-1., -1., -1.], 

[-1., 1., 1.]]) 

>>> r1 

array([[-6., -3., -5. ], 

[ 0., -1., -1.11022302e-16], 

[ 0., 0., -1. ]]) 

>>> piv1 

array([1, 0, 2], dtype=int32) 

>>> q2, r2, piv2 = qr(A, mode='economic', pivoting=1) 

>>> np.allclose(2*q2 - qc, np.zeros((4, 3))) 

True 

 

""" 

if mode not in ['left', 'right']: 

raise ValueError("Mode argument can only be 'left' or 'right' but " 

"not '{}'".format(mode)) 

c = numpy.asarray_chkfinite(c) 

if c.ndim < 2: 

onedim = True 

c = numpy.atleast_2d(c) 

if mode == "left": 

c = c.T 

else: 

onedim = False 

 

a = numpy.atleast_2d(numpy.asarray(a)) # chkfinite done in qr 

M, N = a.shape 

 

if mode == 'left': 

if c.shape[0] != min(M, N + overwrite_c*(M-N)): 

raise ValueError('Array shapes are not compatible for Q @ c' 

' operation: {} vs {}'.format(a.shape, c.shape)) 

else: 

if M != c.shape[1]: 

raise ValueError('Array shapes are not compatible for c @ Q' 

' operation: {} vs {}'.format(c.shape, a.shape)) 

 

raw = qr(a, overwrite_a, None, "raw", pivoting) 

Q, tau = raw[0] 

 

gor_un_mqr, = get_lapack_funcs(('ormqr',), (Q,)) 

if gor_un_mqr.typecode in ('s', 'd'): 

trans = "T" 

else: 

trans = "C" 

 

Q = Q[:, :min(M, N)] 

if M > N and mode == "left" and not overwrite_c: 

if conjugate: 

cc = numpy.zeros((c.shape[1], M), dtype=c.dtype, order="F") 

cc[:, :N] = c.T 

else: 

cc = numpy.zeros((M, c.shape[1]), dtype=c.dtype, order="F") 

cc[:N, :] = c 

trans = "N" 

if conjugate: 

lr = "R" 

else: 

lr = "L" 

overwrite_c = True 

elif c.flags["C_CONTIGUOUS"] and trans == "T" or conjugate: 

cc = c.T 

if mode == "left": 

lr = "R" 

else: 

lr = "L" 

else: 

trans = "N" 

cc = c 

if mode == "left": 

lr = "L" 

else: 

lr = "R" 

cQ, = safecall(gor_un_mqr, "gormqr/gunmqr", lr, trans, Q, tau, cc, 

overwrite_c=overwrite_c) 

if trans != "N": 

cQ = cQ.T 

if mode == "right": 

cQ = cQ[:, :min(M, N)] 

if onedim: 

cQ = cQ.ravel() 

 

return (cQ,) + raw[1:] 

 

 

def rq(a, overwrite_a=False, lwork=None, mode='full', check_finite=True): 

""" 

Compute RQ decomposition of a matrix. 

 

Calculate the decomposition ``A = R Q`` where Q is unitary/orthogonal 

and R upper triangular. 

 

Parameters 

---------- 

a : (M, N) array_like 

Matrix to be decomposed 

overwrite_a : bool, optional 

Whether data in a is overwritten (may improve performance) 

lwork : int, optional 

Work array size, lwork >= a.shape[1]. If None or -1, an optimal size 

is computed. 

mode : {'full', 'r', 'economic'}, optional 

Determines what information is to be returned: either both Q and R 

('full', default), only R ('r') or both Q and R but computed in 

economy-size ('economic', see Notes). 

check_finite : bool, optional 

Whether to check that the input matrix contains only finite numbers. 

Disabling may give a performance gain, but may result in problems 

(crashes, non-termination) if the inputs do contain infinities or NaNs. 

 

Returns 

------- 

R : float or complex ndarray 

Of shape (M, N) or (M, K) for ``mode='economic'``. ``K = min(M, N)``. 

Q : float or complex ndarray 

Of shape (N, N) or (K, N) for ``mode='economic'``. Not returned 

if ``mode='r'``. 

 

Raises 

------ 

LinAlgError 

If decomposition fails. 

 

Notes 

----- 

This is an interface to the LAPACK routines sgerqf, dgerqf, cgerqf, zgerqf, 

sorgrq, dorgrq, cungrq and zungrq. 

 

If ``mode=economic``, the shapes of Q and R are (K, N) and (M, K) instead 

of (N,N) and (M,N), with ``K=min(M,N)``. 

 

Examples 

-------- 

>>> from scipy import linalg 

>>> a = np.random.randn(6, 9) 

>>> r, q = linalg.rq(a) 

>>> np.allclose(a, r @ q) 

True 

>>> r.shape, q.shape 

((6, 9), (9, 9)) 

>>> r2 = linalg.rq(a, mode='r') 

>>> np.allclose(r, r2) 

True 

>>> r3, q3 = linalg.rq(a, mode='economic') 

>>> r3.shape, q3.shape 

((6, 6), (6, 9)) 

 

""" 

if mode not in ['full', 'r', 'economic']: 

raise ValueError( 

"Mode argument should be one of ['full', 'r', 'economic']") 

 

if check_finite: 

a1 = numpy.asarray_chkfinite(a) 

else: 

a1 = numpy.asarray(a) 

if len(a1.shape) != 2: 

raise ValueError('expected matrix') 

M, N = a1.shape 

overwrite_a = overwrite_a or (_datacopied(a1, a)) 

 

gerqf, = get_lapack_funcs(('gerqf',), (a1,)) 

rq, tau = safecall(gerqf, 'gerqf', a1, lwork=lwork, 

overwrite_a=overwrite_a) 

if not mode == 'economic' or N < M: 

R = numpy.triu(rq, N-M) 

else: 

R = numpy.triu(rq[-M:, -M:]) 

 

if mode == 'r': 

return R 

 

gor_un_grq, = get_lapack_funcs(('orgrq',), (rq,)) 

 

if N < M: 

Q, = safecall(gor_un_grq, "gorgrq/gungrq", rq[-N:], tau, lwork=lwork, 

overwrite_a=1) 

elif mode == 'economic': 

Q, = safecall(gor_un_grq, "gorgrq/gungrq", rq, tau, lwork=lwork, 

overwrite_a=1) 

else: 

rq1 = numpy.empty((N, N), dtype=rq.dtype) 

rq1[-M:] = rq 

Q, = safecall(gor_un_grq, "gorgrq/gungrq", rq1, tau, lwork=lwork, 

overwrite_a=1) 

 

return R, Q