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

# TNC Python interface 

# @(#) $Jeannot: tnc.py,v 1.11 2005/01/28 18:27:31 js Exp $ 

 

# Copyright (c) 2004-2005, Jean-Sebastien Roy (js@jeannot.org) 

 

# Permission is hereby granted, free of charge, to any person obtaining a 

# copy of this software and associated documentation files (the 

# "Software"), to deal in the Software without restriction, including 

# without limitation the rights to use, copy, modify, merge, publish, 

# distribute, sublicense, and/or sell copies of the Software, and to 

# permit persons to whom the Software is furnished to do so, subject to 

# the following conditions: 

 

# The above copyright notice and this permission notice shall be included 

# in all copies or substantial portions of the Software. 

 

# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 

# OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 

# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 

# IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY 

# CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 

# TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 

# SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 

 

""" 

TNC: A python interface to the TNC non-linear optimizer 

 

TNC is a non-linear optimizer. To use it, you must provide a function to 

minimize. The function must take one argument: the list of coordinates where to 

evaluate the function; and it must return either a tuple, whose first element is the 

value of the function, and whose second argument is the gradient of the function 

(as a list of values); or None, to abort the minimization. 

""" 

 

from __future__ import division, print_function, absolute_import 

 

from scipy.optimize import moduleTNC, approx_fprime 

from .optimize import MemoizeJac, OptimizeResult, _check_unknown_options 

from numpy import inf, array, zeros, asfarray 

 

__all__ = ['fmin_tnc'] 

 

 

MSG_NONE = 0 # No messages 

MSG_ITER = 1 # One line per iteration 

MSG_INFO = 2 # Informational messages 

MSG_VERS = 4 # Version info 

MSG_EXIT = 8 # Exit reasons 

MSG_ALL = MSG_ITER + MSG_INFO + MSG_VERS + MSG_EXIT 

 

MSGS = { 

MSG_NONE: "No messages", 

MSG_ITER: "One line per iteration", 

MSG_INFO: "Informational messages", 

MSG_VERS: "Version info", 

MSG_EXIT: "Exit reasons", 

MSG_ALL: "All messages" 

} 

 

INFEASIBLE = -1 # Infeasible (lower bound > upper bound) 

LOCALMINIMUM = 0 # Local minimum reached (|pg| ~= 0) 

FCONVERGED = 1 # Converged (|f_n-f_(n-1)| ~= 0) 

XCONVERGED = 2 # Converged (|x_n-x_(n-1)| ~= 0) 

MAXFUN = 3 # Max. number of function evaluations reached 

LSFAIL = 4 # Linear search failed 

CONSTANT = 5 # All lower bounds are equal to the upper bounds 

NOPROGRESS = 6 # Unable to progress 

USERABORT = 7 # User requested end of minimization 

 

RCSTRINGS = { 

INFEASIBLE: "Infeasible (lower bound > upper bound)", 

LOCALMINIMUM: "Local minimum reached (|pg| ~= 0)", 

FCONVERGED: "Converged (|f_n-f_(n-1)| ~= 0)", 

XCONVERGED: "Converged (|x_n-x_(n-1)| ~= 0)", 

MAXFUN: "Max. number of function evaluations reached", 

LSFAIL: "Linear search failed", 

CONSTANT: "All lower bounds are equal to the upper bounds", 

NOPROGRESS: "Unable to progress", 

USERABORT: "User requested end of minimization" 

} 

 

# Changes to interface made by Travis Oliphant, Apr. 2004 for inclusion in 

# SciPy 

 

 

def fmin_tnc(func, x0, fprime=None, args=(), approx_grad=0, 

bounds=None, epsilon=1e-8, scale=None, offset=None, 

messages=MSG_ALL, maxCGit=-1, maxfun=None, eta=-1, 

stepmx=0, accuracy=0, fmin=0, ftol=-1, xtol=-1, pgtol=-1, 

rescale=-1, disp=None, callback=None): 

""" 

Minimize a function with variables subject to bounds, using 

gradient information in a truncated Newton algorithm. This 

method wraps a C implementation of the algorithm. 

 

Parameters 

---------- 

func : callable ``func(x, *args)`` 

Function to minimize. Must do one of: 

 

1. Return f and g, where f is the value of the function and g its 

gradient (a list of floats). 

 

2. Return the function value but supply gradient function 

separately as `fprime`. 

 

3. Return the function value and set ``approx_grad=True``. 

 

If the function returns None, the minimization 

is aborted. 

x0 : array_like 

Initial estimate of minimum. 

fprime : callable ``fprime(x, *args)``, optional 

Gradient of `func`. If None, then either `func` must return the 

function value and the gradient (``f,g = func(x, *args)``) 

or `approx_grad` must be True. 

args : tuple, optional 

Arguments to pass to function. 

approx_grad : bool, optional 

If true, approximate the gradient numerically. 

bounds : list, optional 

(min, max) pairs for each element in x0, defining the 

bounds on that parameter. Use None or +/-inf for one of 

min or max when there is no bound in that direction. 

epsilon : float, optional 

Used if approx_grad is True. The stepsize in a finite 

difference approximation for fprime. 

scale : array_like, optional 

Scaling factors to apply to each variable. If None, the 

factors are up-low for interval bounded variables and 

1+|x| for the others. Defaults to None. 

offset : array_like, optional 

Value to subtract from each variable. If None, the 

offsets are (up+low)/2 for interval bounded variables 

and x for the others. 

messages : int, optional 

Bit mask used to select messages display during 

minimization values defined in the MSGS dict. Defaults to 

MGS_ALL. 

disp : int, optional 

Integer interface to messages. 0 = no message, 5 = all messages 

maxCGit : int, optional 

Maximum number of hessian*vector evaluations per main 

iteration. If maxCGit == 0, the direction chosen is 

-gradient if maxCGit < 0, maxCGit is set to 

max(1,min(50,n/2)). Defaults to -1. 

maxfun : int, optional 

Maximum number of function evaluation. if None, maxfun is 

set to max(100, 10*len(x0)). Defaults to None. 

eta : float, optional 

Severity of the line search. if < 0 or > 1, set to 0.25. 

Defaults to -1. 

stepmx : float, optional 

Maximum step for the line search. May be increased during 

call. If too small, it will be set to 10.0. Defaults to 0. 

accuracy : float, optional 

Relative precision for finite difference calculations. If 

<= machine_precision, set to sqrt(machine_precision). 

Defaults to 0. 

fmin : float, optional 

Minimum function value estimate. Defaults to 0. 

ftol : float, optional 

Precision goal for the value of f in the stopping criterion. 

If ftol < 0.0, ftol is set to 0.0 defaults to -1. 

xtol : float, optional 

Precision goal for the value of x in the stopping 

criterion (after applying x scaling factors). If xtol < 

0.0, xtol is set to sqrt(machine_precision). Defaults to 

-1. 

pgtol : float, optional 

Precision goal for the value of the projected gradient in 

the stopping criterion (after applying x scaling factors). 

If pgtol < 0.0, pgtol is set to 1e-2 * sqrt(accuracy). 

Setting it to 0.0 is not recommended. Defaults to -1. 

rescale : float, optional 

Scaling factor (in log10) used to trigger f value 

rescaling. If 0, rescale at each iteration. If a large 

value, never rescale. If < 0, rescale is set to 1.3. 

callback : callable, optional 

Called after each iteration, as callback(xk), where xk is the 

current parameter vector. 

 

Returns 

------- 

x : ndarray 

The solution. 

nfeval : int 

The number of function evaluations. 

rc : int 

Return code, see below 

 

See also 

-------- 

minimize: Interface to minimization algorithms for multivariate 

functions. See the 'TNC' `method` in particular. 

 

Notes 

----- 

The underlying algorithm is truncated Newton, also called 

Newton Conjugate-Gradient. This method differs from 

scipy.optimize.fmin_ncg in that 

 

1. It wraps a C implementation of the algorithm 

2. It allows each variable to be given an upper and lower bound. 

 

The algorithm incorporates the bound constraints by determining 

the descent direction as in an unconstrained truncated Newton, 

but never taking a step-size large enough to leave the space 

of feasible x's. The algorithm keeps track of a set of 

currently active constraints, and ignores them when computing 

the minimum allowable step size. (The x's associated with the 

active constraint are kept fixed.) If the maximum allowable 

step size is zero then a new constraint is added. At the end 

of each iteration one of the constraints may be deemed no 

longer active and removed. A constraint is considered 

no longer active is if it is currently active 

but the gradient for that variable points inward from the 

constraint. The specific constraint removed is the one 

associated with the variable of largest index whose 

constraint is no longer active. 

 

Return codes are defined as follows:: 

 

-1 : Infeasible (lower bound > upper bound) 

0 : Local minimum reached (|pg| ~= 0) 

1 : Converged (|f_n-f_(n-1)| ~= 0) 

2 : Converged (|x_n-x_(n-1)| ~= 0) 

3 : Max. number of function evaluations reached 

4 : Linear search failed 

5 : All lower bounds are equal to the upper bounds 

6 : Unable to progress 

7 : User requested end of minimization 

 

References 

---------- 

Wright S., Nocedal J. (2006), 'Numerical Optimization' 

 

Nash S.G. (1984), "Newton-Type Minimization Via the Lanczos Method", 

SIAM Journal of Numerical Analysis 21, pp. 770-778 

 

""" 

# handle fprime/approx_grad 

if approx_grad: 

fun = func 

jac = None 

elif fprime is None: 

fun = MemoizeJac(func) 

jac = fun.derivative 

else: 

fun = func 

jac = fprime 

 

if disp is not None: # disp takes precedence over messages 

mesg_num = disp 

else: 

mesg_num = {0:MSG_NONE, 1:MSG_ITER, 2:MSG_INFO, 3:MSG_VERS, 

4:MSG_EXIT, 5:MSG_ALL}.get(messages, MSG_ALL) 

# build options 

opts = {'eps': epsilon, 

'scale': scale, 

'offset': offset, 

'mesg_num': mesg_num, 

'maxCGit': maxCGit, 

'maxiter': maxfun, 

'eta': eta, 

'stepmx': stepmx, 

'accuracy': accuracy, 

'minfev': fmin, 

'ftol': ftol, 

'xtol': xtol, 

'gtol': pgtol, 

'rescale': rescale, 

'disp': False} 

 

res = _minimize_tnc(fun, x0, args, jac, bounds, callback=callback, **opts) 

 

return res['x'], res['nfev'], res['status'] 

 

 

def _minimize_tnc(fun, x0, args=(), jac=None, bounds=None, 

eps=1e-8, scale=None, offset=None, mesg_num=None, 

maxCGit=-1, maxiter=None, eta=-1, stepmx=0, accuracy=0, 

minfev=0, ftol=-1, xtol=-1, gtol=-1, rescale=-1, disp=False, 

callback=None, **unknown_options): 

""" 

Minimize a scalar function of one or more variables using a truncated 

Newton (TNC) algorithm. 

 

Options 

------- 

eps : float 

Step size used for numerical approximation of the jacobian. 

scale : list of floats 

Scaling factors to apply to each variable. If None, the 

factors are up-low for interval bounded variables and 

1+|x] fo the others. Defaults to None 

offset : float 

Value to subtract from each variable. If None, the 

offsets are (up+low)/2 for interval bounded variables 

and x for the others. 

disp : bool 

Set to True to print convergence messages. 

maxCGit : int 

Maximum number of hessian*vector evaluations per main 

iteration. If maxCGit == 0, the direction chosen is 

-gradient if maxCGit < 0, maxCGit is set to 

max(1,min(50,n/2)). Defaults to -1. 

maxiter : int 

Maximum number of function evaluation. if None, `maxiter` is 

set to max(100, 10*len(x0)). Defaults to None. 

eta : float 

Severity of the line search. if < 0 or > 1, set to 0.25. 

Defaults to -1. 

stepmx : float 

Maximum step for the line search. May be increased during 

call. If too small, it will be set to 10.0. Defaults to 0. 

accuracy : float 

Relative precision for finite difference calculations. If 

<= machine_precision, set to sqrt(machine_precision). 

Defaults to 0. 

minfev : float 

Minimum function value estimate. Defaults to 0. 

ftol : float 

Precision goal for the value of f in the stopping criterion. 

If ftol < 0.0, ftol is set to 0.0 defaults to -1. 

xtol : float 

Precision goal for the value of x in the stopping 

criterion (after applying x scaling factors). If xtol < 

0.0, xtol is set to sqrt(machine_precision). Defaults to 

-1. 

gtol : float 

Precision goal for the value of the projected gradient in 

the stopping criterion (after applying x scaling factors). 

If gtol < 0.0, gtol is set to 1e-2 * sqrt(accuracy). 

Setting it to 0.0 is not recommended. Defaults to -1. 

rescale : float 

Scaling factor (in log10) used to trigger f value 

rescaling. If 0, rescale at each iteration. If a large 

value, never rescale. If < 0, rescale is set to 1.3. 

 

""" 

_check_unknown_options(unknown_options) 

epsilon = eps 

maxfun = maxiter 

fmin = minfev 

pgtol = gtol 

 

x0 = asfarray(x0).flatten() 

n = len(x0) 

 

if bounds is None: 

bounds = [(None,None)] * n 

if len(bounds) != n: 

raise ValueError('length of x0 != length of bounds') 

 

if mesg_num is not None: 

messages = {0:MSG_NONE, 1:MSG_ITER, 2:MSG_INFO, 3:MSG_VERS, 

4:MSG_EXIT, 5:MSG_ALL}.get(mesg_num, MSG_ALL) 

elif disp: 

messages = MSG_ALL 

else: 

messages = MSG_NONE 

 

if jac is None: 

def func_and_grad(x): 

f = fun(x, *args) 

g = approx_fprime(x, fun, epsilon, *args) 

return f, g 

else: 

def func_and_grad(x): 

f = fun(x, *args) 

g = jac(x, *args) 

return f, g 

 

""" 

low, up : the bounds (lists of floats) 

if low is None, the lower bounds are removed. 

if up is None, the upper bounds are removed. 

low and up defaults to None 

""" 

low = zeros(n) 

up = zeros(n) 

for i in range(n): 

if bounds[i] is None: 

l, u = -inf, inf 

else: 

l,u = bounds[i] 

if l is None: 

low[i] = -inf 

else: 

low[i] = l 

if u is None: 

up[i] = inf 

else: 

up[i] = u 

 

if scale is None: 

scale = array([]) 

 

if offset is None: 

offset = array([]) 

 

if maxfun is None: 

maxfun = max(100, 10*len(x0)) 

 

rc, nf, nit, x = moduleTNC.minimize(func_and_grad, x0, low, up, scale, 

offset, messages, maxCGit, maxfun, 

eta, stepmx, accuracy, fmin, ftol, 

xtol, pgtol, rescale, callback) 

 

funv, jacv = func_and_grad(x) 

 

return OptimizeResult(x=x, fun=funv, jac=jacv, nfev=nf, nit=nit, status=rc, 

message=RCSTRINGS[rc], success=(-1 < rc < 3)) 

 

 

if __name__ == '__main__': 

# Examples for TNC 

 

def example(): 

print("Example") 

 

# A function to minimize 

def function(x): 

f = pow(x[0],2.0)+pow(abs(x[1]),3.0) 

g = [0,0] 

g[0] = 2.0*x[0] 

g[1] = 3.0*pow(abs(x[1]),2.0) 

if x[1] < 0: 

g[1] = -g[1] 

return f, g 

 

# Optimizer call 

x, nf, rc = fmin_tnc(function, [-7, 3], bounds=([-10, 1], [10, 10])) 

 

print("After", nf, "function evaluations, TNC returned:", RCSTRINGS[rc]) 

print("x =", x) 

print("exact value = [0, 1]") 

print() 

 

example()