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

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

""" 

Routines for removing redundant (linearly dependent) equations from linear 

programming equality constraints. 

""" 

# Author: Matt Haberland 

 

from __future__ import division, print_function, absolute_import 

import numpy as np 

from scipy.linalg import svd 

import scipy 

 

 

def _row_count(A): 

""" 

Counts the number of nonzeros in each row of input array A. 

Nonzeros are defined as any element with absolute value greater than 

tol = 1e-13. This value should probably be an input to the function. 

 

Parameters 

---------- 

A : 2-D array 

An array representing a matrix 

 

Returns 

------- 

rowcount : 1-D array 

Number of nonzeros in each row of A 

 

""" 

tol = 1e-13 

return np.array((abs(A) > tol).sum(axis=1)).flatten() 

 

 

def _get_densest(A, eligibleRows): 

""" 

Returns the index of the densest row of A. Ignores rows that are not 

eligible for consideration. 

 

Parameters 

---------- 

A : 2-D array 

An array representing a matrix 

eligibleRows : 1-D logical array 

Values indicate whether the corresponding row of A is eligible 

to be considered 

 

Returns 

------- 

i_densest : int 

Index of the densest row in A eligible for consideration 

 

""" 

rowCounts = _row_count(A) 

return np.argmax(rowCounts * eligibleRows) 

 

 

def _remove_zero_rows(A, b): 

""" 

Eliminates trivial equations from system of equations defined by Ax = b 

and identifies trivial infeasibilities 

 

Parameters 

---------- 

A : 2-D array 

An array representing the left-hand side of a system of equations 

b : 1-D array 

An array representing the right-hand side of a system of equations 

 

Returns 

------- 

A : 2-D array 

An array representing the left-hand side of a system of equations 

b : 1-D array 

An array representing the right-hand side of a system of equations 

status: int 

An integer indicating the status of the removal operation 

0: No infeasibility identified 

2: Trivially infeasible 

message : str 

A string descriptor of the exit status of the optimization. 

 

""" 

status = 0 

message = "" 

i_zero = _row_count(A) == 0 

A = A[np.logical_not(i_zero), :] 

if not(np.allclose(b[i_zero], 0)): 

status = 2 

message = "There is a zero row in A_eq with a nonzero corresponding " \ 

"entry in b_eq. The problem is infeasible." 

b = b[np.logical_not(i_zero)] 

return A, b, status, message 

 

 

def bg_update_dense(plu, perm_r, v, j): 

LU, p = plu 

 

u = scipy.linalg.solve_triangular(LU, v[perm_r], lower=True, 

unit_diagonal=True) 

LU[:j+1, j] = u[:j+1] 

l = u[j+1:] 

piv = LU[j, j] 

LU[j+1:, j] += (l/piv) 

return LU, p 

 

 

def _remove_redundancy_dense(A, rhs): 

""" 

Eliminates redundant equations from system of equations defined by Ax = b 

and identifies infeasibilities. 

 

Parameters 

---------- 

A : 2-D sparse matrix 

An matrix representing the left-hand side of a system of equations 

rhs : 1-D array 

An array representing the right-hand side of a system of equations 

 

Returns 

---------- 

A : 2-D sparse matrix 

A matrix representing the left-hand side of a system of equations 

rhs : 1-D array 

An array representing the right-hand side of a system of equations 

status: int 

An integer indicating the status of the system 

0: No infeasibility identified 

2: Trivially infeasible 

message : str 

A string descriptor of the exit status of the optimization. 

 

References 

---------- 

.. [2] Andersen, Erling D. "Finding all linearly dependent rows in 

large-scale linear programming." Optimization Methods and Software 

6.3 (1995): 219-227. 

 

""" 

tolapiv = 1e-8 

tolprimal = 1e-8 

status = 0 

message = "" 

inconsistent = ("There is a linear combination of rows of A_eq that " 

"results in zero, suggesting a redundant constraint. " 

"However the same linear combination of b_eq is " 

"nonzero, suggesting that the constraints conflict " 

"and the problem is infeasible.") 

A, rhs, status, message = _remove_zero_rows(A, rhs) 

 

if status != 0: 

return A, rhs, status, message 

 

m, n = A.shape 

 

v = list(range(m)) # Artificial column indices. 

b = list(v) # Basis column indices. 

# This is better as a list than a set because column order of basis matrix 

# needs to be consistent. 

k = set(range(m, m+n)) # Structural column indices. 

d = [] # Indices of dependent rows 

lu = None 

perm_r = None 

 

A_orig = A 

A = np.hstack((np.eye(m), A)) 

e = np.zeros(m) 

 

# Implements basic algorithm from [2] 

# Uses some of the suggested improvements (removing zero rows and 

# Bartels-Golub update idea). 

# Removing column singletons would be easy, but it is not as important 

# because the procedure is performed only on the equality constraint 

# matrix from the original problem - not on the canonical form matrix, 

# which would have many more column singletons due to slack variables 

# from the inequality constraints. 

# The thoughts on "crashing" the initial basis sound useful, but the 

# description of the procedure seems to assume a lot of familiarity with 

# the subject; it is not very explicit. I already went through enough 

# trouble getting the basic algorithm working, so I was not interested in 

# trying to decipher this, too. (Overall, the paper is fraught with 

# mistakes and ambiguities - which is strange, because the rest of 

# Andersen's papers are quite good.) 

 

B = A[:, b] 

for i in v: 

 

e[i] = 1 

if i > 0: 

e[i-1] = 0 

 

try: # fails for i==0 and any time it gets ill-conditioned 

j = b[i-1] 

lu = bg_update_dense(lu, perm_r, A[:, j], i-1) 

except: 

lu = scipy.linalg.lu_factor(B) 

LU, p = lu 

perm_r = list(range(m)) 

for i1, i2 in enumerate(p): 

perm_r[i1], perm_r[i2] = perm_r[i2], perm_r[i1] 

 

pi = scipy.linalg.lu_solve(lu, e, trans=1) 

 

# not efficient, but this is not the time sink... 

js = np.array(list(k-set(b))) 

batch = 50 

dependent = True 

 

# This is a tiny bit faster than looping over columns indivually, 

# like for j in js: if abs(A[:,j].transpose().dot(pi)) > tolapiv: 

for j_index in range(0, len(js), batch): 

j_indices = js[np.arange(j_index, min(j_index+batch, len(js)))] 

 

c = abs(A[:, j_indices].transpose().dot(pi)) 

if (c > tolapiv).any(): 

j = js[j_index + np.argmax(c)] # very independent column 

B[:, i] = A[:, j] 

b[i] = j 

dependent = False 

break 

if dependent: 

bibar = pi.T.dot(rhs.reshape(-1, 1)) 

bnorm = np.linalg.norm(rhs) 

if abs(bibar)/(1+bnorm) > tolprimal: # inconsistent 

status = 2 

message = inconsistent 

return A_orig, rhs, status, message 

else: # dependent 

d.append(i) 

 

keep = set(range(m)) 

keep = list(keep - set(d)) 

return A_orig[keep, :], rhs[keep], status, message 

 

 

def _remove_redundancy_sparse(A, rhs): 

""" 

Eliminates redundant equations from system of equations defined by Ax = b 

and identifies infeasibilities. 

 

Parameters 

---------- 

A : 2-D sparse matrix 

An matrix representing the left-hand side of a system of equations 

rhs : 1-D array 

An array representing the right-hand side of a system of equations 

 

Returns 

------- 

A : 2-D sparse matrix 

A matrix representing the left-hand side of a system of equations 

rhs : 1-D array 

An array representing the right-hand side of a system of equations 

status: int 

An integer indicating the status of the system 

0: No infeasibility identified 

2: Trivially infeasible 

message : str 

A string descriptor of the exit status of the optimization. 

 

References 

---------- 

.. [2] Andersen, Erling D. "Finding all linearly dependent rows in 

large-scale linear programming." Optimization Methods and Software 

6.3 (1995): 219-227. 

 

""" 

 

tolapiv = 1e-8 

tolprimal = 1e-8 

status = 0 

message = "" 

inconsistent = ("There is a linear combination of rows of A_eq that " 

"results in zero, suggesting a redundant constraint. " 

"However the same linear combination of b_eq is " 

"nonzero, suggesting that the constraints conflict " 

"and the problem is infeasible.") 

A, rhs, status, message = _remove_zero_rows(A, rhs) 

 

if status != 0: 

return A, rhs, status, message 

 

m, n = A.shape 

 

v = list(range(m)) # Artificial column indices. 

b = list(v) # Basis column indices. 

# This is better as a list than a set because column order of basis matrix 

# needs to be consistent. 

k = set(range(m, m+n)) # Structural column indices. 

d = [] # Indices of dependent rows 

 

A_orig = A 

A = scipy.sparse.hstack((scipy.sparse.eye(m), A)).tocsc() 

e = np.zeros(m) 

 

# Implements basic algorithm from [2] 

# Uses only one of the suggested improvements (removing zero rows). 

# Removing column singletons would be easy, but it is not as important 

# because the procedure is performed only on the equality constraint 

# matrix from the original problem - not on the canonical form matrix, 

# which would have many more column singletons due to slack variables 

# from the inequality constraints. 

# The thoughts on "crashing" the initial basis sound useful, but the 

# description of the procedure seems to assume a lot of familiarity with 

# the subject; it is not very explicit. I already went through enough 

# trouble getting the basic algorithm working, so I was not interested in 

# trying to decipher this, too. (Overall, the paper is fraught with 

# mistakes and ambiguities - which is strange, because the rest of 

# Andersen's papers are quite good.) 

# I tried and tried and tried to improve performance using the 

# Bartels-Golub update. It works, but it's only practical if the LU 

# factorization can be specialized as described, and that is not possible 

# until the Scipy SuperLU interface permits control over column 

# permutation - see issue #7700. 

 

for i in v: 

B = A[:, b] 

 

e[i] = 1 

if i > 0: 

e[i-1] = 0 

 

pi = scipy.sparse.linalg.spsolve(B.transpose(), e).reshape(-1, 1) 

 

js = list(k-set(b)) # not efficient, but this is not the time sink... 

 

# Due to overhead, it tends to be faster (for problems tested) to 

# compute the full matrix-vector product rather than individual 

# vector-vector products (with the chance of terminating as soon 

# as any are nonzero). For very large matrices, it might be worth 

# it to compute, say, 100 or 1000 at a time and stop when a nonzero 

# is found. 

 

c = (np.abs(A[:, js].transpose().dot(pi)) > tolapiv).nonzero()[0] 

if len(c) > 0: # independent 

j = js[c[0]] 

# in a previous commit, the previous line was changed to choose 

# index j corresponding with the maximum dot product. 

# While this avoided issues with almost 

# singular matrices, it slowed the routine in most NETLIB tests. 

# I think this is because these columns were denser than the 

# first column with nonzero dot product (c[0]). 

# It would be nice to have a heuristic that balances sparsity with 

# high dot product, but I don't think it's worth the time to 

# develop one right now. Bartels-Golub update is a much higher 

# priority. 

b[i] = j # replace artificial column 

else: 

bibar = pi.T.dot(rhs.reshape(-1, 1)) 

bnorm = np.linalg.norm(rhs) 

if abs(bibar)/(1 + bnorm) > tolprimal: 

status = 2 

message = inconsistent 

return A_orig, rhs, status, message 

else: # dependent 

d.append(i) 

 

keep = set(range(m)) 

keep = list(keep - set(d)) 

return A_orig[keep, :], rhs[keep], status, message 

 

 

def _remove_redundancy(A, b): 

""" 

Eliminates redundant equations from system of equations defined by Ax = b 

and identifies infeasibilities. 

 

Parameters 

---------- 

A : 2-D array 

An array representing the left-hand side of a system of equations 

b : 1-D array 

An array representing the right-hand side of a system of equations 

 

Returns 

------- 

A : 2-D array 

An array representing the left-hand side of a system of equations 

b : 1-D array 

An array representing the right-hand side of a system of equations 

status: int 

An integer indicating the status of the system 

0: No infeasibility identified 

2: Trivially infeasible 

message : str 

A string descriptor of the exit status of the optimization. 

 

References 

---------- 

.. [2] Andersen, Erling D. "Finding all linearly dependent rows in 

large-scale linear programming." Optimization Methods and Software 

6.3 (1995): 219-227. 

 

""" 

 

A, b, status, message = _remove_zero_rows(A, b) 

 

if status != 0: 

return A, b, status, message 

 

U, s, Vh = svd(A) 

eps = np.finfo(float).eps 

tol = s.max() * max(A.shape) * eps 

 

m, n = A.shape 

s_min = s[-1] if m <= n else 0 

 

# this algorithm is faster than that of [2] when the nullspace is small 

# but it could probably be improvement by randomized algorithms and with 

# a sparse implementation. 

# it relies on repeated singular value decomposition to find linearly 

# dependent rows (as identified by columns of U that correspond with zero 

# singular values). Unfortunately, only one row can be removed per 

# decomposition (I tried otherwise; doing so can cause problems.) 

# It would be nice if we could do truncated SVD like sp.sparse.linalg.svds 

# but that function is unreliable at finding singular values near zero. 

# Finding max eigenvalue L of A A^T, then largest eigenvalue (and 

# associated eigenvector) of -A A^T + L I (I is identity) via power 

# iteration would also work in theory, but is only efficient if the 

# smallest nonzero eigenvalue of A A^T is close to the largest nonzero 

# eigenvalue. 

 

while abs(s_min) < tol: 

v = U[:, -1] # TODO: return these so user can eliminate from problem? 

# rows need to be represented in significant amount 

eligibleRows = np.abs(v) > tol * 10e6 

if not np.any(eligibleRows) or np.any(np.abs(v.dot(A)) > tol): 

status = 4 

message = ("Due to numerical issues, redundant equality " 

"constraints could not be removed automatically. " 

"Try providing your constraint matrices as sparse " 

"matrices to activate sparse presolve, try turning " 

"off redundancy removal, or try turning off presolve " 

"altogether.") 

break 

if np.any(np.abs(v.dot(b)) > tol): 

status = 2 

message = ("There is a linear combination of rows of A_eq that " 

"results in zero, suggesting a redundant constraint. " 

"However the same linear combination of b_eq is " 

"nonzero, suggesting that the constraints conflict " 

"and the problem is infeasible.") 

break 

 

i_remove = _get_densest(A, eligibleRows) 

A = np.delete(A, i_remove, axis=0) 

b = np.delete(b, i_remove) 

U, s, Vh = svd(A) 

m, n = A.shape 

s_min = s[-1] if m <= n else 0 

 

return A, b, status, message