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

452

453

454

455

456

457

458

459

460

461

462

463

464

465

466

467

468

469

470

471

472

473

474

475

476

477

478

479

480

481

482

483

484

485

486

487

488

489

490

491

492

493

494

495

496

497

498

499

500

501

502

503

504

505

506

507

508

509

510

511

512

513

514

515

516

517

518

519

520

521

522

from __future__ import division, print_function, absolute_import 

import numpy as np 

import scipy.sparse as sps 

from ._numdiff import approx_derivative, group_columns 

from ._hessian_update_strategy import HessianUpdateStrategy 

from scipy.sparse.linalg import LinearOperator 

from copy import deepcopy 

 

 

FD_METHODS = ('2-point', '3-point', 'cs') 

 

 

class ScalarFunction(object): 

"""Scalar function and its derivatives. 

 

This class defines a scalar function F: R^n->R and methods for 

computing or approximating its first and second derivatives. 

 

Notes 

----- 

This class implements a memoization logic. There are methods `fun`, 

`grad`, hess` and corresponding attributes `f`, `g` and `H`. The following 

things should be considered: 

 

1. Use only public methods `fun`, `grad` and `hess`. 

2. After one of the methods is called, the corresponding attribute 

will be set. However, a subsequent call with a different argument 

of *any* of the methods may overwrite the attribute. 

""" 

def __init__(self, fun, x0, args, grad, hess, finite_diff_rel_step, 

finite_diff_bounds): 

if not callable(grad) and grad not in FD_METHODS: 

raise ValueError("`grad` must be either callable or one of {}." 

.format(FD_METHODS)) 

 

if not (callable(hess) or hess in FD_METHODS 

or isinstance(hess, HessianUpdateStrategy)): 

raise ValueError("`hess` must be either callable," 

"HessianUpdateStrategy or one of {}." 

.format(FD_METHODS)) 

 

if grad in FD_METHODS and hess in FD_METHODS: 

raise ValueError("Whenever the gradient is estimated via " 

"finite-differences, we require the Hessian " 

"to be estimated using one of the " 

"quasi-Newton strategies.") 

 

self.x = np.atleast_1d(x0).astype(float) 

self.n = self.x.size 

self.nfev = 0 

self.ngev = 0 

self.nhev = 0 

self.f_updated = False 

self.g_updated = False 

self.H_updated = False 

 

finite_diff_options = {} 

if grad in FD_METHODS: 

finite_diff_options["method"] = grad 

finite_diff_options["rel_step"] = finite_diff_rel_step 

finite_diff_options["bounds"] = finite_diff_bounds 

if hess in FD_METHODS: 

finite_diff_options["method"] = hess 

finite_diff_options["rel_step"] = finite_diff_rel_step 

finite_diff_options["as_linear_operator"] = True 

 

# Function evaluation 

def fun_wrapped(x): 

self.nfev += 1 

return fun(x, *args) 

 

def update_fun(): 

self.f = fun_wrapped(self.x) 

 

self._update_fun_impl = update_fun 

self._update_fun() 

 

# Gradient evaluation 

if callable(grad): 

def grad_wrapped(x): 

self.ngev += 1 

return np.atleast_1d(grad(x, *args)) 

 

def update_grad(): 

self.g = grad_wrapped(self.x) 

 

elif grad in FD_METHODS: 

def update_grad(): 

self._update_fun() 

self.g = approx_derivative(fun_wrapped, self.x, f0=self.f, 

**finite_diff_options) 

 

self._update_grad_impl = update_grad 

self._update_grad() 

 

# Hessian Evaluation 

if callable(hess): 

self.H = hess(x0, *args) 

self.H_updated = True 

self.nhev += 1 

 

if sps.issparse(self.H): 

def hess_wrapped(x): 

self.nhev += 1 

return sps.csr_matrix(hess(x, *args)) 

self.H = sps.csr_matrix(self.H) 

 

elif isinstance(self.H, LinearOperator): 

def hess_wrapped(x): 

self.nhev += 1 

return hess(x, *args) 

 

else: 

def hess_wrapped(x): 

self.nhev += 1 

return np.atleast_2d(np.asarray(hess(x, *args))) 

self.H = np.atleast_2d(np.asarray(self.H)) 

 

def update_hess(): 

self.H = hess_wrapped(self.x) 

 

elif hess in FD_METHODS: 

def update_hess(): 

self._update_grad() 

self.H = approx_derivative(grad_wrapped, self.x, f0=self.g, 

**finite_diff_options) 

return self.H 

 

update_hess() 

self.H_updated = True 

elif isinstance(hess, HessianUpdateStrategy): 

self.H = hess 

self.H.initialize(self.n, 'hess') 

self.H_updated = True 

self.x_prev = None 

self.g_prev = None 

 

def update_hess(): 

self._update_grad() 

self.H.update(self.x - self.x_prev, self.g - self.g_prev) 

 

self._update_hess_impl = update_hess 

 

if isinstance(hess, HessianUpdateStrategy): 

def update_x(x): 

self._update_grad() 

self.x_prev = self.x 

self.g_prev = self.g 

 

self.x = x 

self.f_updated = False 

self.g_updated = False 

self.H_updated = False 

self._update_hess() 

else: 

def update_x(x): 

self.x = x 

self.f_updated = False 

self.g_updated = False 

self.H_updated = False 

self._update_x_impl = update_x 

 

def _update_fun(self): 

if not self.f_updated: 

self._update_fun_impl() 

self.f_updated = True 

 

def _update_grad(self): 

if not self.g_updated: 

self._update_grad_impl() 

self.g_updated = True 

 

def _update_hess(self): 

if not self.H_updated: 

self._update_hess_impl() 

self.H_updated = True 

 

def fun(self, x): 

if not np.array_equal(x, self.x): 

self._update_x_impl(x) 

self._update_fun() 

return self.f 

 

def grad(self, x): 

if not np.array_equal(x, self.x): 

self._update_x_impl(x) 

self._update_grad() 

return self.g 

 

def hess(self, x): 

if not np.array_equal(x, self.x): 

self._update_x_impl(x) 

self._update_hess() 

return self.H 

 

 

class VectorFunction(object): 

"""Vector function and its derivatives. 

 

This class defines a vector function F: R^n->R^m and methods for 

computing or approximating its first and second derivatives. 

 

Notes 

----- 

This class implements a memoization logic. There are methods `fun`, 

`jac`, hess` and corresponding attributes `f`, `J` and `H`. The following 

things should be considered: 

 

1. Use only public methods `fun`, `jac` and `hess`. 

2. After one of the methods is called, the corresponding attribute 

will be set. However, a subsequent call with a different argument 

of *any* of the methods may overwrite the attribute. 

""" 

def __init__(self, fun, x0, jac, hess, 

finite_diff_rel_step, finite_diff_jac_sparsity, 

finite_diff_bounds, sparse_jacobian): 

if not callable(jac) and jac not in FD_METHODS: 

raise ValueError("`jac` must be either callable or one of {}." 

.format(FD_METHODS)) 

 

if not (callable(hess) or hess in FD_METHODS 

or isinstance(hess, HessianUpdateStrategy)): 

raise ValueError("`hess` must be either callable," 

"HessianUpdateStrategy or one of {}." 

.format(FD_METHODS)) 

 

if jac in FD_METHODS and hess in FD_METHODS: 

raise ValueError("Whenever the Jacobian is estimated via " 

"finite-differences, we require the Hessian to " 

"be estimated using one of the quasi-Newton " 

"strategies.") 

 

self.x = np.atleast_1d(x0).astype(float) 

self.n = self.x.size 

self.nfev = 0 

self.njev = 0 

self.nhev = 0 

self.f_updated = False 

self.J_updated = False 

self.H_updated = False 

 

finite_diff_options = {} 

if jac in FD_METHODS: 

finite_diff_options["method"] = jac 

finite_diff_options["rel_step"] = finite_diff_rel_step 

if finite_diff_jac_sparsity is not None: 

sparsity_groups = group_columns(finite_diff_jac_sparsity) 

finite_diff_options["sparsity"] = (finite_diff_jac_sparsity, 

sparsity_groups) 

finite_diff_options["bounds"] = finite_diff_bounds 

self.x_diff = np.copy(self.x) 

if hess in FD_METHODS: 

finite_diff_options["method"] = hess 

finite_diff_options["rel_step"] = finite_diff_rel_step 

finite_diff_options["as_linear_operator"] = True 

self.x_diff = np.copy(self.x) 

if jac in FD_METHODS and hess in FD_METHODS: 

raise ValueError("Whenever the Jacobian is estimated via " 

"finite-differences, we require the Hessian to " 

"be estimated using one of the quasi-Newton " 

"strategies.") 

 

# Function evaluation 

def fun_wrapped(x): 

self.nfev += 1 

return np.atleast_1d(fun(x)) 

 

def update_fun(): 

self.f = fun_wrapped(self.x) 

 

self._update_fun_impl = update_fun 

update_fun() 

 

self.v = np.zeros_like(self.f) 

self.m = self.v.size 

 

# Jacobian Evaluation 

if callable(jac): 

self.J = jac(self.x) 

self.J_updated = True 

self.njev += 1 

 

if (sparse_jacobian or 

sparse_jacobian is None and sps.issparse(self.J)): 

def jac_wrapped(x): 

self.njev += 1 

return sps.csr_matrix(jac(x)) 

self.J = sps.csr_matrix(self.J) 

self.sparse_jacobian = True 

 

elif sps.issparse(self.J): 

def jac_wrapped(x): 

self.njev += 1 

return jac(x).toarray() 

self.J = self.J.toarray() 

self.sparse_jacobian = False 

 

else: 

def jac_wrapped(x): 

self.njev += 1 

return np.atleast_2d(jac(x)) 

self.J = np.atleast_2d(self.J) 

self.sparse_jacobian = False 

 

def update_jac(): 

self.J = jac_wrapped(self.x) 

 

elif jac in FD_METHODS: 

self.J = approx_derivative(fun_wrapped, self.x, f0=self.f, 

**finite_diff_options) 

self.J_updated = True 

 

if (sparse_jacobian or 

sparse_jacobian is None and sps.issparse(self.J)): 

def update_jac(): 

self._update_fun() 

self.J = sps.csr_matrix( 

approx_derivative(fun_wrapped, self.x, f0=self.f, 

**finite_diff_options)) 

self.J = sps.csr_matrix(self.J) 

self.sparse_jacobian = True 

 

elif sps.issparse(self.J): 

def update_jac(): 

self._update_fun() 

self.J = approx_derivative(fun_wrapped, self.x, f0=self.f, 

**finite_diff_options).toarray() 

self.J = self.J.toarray() 

self.sparse_jacobian = False 

 

else: 

def update_jac(): 

self._update_fun() 

self.J = np.atleast_2d( 

approx_derivative(fun_wrapped, self.x, f0=self.f, 

**finite_diff_options)) 

self.J = np.atleast_2d(self.J) 

self.sparse_jacobian = False 

 

self._update_jac_impl = update_jac 

 

# Define Hessian 

if callable(hess): 

self.H = hess(self.x, self.v) 

self.H_updated = True 

self.nhev += 1 

 

if sps.issparse(self.H): 

def hess_wrapped(x, v): 

self.nhev += 1 

return sps.csr_matrix(hess(x, v)) 

self.H = sps.csr_matrix(self.H) 

 

elif isinstance(self.H, LinearOperator): 

def hess_wrapped(x, v): 

self.nhev += 1 

return hess(x, v) 

 

else: 

def hess_wrapped(x, v): 

self.nhev += 1 

return np.atleast_2d(np.asarray(hess(x, v))) 

self.H = np.atleast_2d(np.asarray(self.H)) 

 

def update_hess(): 

self.H = hess_wrapped(self.x, self.v) 

elif hess in FD_METHODS: 

def jac_dot_v(x, v): 

return jac_wrapped(x).T.dot(v) 

 

def update_hess(): 

self._update_jac() 

self.H = approx_derivative(jac_dot_v, self.x, 

f0=self.J.T.dot(self.v), 

args=(self.v,), 

**finite_diff_options) 

update_hess() 

self.H_updated = True 

elif isinstance(hess, HessianUpdateStrategy): 

self.H = hess 

self.H.initialize(self.n, 'hess') 

self.H_updated = True 

self.x_prev = None 

self.J_prev = None 

 

def update_hess(): 

self._update_jac() 

# When v is updated before x was updated, then x_prev and 

# J_prev are None and we need this check. 

if self.x_prev is not None and self.J_prev is not None: 

delta_x = self.x - self.x_prev 

delta_g = self.J.T.dot(self.v) - self.J_prev.T.dot(self.v) 

self.H.update(delta_x, delta_g) 

 

self._update_hess_impl = update_hess 

 

if isinstance(hess, HessianUpdateStrategy): 

def update_x(x): 

self._update_jac() 

self.x_prev = self.x 

self.J_prev = self.J 

self.x = x 

self.f_updated = False 

self.J_updated = False 

self.H_updated = False 

self._update_hess() 

else: 

def update_x(x): 

self.x = x 

self.f_updated = False 

self.J_updated = False 

self.H_updated = False 

 

self._update_x_impl = update_x 

 

def _update_v(self, v): 

if not np.array_equal(v, self.v): 

self.v = v 

self.H_updated = False 

 

def _update_x(self, x): 

if not np.array_equal(x, self.x): 

self._update_x_impl(x) 

 

def _update_fun(self): 

if not self.f_updated: 

self._update_fun_impl() 

self.f_updated = True 

 

def _update_jac(self): 

if not self.J_updated: 

self._update_jac_impl() 

self.J_updated = True 

 

def _update_hess(self): 

if not self.H_updated: 

self._update_hess_impl() 

self.H_updated = True 

 

def fun(self, x): 

self._update_x(x) 

self._update_fun() 

return self.f 

 

def jac(self, x): 

self._update_x(x) 

self._update_jac() 

return self.J 

 

def hess(self, x, v): 

# v should be updated before x. 

self._update_v(v) 

self._update_x(x) 

self._update_hess() 

return self.H 

 

 

class LinearVectorFunction(object): 

"""Linear vector function and its derivatives. 

 

Defines a linear function F = A x, where x is n-dimensional vector and 

A is m-by-n matrix. The Jacobian is constant and equals to A. The Hessian 

is identically zero and it is returned as a csr matrix. 

""" 

def __init__(self, A, x0, sparse_jacobian): 

if sparse_jacobian or sparse_jacobian is None and sps.issparse(A): 

self.J = sps.csr_matrix(A) 

self.sparse_jacobian = True 

elif sps.issparse(A): 

self.J = A.toarray() 

self.sparse_jacobian = False 

else: 

self.J = np.atleast_2d(A) 

self.sparse_jacobian = False 

 

self.m, self.n = self.J.shape 

 

self.x = np.atleast_1d(x0).astype(float) 

self.f = self.J.dot(self.x) 

self.f_updated = True 

 

self.v = np.zeros(self.m, dtype=float) 

self.H = sps.csr_matrix((self.n, self.n)) 

 

def _update_x(self, x): 

if not np.array_equal(x, self.x): 

self.x = x 

self.f_updated = False 

 

def fun(self, x): 

self._update_x(x) 

if not self.f_updated: 

self.f = self.J.dot(x) 

self.f_updated = True 

return self.f 

 

def jac(self, x): 

self._update_x(x) 

return self.J 

 

def hess(self, x, v): 

self._update_x(x) 

self.v = v 

return self.H 

 

 

class IdentityVectorFunction(LinearVectorFunction): 

"""Identity vector function and its derivatives. 

 

The Jacobian is the identity matrix, returned as a dense array when 

`sparse_jacobian=False` and as a csr matrix otherwise. The Hessian is 

identically zero and it is returned as a csr matrix. 

""" 

def __init__(self, x0, sparse_jacobian): 

n = len(x0) 

if sparse_jacobian or sparse_jacobian is None: 

A = sps.eye(n, format='csr') 

sparse_jacobian = True 

else: 

A = np.eye(n) 

sparse_jacobian = False 

super(IdentityVectorFunction, self).__init__(A, x0, sparse_jacobian)