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

523

524

525

526

527

528

529

530

531

532

533

534

535

536

537

538

539

540

541

542

543

544

545

546

547

548

549

550

551

552

553

554

555

556

557

558

559

560

561

562

563

564

565

566

567

568

569

570

571

572

573

574

575

576

577

578

579

580

581

582

583

584

585

586

587

588

589

590

591

592

593

594

595

596

597

598

599

600

601

602

603

604

605

606

607

608

609

610

611

612

613

614

615

616

617

618

619

620

621

622

623

624

625

626

627

628

629

630

631

632

633

634

635

636

637

638

639

640

641

642

643

644

645

646

647

648

649

650

651

652

653

654

655

656

657

658

659

660

661

662

663

664

665

666

667

668

669

670

671

672

673

674

675

676

677

678

679

680

681

682

683

684

685

686

687

688

689

690

691

692

693

694

695

696

697

698

699

700

701

702

703

704

705

706

707

708

709

710

711

712

713

714

715

716

717

718

719

720

721

722

723

724

725

726

727

728

729

730

731

732

733

734

735

736

737

738

739

740

741

742

743

744

745

746

747

748

749

750

751

752

753

754

755

756

757

758

759

760

761

762

763

764

765

766

767

768

769

770

771

772

773

774

775

776

777

778

779

780

781

782

783

784

785

786

787

788

789

790

791

792

793

794

795

796

797

798

799

800

801

802

803

804

805

806

807

808

809

810

811

812

813

814

815

816

817

818

819

820

821

822

823

824

825

826

827

828

829

830

831

832

833

834

835

836

837

838

839

840

841

842

843

844

845

846

847

848

849

850

851

852

853

854

855

856

857

858

859

860

861

862

863

864

865

866

867

868

869

870

871

872

873

874

875

876

877

878

879

880

881

882

883

884

885

886

887

888

889

890

891

892

893

894

895

896

897

898

899

900

901

902

903

904

from __future__ import division, print_function, absolute_import 

 

import threading 

import warnings 

from . import _minpack 

 

import numpy as np 

from numpy import (atleast_1d, dot, take, triu, shape, eye, 

transpose, zeros, product, greater, array, 

all, where, isscalar, asarray, inf, abs, 

finfo, inexact, issubdtype, dtype) 

from scipy.linalg import svd, cholesky, solve_triangular, LinAlgError 

from scipy._lib._util import _asarray_validated, _lazywhere 

from .optimize import OptimizeResult, _check_unknown_options, OptimizeWarning 

from ._lsq import least_squares 

from ._lsq.common import make_strictly_feasible 

from ._lsq.least_squares import prepare_bounds 

 

_MINPACK_LOCK = threading.RLock() 

error = _minpack.error 

 

__all__ = ['fsolve', 'leastsq', 'fixed_point', 'curve_fit'] 

 

 

def _check_func(checker, argname, thefunc, x0, args, numinputs, 

output_shape=None): 

res = atleast_1d(thefunc(*((x0[:numinputs],) + args))) 

if (output_shape is not None) and (shape(res) != output_shape): 

if (output_shape[0] != 1): 

if len(output_shape) > 1: 

if output_shape[1] == 1: 

return shape(res) 

msg = "%s: there is a mismatch between the input and output " \ 

"shape of the '%s' argument" % (checker, argname) 

func_name = getattr(thefunc, '__name__', None) 

if func_name: 

msg += " '%s'." % func_name 

else: 

msg += "." 

msg += 'Shape should be %s but it is %s.' % (output_shape, shape(res)) 

raise TypeError(msg) 

if issubdtype(res.dtype, inexact): 

dt = res.dtype 

else: 

dt = dtype(float) 

return shape(res), dt 

 

 

def fsolve(func, x0, args=(), fprime=None, full_output=0, 

col_deriv=0, xtol=1.49012e-8, maxfev=0, band=None, 

epsfcn=None, factor=100, diag=None): 

""" 

Find the roots of a function. 

 

Return the roots of the (non-linear) equations defined by 

``func(x) = 0`` given a starting estimate. 

 

Parameters 

---------- 

func : callable ``f(x, *args)`` 

A function that takes at least one (possibly vector) argument, 

and returns a value of the same length. 

x0 : ndarray 

The starting estimate for the roots of ``func(x) = 0``. 

args : tuple, optional 

Any extra arguments to `func`. 

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

A function to compute the Jacobian of `func` with derivatives 

across the rows. By default, the Jacobian will be estimated. 

full_output : bool, optional 

If True, return optional outputs. 

col_deriv : bool, optional 

Specify whether the Jacobian function computes derivatives down 

the columns (faster, because there is no transpose operation). 

xtol : float, optional 

The calculation will terminate if the relative error between two 

consecutive iterates is at most `xtol`. 

maxfev : int, optional 

The maximum number of calls to the function. If zero, then 

``100*(N+1)`` is the maximum where N is the number of elements 

in `x0`. 

band : tuple, optional 

If set to a two-sequence containing the number of sub- and 

super-diagonals within the band of the Jacobi matrix, the 

Jacobi matrix is considered banded (only for ``fprime=None``). 

epsfcn : float, optional 

A suitable step length for the forward-difference 

approximation of the Jacobian (for ``fprime=None``). If 

`epsfcn` is less than the machine precision, it is assumed 

that the relative errors in the functions are of the order of 

the machine precision. 

factor : float, optional 

A parameter determining the initial step bound 

(``factor * || diag * x||``). Should be in the interval 

``(0.1, 100)``. 

diag : sequence, optional 

N positive entries that serve as a scale factors for the 

variables. 

 

Returns 

------- 

x : ndarray 

The solution (or the result of the last iteration for 

an unsuccessful call). 

infodict : dict 

A dictionary of optional outputs with the keys: 

 

``nfev`` 

number of function calls 

``njev`` 

number of Jacobian calls 

``fvec`` 

function evaluated at the output 

``fjac`` 

the orthogonal matrix, q, produced by the QR 

factorization of the final approximate Jacobian 

matrix, stored column wise 

``r`` 

upper triangular matrix produced by QR factorization 

of the same matrix 

``qtf`` 

the vector ``(transpose(q) * fvec)`` 

 

ier : int 

An integer flag. Set to 1 if a solution was found, otherwise refer 

to `mesg` for more information. 

mesg : str 

If no solution is found, `mesg` details the cause of failure. 

 

See Also 

-------- 

root : Interface to root finding algorithms for multivariate 

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

 

Notes 

----- 

``fsolve`` is a wrapper around MINPACK's hybrd and hybrj algorithms. 

 

""" 

options = {'col_deriv': col_deriv, 

'xtol': xtol, 

'maxfev': maxfev, 

'band': band, 

'eps': epsfcn, 

'factor': factor, 

'diag': diag} 

 

res = _root_hybr(func, x0, args, jac=fprime, **options) 

if full_output: 

x = res['x'] 

info = dict((k, res.get(k)) 

for k in ('nfev', 'njev', 'fjac', 'r', 'qtf') if k in res) 

info['fvec'] = res['fun'] 

return x, info, res['status'], res['message'] 

else: 

status = res['status'] 

msg = res['message'] 

if status == 0: 

raise TypeError(msg) 

elif status == 1: 

pass 

elif status in [2, 3, 4, 5]: 

warnings.warn(msg, RuntimeWarning) 

else: 

raise TypeError(msg) 

return res['x'] 

 

 

def _root_hybr(func, x0, args=(), jac=None, 

col_deriv=0, xtol=1.49012e-08, maxfev=0, band=None, eps=None, 

factor=100, diag=None, **unknown_options): 

""" 

Find the roots of a multivariate function using MINPACK's hybrd and 

hybrj routines (modified Powell method). 

 

Options 

------- 

col_deriv : bool 

Specify whether the Jacobian function computes derivatives down 

the columns (faster, because there is no transpose operation). 

xtol : float 

The calculation will terminate if the relative error between two 

consecutive iterates is at most `xtol`. 

maxfev : int 

The maximum number of calls to the function. If zero, then 

``100*(N+1)`` is the maximum where N is the number of elements 

in `x0`. 

band : tuple 

If set to a two-sequence containing the number of sub- and 

super-diagonals within the band of the Jacobi matrix, the 

Jacobi matrix is considered banded (only for ``fprime=None``). 

eps : float 

A suitable step length for the forward-difference 

approximation of the Jacobian (for ``fprime=None``). If 

`eps` is less than the machine precision, it is assumed 

that the relative errors in the functions are of the order of 

the machine precision. 

factor : float 

A parameter determining the initial step bound 

(``factor * || diag * x||``). Should be in the interval 

``(0.1, 100)``. 

diag : sequence 

N positive entries that serve as a scale factors for the 

variables. 

 

""" 

_check_unknown_options(unknown_options) 

epsfcn = eps 

 

x0 = asarray(x0).flatten() 

n = len(x0) 

if not isinstance(args, tuple): 

args = (args,) 

shape, dtype = _check_func('fsolve', 'func', func, x0, args, n, (n,)) 

if epsfcn is None: 

epsfcn = finfo(dtype).eps 

Dfun = jac 

if Dfun is None: 

if band is None: 

ml, mu = -10, -10 

else: 

ml, mu = band[:2] 

if maxfev == 0: 

maxfev = 200 * (n + 1) 

with _MINPACK_LOCK: 

retval = _minpack._hybrd(func, x0, args, 1, xtol, maxfev, 

ml, mu, epsfcn, factor, diag) 

else: 

_check_func('fsolve', 'fprime', Dfun, x0, args, n, (n, n)) 

if (maxfev == 0): 

maxfev = 100 * (n + 1) 

with _MINPACK_LOCK: 

retval = _minpack._hybrj(func, Dfun, x0, args, 1, 

col_deriv, xtol, maxfev, factor, diag) 

 

x, status = retval[0], retval[-1] 

 

errors = {0: "Improper input parameters were entered.", 

1: "The solution converged.", 

2: "The number of calls to function has " 

"reached maxfev = %d." % maxfev, 

3: "xtol=%f is too small, no further improvement " 

"in the approximate\n solution " 

"is possible." % xtol, 

4: "The iteration is not making good progress, as measured " 

"by the \n improvement from the last five " 

"Jacobian evaluations.", 

5: "The iteration is not making good progress, " 

"as measured by the \n improvement from the last " 

"ten iterations.", 

'unknown': "An error occurred."} 

 

info = retval[1] 

info['fun'] = info.pop('fvec') 

sol = OptimizeResult(x=x, success=(status == 1), status=status) 

sol.update(info) 

try: 

sol['message'] = errors[status] 

except KeyError: 

sol['message'] = errors['unknown'] 

 

return sol 

 

 

def leastsq(func, x0, args=(), Dfun=None, full_output=0, 

col_deriv=0, ftol=1.49012e-8, xtol=1.49012e-8, 

gtol=0.0, maxfev=0, epsfcn=None, factor=100, diag=None): 

""" 

Minimize the sum of squares of a set of equations. 

 

:: 

 

x = arg min(sum(func(y)**2,axis=0)) 

y 

 

Parameters 

---------- 

func : callable 

should take at least one (possibly length N vector) argument and 

returns M floating point numbers. It must not return NaNs or 

fitting might fail. 

x0 : ndarray 

The starting estimate for the minimization. 

args : tuple, optional 

Any extra arguments to func are placed in this tuple. 

Dfun : callable, optional 

A function or method to compute the Jacobian of func with derivatives 

across the rows. If this is None, the Jacobian will be estimated. 

full_output : bool, optional 

non-zero to return all optional outputs. 

col_deriv : bool, optional 

non-zero to specify that the Jacobian function computes derivatives 

down the columns (faster, because there is no transpose operation). 

ftol : float, optional 

Relative error desired in the sum of squares. 

xtol : float, optional 

Relative error desired in the approximate solution. 

gtol : float, optional 

Orthogonality desired between the function vector and the columns of 

the Jacobian. 

maxfev : int, optional 

The maximum number of calls to the function. If `Dfun` is provided 

then the default `maxfev` is 100*(N+1) where N is the number of elements 

in x0, otherwise the default `maxfev` is 200*(N+1). 

epsfcn : float, optional 

A variable used in determining a suitable step length for the forward- 

difference approximation of the Jacobian (for Dfun=None). 

Normally the actual step length will be sqrt(epsfcn)*x 

If epsfcn is less than the machine precision, it is assumed that the 

relative errors are of the order of the machine precision. 

factor : float, optional 

A parameter determining the initial step bound 

(``factor * || diag * x||``). Should be in interval ``(0.1, 100)``. 

diag : sequence, optional 

N positive entries that serve as a scale factors for the variables. 

 

Returns 

------- 

x : ndarray 

The solution (or the result of the last iteration for an unsuccessful 

call). 

cov_x : ndarray 

Uses the fjac and ipvt optional outputs to construct an 

estimate of the jacobian around the solution. None if a 

singular matrix encountered (indicates very flat curvature in 

some direction). This matrix must be multiplied by the 

residual variance to get the covariance of the 

parameter estimates -- see curve_fit. 

infodict : dict 

a dictionary of optional outputs with the key s: 

 

``nfev`` 

The number of function calls 

``fvec`` 

The function evaluated at the output 

``fjac`` 

A permutation of the R matrix of a QR 

factorization of the final approximate 

Jacobian matrix, stored column wise. 

Together with ipvt, the covariance of the 

estimate can be approximated. 

``ipvt`` 

An integer array of length N which defines 

a permutation matrix, p, such that 

fjac*p = q*r, where r is upper triangular 

with diagonal elements of nonincreasing 

magnitude. Column j of p is column ipvt(j) 

of the identity matrix. 

``qtf`` 

The vector (transpose(q) * fvec). 

 

mesg : str 

A string message giving information about the cause of failure. 

ier : int 

An integer flag. If it is equal to 1, 2, 3 or 4, the solution was 

found. Otherwise, the solution was not found. In either case, the 

optional output variable 'mesg' gives more information. 

 

Notes 

----- 

"leastsq" is a wrapper around MINPACK's lmdif and lmder algorithms. 

 

cov_x is a Jacobian approximation to the Hessian of the least squares 

objective function. 

This approximation assumes that the objective function is based on the 

difference between some observed target data (ydata) and a (non-linear) 

function of the parameters `f(xdata, params)` :: 

 

func(params) = ydata - f(xdata, params) 

 

so that the objective function is :: 

 

min sum((ydata - f(xdata, params))**2, axis=0) 

params 

 

The solution, `x`, is always a 1D array, regardless of the shape of `x0`, 

or whether `x0` is a scalar. 

""" 

x0 = asarray(x0).flatten() 

n = len(x0) 

if not isinstance(args, tuple): 

args = (args,) 

shape, dtype = _check_func('leastsq', 'func', func, x0, args, n) 

m = shape[0] 

if n > m: 

raise TypeError('Improper input: N=%s must not exceed M=%s' % (n, m)) 

if epsfcn is None: 

epsfcn = finfo(dtype).eps 

if Dfun is None: 

if maxfev == 0: 

maxfev = 200*(n + 1) 

with _MINPACK_LOCK: 

retval = _minpack._lmdif(func, x0, args, full_output, ftol, xtol, 

gtol, maxfev, epsfcn, factor, diag) 

else: 

if col_deriv: 

_check_func('leastsq', 'Dfun', Dfun, x0, args, n, (n, m)) 

else: 

_check_func('leastsq', 'Dfun', Dfun, x0, args, n, (m, n)) 

if maxfev == 0: 

maxfev = 100 * (n + 1) 

with _MINPACK_LOCK: 

retval = _minpack._lmder(func, Dfun, x0, args, full_output, 

col_deriv, ftol, xtol, gtol, maxfev, 

factor, diag) 

 

errors = {0: ["Improper input parameters.", TypeError], 

1: ["Both actual and predicted relative reductions " 

"in the sum of squares\n are at most %f" % ftol, None], 

2: ["The relative error between two consecutive " 

"iterates is at most %f" % xtol, None], 

3: ["Both actual and predicted relative reductions in " 

"the sum of squares\n are at most %f and the " 

"relative error between two consecutive " 

"iterates is at \n most %f" % (ftol, xtol), None], 

4: ["The cosine of the angle between func(x) and any " 

"column of the\n Jacobian is at most %f in " 

"absolute value" % gtol, None], 

5: ["Number of calls to function has reached " 

"maxfev = %d." % maxfev, ValueError], 

6: ["ftol=%f is too small, no further reduction " 

"in the sum of squares\n is possible.""" % ftol, 

ValueError], 

7: ["xtol=%f is too small, no further improvement in " 

"the approximate\n solution is possible." % xtol, 

ValueError], 

8: ["gtol=%f is too small, func(x) is orthogonal to the " 

"columns of\n the Jacobian to machine " 

"precision." % gtol, ValueError], 

'unknown': ["Unknown error.", TypeError]} 

 

info = retval[-1] # The FORTRAN return value 

 

if info not in [1, 2, 3, 4] and not full_output: 

if info in [5, 6, 7, 8]: 

warnings.warn(errors[info][0], RuntimeWarning) 

else: 

try: 

raise errors[info][1](errors[info][0]) 

except KeyError: 

raise errors['unknown'][1](errors['unknown'][0]) 

 

mesg = errors[info][0] 

if full_output: 

cov_x = None 

if info in [1, 2, 3, 4]: 

from numpy.dual import inv 

perm = take(eye(n), retval[1]['ipvt'] - 1, 0) 

r = triu(transpose(retval[1]['fjac'])[:n, :]) 

R = dot(r, perm) 

try: 

cov_x = inv(dot(transpose(R), R)) 

except (LinAlgError, ValueError): 

pass 

return (retval[0], cov_x) + retval[1:-1] + (mesg, info) 

else: 

return (retval[0], info) 

 

 

def _wrap_func(func, xdata, ydata, transform): 

if transform is None: 

def func_wrapped(params): 

return func(xdata, *params) - ydata 

elif transform.ndim == 1: 

def func_wrapped(params): 

return transform * (func(xdata, *params) - ydata) 

else: 

# Chisq = (y - yd)^T C^{-1} (y-yd) 

# transform = L such that C = L L^T 

# C^{-1} = L^{-T} L^{-1} 

# Chisq = (y - yd)^T L^{-T} L^{-1} (y-yd) 

# Define (y-yd)' = L^{-1} (y-yd) 

# by solving 

# L (y-yd)' = (y-yd) 

# and minimize (y-yd)'^T (y-yd)' 

def func_wrapped(params): 

return solve_triangular(transform, func(xdata, *params) - ydata, lower=True) 

return func_wrapped 

 

 

def _wrap_jac(jac, xdata, transform): 

if transform is None: 

def jac_wrapped(params): 

return jac(xdata, *params) 

elif transform.ndim == 1: 

def jac_wrapped(params): 

return transform[:, np.newaxis] * np.asarray(jac(xdata, *params)) 

else: 

def jac_wrapped(params): 

return solve_triangular(transform, np.asarray(jac(xdata, *params)), lower=True) 

return jac_wrapped 

 

 

def _initialize_feasible(lb, ub): 

p0 = np.ones_like(lb) 

lb_finite = np.isfinite(lb) 

ub_finite = np.isfinite(ub) 

 

mask = lb_finite & ub_finite 

p0[mask] = 0.5 * (lb[mask] + ub[mask]) 

 

mask = lb_finite & ~ub_finite 

p0[mask] = lb[mask] + 1 

 

mask = ~lb_finite & ub_finite 

p0[mask] = ub[mask] - 1 

 

return p0 

 

 

def curve_fit(f, xdata, ydata, p0=None, sigma=None, absolute_sigma=False, 

check_finite=True, bounds=(-np.inf, np.inf), method=None, 

jac=None, **kwargs): 

""" 

Use non-linear least squares to fit a function, f, to data. 

 

Assumes ``ydata = f(xdata, *params) + eps`` 

 

Parameters 

---------- 

f : callable 

The model function, f(x, ...). It must take the independent 

variable as the first argument and the parameters to fit as 

separate remaining arguments. 

xdata : An M-length sequence or an (k,M)-shaped array for functions with k predictors 

The independent variable where the data is measured. 

ydata : M-length sequence 

The dependent data --- nominally f(xdata, ...) 

p0 : None, scalar, or N-length sequence, optional 

Initial guess for the parameters. If None, then the initial 

values will all be 1 (if the number of parameters for the function 

can be determined using introspection, otherwise a ValueError 

is raised). 

sigma : None or M-length sequence or MxM array, optional 

Determines the uncertainty in `ydata`. If we define residuals as 

``r = ydata - f(xdata, *popt)``, then the interpretation of `sigma` 

depends on its number of dimensions: 

 

- A 1-d `sigma` should contain values of standard deviations of 

errors in `ydata`. In this case, the optimized function is 

``chisq = sum((r / sigma) ** 2)``. 

 

- A 2-d `sigma` should contain the covariance matrix of 

errors in `ydata`. In this case, the optimized function is 

``chisq = r.T @ inv(sigma) @ r``. 

 

.. versionadded:: 0.19 

 

None (default) is equivalent of 1-d `sigma` filled with ones. 

absolute_sigma : bool, optional 

If True, `sigma` is used in an absolute sense and the estimated parameter 

covariance `pcov` reflects these absolute values. 

 

If False, only the relative magnitudes of the `sigma` values matter. 

The returned parameter covariance matrix `pcov` is based on scaling 

`sigma` by a constant factor. This constant is set by demanding that the 

reduced `chisq` for the optimal parameters `popt` when using the 

*scaled* `sigma` equals unity. In other words, `sigma` is scaled to 

match the sample variance of the residuals after the fit. 

Mathematically, 

``pcov(absolute_sigma=False) = pcov(absolute_sigma=True) * chisq(popt)/(M-N)`` 

check_finite : bool, optional 

If True, check that the input arrays do not contain nans of infs, 

and raise a ValueError if they do. Setting this parameter to 

False may silently produce nonsensical results if the input arrays 

do contain nans. Default is True. 

bounds : 2-tuple of array_like, optional 

Lower and upper bounds on parameters. Defaults to no bounds. 

Each element of the tuple must be either an array with the length equal 

to the number of parameters, or a scalar (in which case the bound is 

taken to be the same for all parameters.) Use ``np.inf`` with an 

appropriate sign to disable bounds on all or some parameters. 

 

.. versionadded:: 0.17 

method : {'lm', 'trf', 'dogbox'}, optional 

Method to use for optimization. See `least_squares` for more details. 

Default is 'lm' for unconstrained problems and 'trf' if `bounds` are 

provided. The method 'lm' won't work when the number of observations 

is less than the number of variables, use 'trf' or 'dogbox' in this 

case. 

 

.. versionadded:: 0.17 

jac : callable, string or None, optional 

Function with signature ``jac(x, ...)`` which computes the Jacobian 

matrix of the model function with respect to parameters as a dense 

array_like structure. It will be scaled according to provided `sigma`. 

If None (default), the Jacobian will be estimated numerically. 

String keywords for 'trf' and 'dogbox' methods can be used to select 

a finite difference scheme, see `least_squares`. 

 

.. versionadded:: 0.18 

kwargs 

Keyword arguments passed to `leastsq` for ``method='lm'`` or 

`least_squares` otherwise. 

 

Returns 

------- 

popt : array 

Optimal values for the parameters so that the sum of the squared 

residuals of ``f(xdata, *popt) - ydata`` is minimized 

pcov : 2d array 

The estimated covariance of popt. The diagonals provide the variance 

of the parameter estimate. To compute one standard deviation errors 

on the parameters use ``perr = np.sqrt(np.diag(pcov))``. 

 

How the `sigma` parameter affects the estimated covariance 

depends on `absolute_sigma` argument, as described above. 

 

If the Jacobian matrix at the solution doesn't have a full rank, then 

'lm' method returns a matrix filled with ``np.inf``, on the other hand 

'trf' and 'dogbox' methods use Moore-Penrose pseudoinverse to compute 

the covariance matrix. 

 

Raises 

------ 

ValueError 

if either `ydata` or `xdata` contain NaNs, or if incompatible options 

are used. 

 

RuntimeError 

if the least-squares minimization fails. 

 

OptimizeWarning 

if covariance of the parameters can not be estimated. 

 

See Also 

-------- 

least_squares : Minimize the sum of squares of nonlinear functions. 

scipy.stats.linregress : Calculate a linear least squares regression for 

two sets of measurements. 

 

Notes 

----- 

With ``method='lm'``, the algorithm uses the Levenberg-Marquardt algorithm 

through `leastsq`. Note that this algorithm can only deal with 

unconstrained problems. 

 

Box constraints can be handled by methods 'trf' and 'dogbox'. Refer to 

the docstring of `least_squares` for more information. 

 

Examples 

-------- 

>>> import numpy as np 

>>> import matplotlib.pyplot as plt 

>>> from scipy.optimize import curve_fit 

 

>>> def func(x, a, b, c): 

... return a * np.exp(-b * x) + c 

 

Define the data to be fit with some noise: 

 

>>> xdata = np.linspace(0, 4, 50) 

>>> y = func(xdata, 2.5, 1.3, 0.5) 

>>> np.random.seed(1729) 

>>> y_noise = 0.2 * np.random.normal(size=xdata.size) 

>>> ydata = y + y_noise 

>>> plt.plot(xdata, ydata, 'b-', label='data') 

 

Fit for the parameters a, b, c of the function `func`: 

 

>>> popt, pcov = curve_fit(func, xdata, ydata) 

>>> popt 

array([ 2.55423706, 1.35190947, 0.47450618]) 

>>> plt.plot(xdata, func(xdata, *popt), 'r-', 

... label='fit: a=%5.3f, b=%5.3f, c=%5.3f' % tuple(popt)) 

 

Constrain the optimization to the region of ``0 <= a <= 3``, 

``0 <= b <= 1`` and ``0 <= c <= 0.5``: 

 

>>> popt, pcov = curve_fit(func, xdata, ydata, bounds=(0, [3., 1., 0.5])) 

>>> popt 

array([ 2.43708906, 1. , 0.35015434]) 

>>> plt.plot(xdata, func(xdata, *popt), 'g--', 

... label='fit: a=%5.3f, b=%5.3f, c=%5.3f' % tuple(popt)) 

 

>>> plt.xlabel('x') 

>>> plt.ylabel('y') 

>>> plt.legend() 

>>> plt.show() 

 

""" 

if p0 is None: 

# determine number of parameters by inspecting the function 

from scipy._lib._util import getargspec_no_self as _getargspec 

args, varargs, varkw, defaults = _getargspec(f) 

if len(args) < 2: 

raise ValueError("Unable to determine number of fit parameters.") 

n = len(args) - 1 

else: 

p0 = np.atleast_1d(p0) 

n = p0.size 

 

lb, ub = prepare_bounds(bounds, n) 

if p0 is None: 

p0 = _initialize_feasible(lb, ub) 

 

bounded_problem = np.any((lb > -np.inf) | (ub < np.inf)) 

if method is None: 

if bounded_problem: 

method = 'trf' 

else: 

method = 'lm' 

 

if method == 'lm' and bounded_problem: 

raise ValueError("Method 'lm' only works for unconstrained problems. " 

"Use 'trf' or 'dogbox' instead.") 

 

# NaNs can not be handled 

if check_finite: 

ydata = np.asarray_chkfinite(ydata) 

else: 

ydata = np.asarray(ydata) 

 

if isinstance(xdata, (list, tuple, np.ndarray)): 

# `xdata` is passed straight to the user-defined `f`, so allow 

# non-array_like `xdata`. 

if check_finite: 

xdata = np.asarray_chkfinite(xdata) 

else: 

xdata = np.asarray(xdata) 

 

# Determine type of sigma 

if sigma is not None: 

sigma = np.asarray(sigma) 

 

# if 1-d, sigma are errors, define transform = 1/sigma 

if sigma.shape == (ydata.size, ): 

transform = 1.0 / sigma 

# if 2-d, sigma is the covariance matrix, 

# define transform = L such that L L^T = C 

elif sigma.shape == (ydata.size, ydata.size): 

try: 

# scipy.linalg.cholesky requires lower=True to return L L^T = A 

transform = cholesky(sigma, lower=True) 

except LinAlgError: 

raise ValueError("`sigma` must be positive definite.") 

else: 

raise ValueError("`sigma` has incorrect shape.") 

else: 

transform = None 

 

func = _wrap_func(f, xdata, ydata, transform) 

if callable(jac): 

jac = _wrap_jac(jac, xdata, transform) 

elif jac is None and method != 'lm': 

jac = '2-point' 

 

if method == 'lm': 

# Remove full_output from kwargs, otherwise we're passing it in twice. 

return_full = kwargs.pop('full_output', False) 

res = leastsq(func, p0, Dfun=jac, full_output=1, **kwargs) 

popt, pcov, infodict, errmsg, ier = res 

cost = np.sum(infodict['fvec'] ** 2) 

if ier not in [1, 2, 3, 4]: 

raise RuntimeError("Optimal parameters not found: " + errmsg) 

else: 

# Rename maxfev (leastsq) to max_nfev (least_squares), if specified. 

if 'max_nfev' not in kwargs: 

kwargs['max_nfev'] = kwargs.pop('maxfev', None) 

 

res = least_squares(func, p0, jac=jac, bounds=bounds, method=method, 

**kwargs) 

 

if not res.success: 

raise RuntimeError("Optimal parameters not found: " + res.message) 

 

cost = 2 * res.cost # res.cost is half sum of squares! 

popt = res.x 

 

# Do Moore-Penrose inverse discarding zero singular values. 

_, s, VT = svd(res.jac, full_matrices=False) 

threshold = np.finfo(float).eps * max(res.jac.shape) * s[0] 

s = s[s > threshold] 

VT = VT[:s.size] 

pcov = np.dot(VT.T / s**2, VT) 

return_full = False 

 

warn_cov = False 

if pcov is None: 

# indeterminate covariance 

pcov = zeros((len(popt), len(popt)), dtype=float) 

pcov.fill(inf) 

warn_cov = True 

elif not absolute_sigma: 

if ydata.size > p0.size: 

s_sq = cost / (ydata.size - p0.size) 

pcov = pcov * s_sq 

else: 

pcov.fill(inf) 

warn_cov = True 

 

if warn_cov: 

warnings.warn('Covariance of the parameters could not be estimated', 

category=OptimizeWarning) 

 

if return_full: 

return popt, pcov, infodict, errmsg, ier 

else: 

return popt, pcov 

 

 

def check_gradient(fcn, Dfcn, x0, args=(), col_deriv=0): 

"""Perform a simple check on the gradient for correctness. 

 

""" 

 

x = atleast_1d(x0) 

n = len(x) 

x = x.reshape((n,)) 

fvec = atleast_1d(fcn(x, *args)) 

m = len(fvec) 

fvec = fvec.reshape((m,)) 

ldfjac = m 

fjac = atleast_1d(Dfcn(x, *args)) 

fjac = fjac.reshape((m, n)) 

if col_deriv == 0: 

fjac = transpose(fjac) 

 

xp = zeros((n,), float) 

err = zeros((m,), float) 

fvecp = None 

with _MINPACK_LOCK: 

_minpack._chkder(m, n, x, fvec, fjac, ldfjac, xp, fvecp, 1, err) 

 

fvecp = atleast_1d(fcn(xp, *args)) 

fvecp = fvecp.reshape((m,)) 

with _MINPACK_LOCK: 

_minpack._chkder(m, n, x, fvec, fjac, ldfjac, xp, fvecp, 2, err) 

 

good = (product(greater(err, 0.5), axis=0)) 

 

return (good, err) 

 

 

def _del2(p0, p1, d): 

return p0 - np.square(p1 - p0) / d 

 

 

def _relerr(actual, desired): 

return (actual - desired) / desired 

 

 

def _fixed_point_helper(func, x0, args, xtol, maxiter, use_accel): 

p0 = x0 

for i in range(maxiter): 

p1 = func(p0, *args) 

if use_accel: 

p2 = func(p1, *args) 

d = p2 - 2.0 * p1 + p0 

p = _lazywhere(d != 0, (p0, p1, d), f=_del2, fillvalue=p2) 

else: 

p = p1 

relerr = _lazywhere(p0 != 0, (p, p0), f=_relerr, fillvalue=p) 

if np.all(np.abs(relerr) < xtol): 

return p 

p0 = p 

msg = "Failed to converge after %d iterations, value is %s" % (maxiter, p) 

raise RuntimeError(msg) 

 

 

def fixed_point(func, x0, args=(), xtol=1e-8, maxiter=500, method='del2'): 

""" 

Find a fixed point of the function. 

 

Given a function of one or more variables and a starting point, find a 

fixed-point of the function: i.e. where ``func(x0) == x0``. 

 

Parameters 

---------- 

func : function 

Function to evaluate. 

x0 : array_like 

Fixed point of function. 

args : tuple, optional 

Extra arguments to `func`. 

xtol : float, optional 

Convergence tolerance, defaults to 1e-08. 

maxiter : int, optional 

Maximum number of iterations, defaults to 500. 

method : {"del2", "iteration"}, optional 

Method of finding the fixed-point, defaults to "del2" 

which uses Steffensen's Method with Aitken's ``Del^2`` 

convergence acceleration [1]_. The "iteration" method simply iterates 

the function until convergence is detected, without attempting to 

accelerate the convergence. 

 

References 

---------- 

.. [1] Burden, Faires, "Numerical Analysis", 5th edition, pg. 80 

 

Examples 

-------- 

>>> from scipy import optimize 

>>> def func(x, c1, c2): 

... return np.sqrt(c1/(x+c2)) 

>>> c1 = np.array([10,12.]) 

>>> c2 = np.array([3, 5.]) 

>>> optimize.fixed_point(func, [1.2, 1.3], args=(c1,c2)) 

array([ 1.4920333 , 1.37228132]) 

 

""" 

use_accel = {'del2': True, 'iteration': False}[method] 

x0 = _asarray_validated(x0, as_inexact=True) 

return _fixed_point_helper(func, x0, args, xtol, maxiter, use_accel)