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

905

906

907

908

909

910

911

912

913

914

915

916

917

918

919

920

921

922

923

924

925

926

927

928

929

930

931

932

933

934

935

936

937

938

939

940

941

942

943

944

945

946

947

948

949

950

951

952

953

954

955

956

957

958

959

960

961

962

963

964

965

966

967

968

969

970

971

972

973

974

975

976

977

978

979

980

981

982

983

984

985

986

987

988

989

990

991

992

993

994

995

996

997

998

999

1000

1001

1002

1003

1004

1005

1006

1007

1008

1009

1010

1011

1012

1013

1014

1015

1016

1017

1018

1019

1020

1021

1022

1023

1024

1025

1026

1027

1028

1029

1030

1031

1032

1033

1034

1035

1036

1037

1038

1039

1040

1041

1042

1043

1044

1045

1046

1047

1048

1049

1050

1051

1052

1053

1054

1055

1056

1057

1058

1059

1060

1061

1062

1063

1064

1065

1066

1067

1068

1069

1070

1071

1072

1073

1074

1075

1076

1077

1078

1079

1080

1081

1082

1083

1084

1085

1086

1087

1088

1089

1090

1091

1092

1093

1094

1095

1096

1097

1098

1099

1100

1101

1102

""" 

Histogram-related functions 

""" 

from __future__ import division, absolute_import, print_function 

 

import functools 

import operator 

import warnings 

 

import numpy as np 

from numpy.compat.py3k import basestring 

from numpy.core import overrides 

 

__all__ = ['histogram', 'histogramdd', 'histogram_bin_edges'] 

 

array_function_dispatch = functools.partial( 

overrides.array_function_dispatch, module='numpy') 

 

# range is a keyword argument to many functions, so save the builtin so they can 

# use it. 

_range = range 

 

 

def _hist_bin_sqrt(x, range): 

""" 

Square root histogram bin estimator. 

 

Bin width is inversely proportional to the data size. Used by many 

programs for its simplicity. 

 

Parameters 

---------- 

x : array_like 

Input data that is to be histogrammed, trimmed to range. May not 

be empty. 

 

Returns 

------- 

h : An estimate of the optimal bin width for the given data. 

""" 

del range # unused 

return x.ptp() / np.sqrt(x.size) 

 

 

def _hist_bin_sturges(x, range): 

""" 

Sturges histogram bin estimator. 

 

A very simplistic estimator based on the assumption of normality of 

the data. This estimator has poor performance for non-normal data, 

which becomes especially obvious for large data sets. The estimate 

depends only on size of the data. 

 

Parameters 

---------- 

x : array_like 

Input data that is to be histogrammed, trimmed to range. May not 

be empty. 

 

Returns 

------- 

h : An estimate of the optimal bin width for the given data. 

""" 

del range # unused 

return x.ptp() / (np.log2(x.size) + 1.0) 

 

 

def _hist_bin_rice(x, range): 

""" 

Rice histogram bin estimator. 

 

Another simple estimator with no normality assumption. It has better 

performance for large data than Sturges, but tends to overestimate 

the number of bins. The number of bins is proportional to the cube 

root of data size (asymptotically optimal). The estimate depends 

only on size of the data. 

 

Parameters 

---------- 

x : array_like 

Input data that is to be histogrammed, trimmed to range. May not 

be empty. 

 

Returns 

------- 

h : An estimate of the optimal bin width for the given data. 

""" 

del range # unused 

return x.ptp() / (2.0 * x.size ** (1.0 / 3)) 

 

 

def _hist_bin_scott(x, range): 

""" 

Scott histogram bin estimator. 

 

The binwidth is proportional to the standard deviation of the data 

and inversely proportional to the cube root of data size 

(asymptotically optimal). 

 

Parameters 

---------- 

x : array_like 

Input data that is to be histogrammed, trimmed to range. May not 

be empty. 

 

Returns 

------- 

h : An estimate of the optimal bin width for the given data. 

""" 

del range # unused 

return (24.0 * np.pi**0.5 / x.size)**(1.0 / 3.0) * np.std(x) 

 

 

def _hist_bin_stone(x, range): 

""" 

Histogram bin estimator based on minimizing the estimated integrated squared error (ISE). 

 

The number of bins is chosen by minimizing the estimated ISE against the unknown true distribution. 

The ISE is estimated using cross-validation and can be regarded as a generalization of Scott's rule. 

https://en.wikipedia.org/wiki/Histogram#Scott.27s_normal_reference_rule 

 

This paper by Stone appears to be the origination of this rule. 

http://digitalassets.lib.berkeley.edu/sdtr/ucb/text/34.pdf 

 

Parameters 

---------- 

x : array_like 

Input data that is to be histogrammed, trimmed to range. May not 

be empty. 

range : (float, float) 

The lower and upper range of the bins. 

 

Returns 

------- 

h : An estimate of the optimal bin width for the given data. 

""" 

 

n = x.size 

ptp_x = np.ptp(x) 

if n <= 1 or ptp_x == 0: 

return 0 

 

def jhat(nbins): 

hh = ptp_x / nbins 

p_k = np.histogram(x, bins=nbins, range=range)[0] / n 

return (2 - (n + 1) * p_k.dot(p_k)) / hh 

 

nbins_upper_bound = max(100, int(np.sqrt(n))) 

nbins = min(_range(1, nbins_upper_bound + 1), key=jhat) 

if nbins == nbins_upper_bound: 

warnings.warn("The number of bins estimated may be suboptimal.", RuntimeWarning, stacklevel=2) 

return ptp_x / nbins 

 

 

def _hist_bin_doane(x, range): 

""" 

Doane's histogram bin estimator. 

 

Improved version of Sturges' formula which works better for 

non-normal data. See 

stats.stackexchange.com/questions/55134/doanes-formula-for-histogram-binning 

 

Parameters 

---------- 

x : array_like 

Input data that is to be histogrammed, trimmed to range. May not 

be empty. 

 

Returns 

------- 

h : An estimate of the optimal bin width for the given data. 

""" 

del range # unused 

if x.size > 2: 

sg1 = np.sqrt(6.0 * (x.size - 2) / ((x.size + 1.0) * (x.size + 3))) 

sigma = np.std(x) 

if sigma > 0.0: 

# These three operations add up to 

# g1 = np.mean(((x - np.mean(x)) / sigma)**3) 

# but use only one temp array instead of three 

temp = x - np.mean(x) 

np.true_divide(temp, sigma, temp) 

np.power(temp, 3, temp) 

g1 = np.mean(temp) 

return x.ptp() / (1.0 + np.log2(x.size) + 

np.log2(1.0 + np.absolute(g1) / sg1)) 

return 0.0 

 

 

def _hist_bin_fd(x, range): 

""" 

The Freedman-Diaconis histogram bin estimator. 

 

The Freedman-Diaconis rule uses interquartile range (IQR) to 

estimate binwidth. It is considered a variation of the Scott rule 

with more robustness as the IQR is less affected by outliers than 

the standard deviation. However, the IQR depends on fewer points 

than the standard deviation, so it is less accurate, especially for 

long tailed distributions. 

 

If the IQR is 0, this function returns 1 for the number of bins. 

Binwidth is inversely proportional to the cube root of data size 

(asymptotically optimal). 

 

Parameters 

---------- 

x : array_like 

Input data that is to be histogrammed, trimmed to range. May not 

be empty. 

 

Returns 

------- 

h : An estimate of the optimal bin width for the given data. 

""" 

del range # unused 

iqr = np.subtract(*np.percentile(x, [75, 25])) 

return 2.0 * iqr * x.size ** (-1.0 / 3.0) 

 

 

def _hist_bin_auto(x, range): 

""" 

Histogram bin estimator that uses the minimum width of the 

Freedman-Diaconis and Sturges estimators if the FD bandwidth is non zero 

and the Sturges estimator if the FD bandwidth is 0. 

 

The FD estimator is usually the most robust method, but its width 

estimate tends to be too large for small `x` and bad for data with limited 

variance. The Sturges estimator is quite good for small (<1000) datasets 

and is the default in the R language. This method gives good off the shelf 

behaviour. 

 

.. versionchanged:: 1.15.0 

If there is limited variance the IQR can be 0, which results in the 

FD bin width being 0 too. This is not a valid bin width, so 

``np.histogram_bin_edges`` chooses 1 bin instead, which may not be optimal. 

If the IQR is 0, it's unlikely any variance based estimators will be of 

use, so we revert to the sturges estimator, which only uses the size of the 

dataset in its calculation. 

 

Parameters 

---------- 

x : array_like 

Input data that is to be histogrammed, trimmed to range. May not 

be empty. 

 

Returns 

------- 

h : An estimate of the optimal bin width for the given data. 

 

See Also 

-------- 

_hist_bin_fd, _hist_bin_sturges 

""" 

fd_bw = _hist_bin_fd(x, range) 

sturges_bw = _hist_bin_sturges(x, range) 

del range # unused 

if fd_bw: 

return min(fd_bw, sturges_bw) 

else: 

# limited variance, so we return a len dependent bw estimator 

return sturges_bw 

 

# Private dict initialized at module load time 

_hist_bin_selectors = {'stone': _hist_bin_stone, 

'auto': _hist_bin_auto, 

'doane': _hist_bin_doane, 

'fd': _hist_bin_fd, 

'rice': _hist_bin_rice, 

'scott': _hist_bin_scott, 

'sqrt': _hist_bin_sqrt, 

'sturges': _hist_bin_sturges} 

 

 

def _ravel_and_check_weights(a, weights): 

""" Check a and weights have matching shapes, and ravel both """ 

a = np.asarray(a) 

 

# Ensure that the array is a "subtractable" dtype 

if a.dtype == np.bool_: 

warnings.warn("Converting input from {} to {} for compatibility." 

.format(a.dtype, np.uint8), 

RuntimeWarning, stacklevel=2) 

a = a.astype(np.uint8) 

 

if weights is not None: 

weights = np.asarray(weights) 

if weights.shape != a.shape: 

raise ValueError( 

'weights should have the same shape as a.') 

weights = weights.ravel() 

a = a.ravel() 

return a, weights 

 

 

def _get_outer_edges(a, range): 

""" 

Determine the outer bin edges to use, from either the data or the range 

argument 

""" 

if range is not None: 

first_edge, last_edge = range 

if first_edge > last_edge: 

raise ValueError( 

'max must be larger than min in range parameter.') 

if not (np.isfinite(first_edge) and np.isfinite(last_edge)): 

raise ValueError( 

"supplied range of [{}, {}] is not finite".format(first_edge, last_edge)) 

elif a.size == 0: 

# handle empty arrays. Can't determine range, so use 0-1. 

first_edge, last_edge = 0, 1 

else: 

first_edge, last_edge = a.min(), a.max() 

if not (np.isfinite(first_edge) and np.isfinite(last_edge)): 

raise ValueError( 

"autodetected range of [{}, {}] is not finite".format(first_edge, last_edge)) 

 

# expand empty range to avoid divide by zero 

if first_edge == last_edge: 

first_edge = first_edge - 0.5 

last_edge = last_edge + 0.5 

 

return first_edge, last_edge 

 

 

def _unsigned_subtract(a, b): 

""" 

Subtract two values where a >= b, and produce an unsigned result 

 

This is needed when finding the difference between the upper and lower 

bound of an int16 histogram 

""" 

# coerce to a single type 

signed_to_unsigned = { 

np.byte: np.ubyte, 

np.short: np.ushort, 

np.intc: np.uintc, 

np.int_: np.uint, 

np.longlong: np.ulonglong 

} 

dt = np.result_type(a, b) 

try: 

dt = signed_to_unsigned[dt.type] 

except KeyError: 

return np.subtract(a, b, dtype=dt) 

else: 

# we know the inputs are integers, and we are deliberately casting 

# signed to unsigned 

return np.subtract(a, b, casting='unsafe', dtype=dt) 

 

 

def _get_bin_edges(a, bins, range, weights): 

""" 

Computes the bins used internally by `histogram`. 

 

Parameters 

========== 

a : ndarray 

Ravelled data array 

bins, range 

Forwarded arguments from `histogram`. 

weights : ndarray, optional 

Ravelled weights array, or None 

 

Returns 

======= 

bin_edges : ndarray 

Array of bin edges 

uniform_bins : (Number, Number, int): 

The upper bound, lowerbound, and number of bins, used in the optimized 

implementation of `histogram` that works on uniform bins. 

""" 

# parse the overloaded bins argument 

n_equal_bins = None 

bin_edges = None 

 

if isinstance(bins, basestring): 

bin_name = bins 

# if `bins` is a string for an automatic method, 

# this will replace it with the number of bins calculated 

if bin_name not in _hist_bin_selectors: 

raise ValueError( 

"{!r} is not a valid estimator for `bins`".format(bin_name)) 

if weights is not None: 

raise TypeError("Automated estimation of the number of " 

"bins is not supported for weighted data") 

 

first_edge, last_edge = _get_outer_edges(a, range) 

 

# truncate the range if needed 

if range is not None: 

keep = (a >= first_edge) 

keep &= (a <= last_edge) 

if not np.logical_and.reduce(keep): 

a = a[keep] 

 

if a.size == 0: 

n_equal_bins = 1 

else: 

# Do not call selectors on empty arrays 

width = _hist_bin_selectors[bin_name](a, (first_edge, last_edge)) 

if width: 

n_equal_bins = int(np.ceil(_unsigned_subtract(last_edge, first_edge) / width)) 

else: 

# Width can be zero for some estimators, e.g. FD when 

# the IQR of the data is zero. 

n_equal_bins = 1 

 

elif np.ndim(bins) == 0: 

try: 

n_equal_bins = operator.index(bins) 

except TypeError: 

raise TypeError( 

'`bins` must be an integer, a string, or an array') 

if n_equal_bins < 1: 

raise ValueError('`bins` must be positive, when an integer') 

 

first_edge, last_edge = _get_outer_edges(a, range) 

 

elif np.ndim(bins) == 1: 

bin_edges = np.asarray(bins) 

if np.any(bin_edges[:-1] > bin_edges[1:]): 

raise ValueError( 

'`bins` must increase monotonically, when an array') 

 

else: 

raise ValueError('`bins` must be 1d, when an array') 

 

if n_equal_bins is not None: 

# gh-10322 means that type resolution rules are dependent on array 

# shapes. To avoid this causing problems, we pick a type now and stick 

# with it throughout. 

bin_type = np.result_type(first_edge, last_edge, a) 

if np.issubdtype(bin_type, np.integer): 

bin_type = np.result_type(bin_type, float) 

 

# bin edges must be computed 

bin_edges = np.linspace( 

first_edge, last_edge, n_equal_bins + 1, 

endpoint=True, dtype=bin_type) 

return bin_edges, (first_edge, last_edge, n_equal_bins) 

else: 

return bin_edges, None 

 

 

def _search_sorted_inclusive(a, v): 

""" 

Like `searchsorted`, but where the last item in `v` is placed on the right. 

 

In the context of a histogram, this makes the last bin edge inclusive 

""" 

return np.concatenate(( 

a.searchsorted(v[:-1], 'left'), 

a.searchsorted(v[-1:], 'right') 

)) 

 

 

def _histogram_bin_edges_dispatcher(a, bins=None, range=None, weights=None): 

return (a, bins, weights) 

 

 

@array_function_dispatch(_histogram_bin_edges_dispatcher) 

def histogram_bin_edges(a, bins=10, range=None, weights=None): 

r""" 

Function to calculate only the edges of the bins used by the `histogram` function. 

 

Parameters 

---------- 

a : array_like 

Input data. The histogram is computed over the flattened array. 

bins : int or sequence of scalars or str, optional 

If `bins` is an int, it defines the number of equal-width 

bins in the given range (10, by default). If `bins` is a 

sequence, it defines the bin edges, including the rightmost 

edge, allowing for non-uniform bin widths. 

 

If `bins` is a string from the list below, `histogram_bin_edges` will use 

the method chosen to calculate the optimal bin width and 

consequently the number of bins (see `Notes` for more detail on 

the estimators) from the data that falls within the requested 

range. While the bin width will be optimal for the actual data 

in the range, the number of bins will be computed to fill the 

entire range, including the empty portions. For visualisation, 

using the 'auto' option is suggested. Weighted data is not 

supported for automated bin size selection. 

 

'auto' 

Maximum of the 'sturges' and 'fd' estimators. Provides good 

all around performance. 

 

'fd' (Freedman Diaconis Estimator) 

Robust (resilient to outliers) estimator that takes into 

account data variability and data size. 

 

'doane' 

An improved version of Sturges' estimator that works better 

with non-normal datasets. 

 

'scott' 

Less robust estimator that that takes into account data 

variability and data size. 

 

'stone' 

Estimator based on leave-one-out cross-validation estimate of 

the integrated squared error. Can be regarded as a generalization 

of Scott's rule. 

 

'rice' 

Estimator does not take variability into account, only data 

size. Commonly overestimates number of bins required. 

 

'sturges' 

R's default method, only accounts for data size. Only 

optimal for gaussian data and underestimates number of bins 

for large non-gaussian datasets. 

 

'sqrt' 

Square root (of data size) estimator, used by Excel and 

other programs for its speed and simplicity. 

 

range : (float, float), optional 

The lower and upper range of the bins. If not provided, range 

is simply ``(a.min(), a.max())``. Values outside the range are 

ignored. The first element of the range must be less than or 

equal to the second. `range` affects the automatic bin 

computation as well. While bin width is computed to be optimal 

based on the actual data within `range`, the bin count will fill 

the entire range including portions containing no data. 

 

weights : array_like, optional 

An array of weights, of the same shape as `a`. Each value in 

`a` only contributes its associated weight towards the bin count 

(instead of 1). This is currently not used by any of the bin estimators, 

but may be in the future. 

 

Returns 

------- 

bin_edges : array of dtype float 

The edges to pass into `histogram` 

 

See Also 

-------- 

histogram 

 

Notes 

----- 

The methods to estimate the optimal number of bins are well founded 

in literature, and are inspired by the choices R provides for 

histogram visualisation. Note that having the number of bins 

proportional to :math:`n^{1/3}` is asymptotically optimal, which is 

why it appears in most estimators. These are simply plug-in methods 

that give good starting points for number of bins. In the equations 

below, :math:`h` is the binwidth and :math:`n_h` is the number of 

bins. All estimators that compute bin counts are recast to bin width 

using the `ptp` of the data. The final bin count is obtained from 

``np.round(np.ceil(range / h))``. 

 

'Auto' (maximum of the 'Sturges' and 'FD' estimators) 

A compromise to get a good value. For small datasets the Sturges 

value will usually be chosen, while larger datasets will usually 

default to FD. Avoids the overly conservative behaviour of FD 

and Sturges for small and large datasets respectively. 

Switchover point is usually :math:`a.size \approx 1000`. 

 

'FD' (Freedman Diaconis Estimator) 

.. math:: h = 2 \frac{IQR}{n^{1/3}} 

 

The binwidth is proportional to the interquartile range (IQR) 

and inversely proportional to cube root of a.size. Can be too 

conservative for small datasets, but is quite good for large 

datasets. The IQR is very robust to outliers. 

 

'Scott' 

.. math:: h = \sigma \sqrt[3]{\frac{24 * \sqrt{\pi}}{n}} 

 

The binwidth is proportional to the standard deviation of the 

data and inversely proportional to cube root of ``x.size``. Can 

be too conservative for small datasets, but is quite good for 

large datasets. The standard deviation is not very robust to 

outliers. Values are very similar to the Freedman-Diaconis 

estimator in the absence of outliers. 

 

'Rice' 

.. math:: n_h = 2n^{1/3} 

 

The number of bins is only proportional to cube root of 

``a.size``. It tends to overestimate the number of bins and it 

does not take into account data variability. 

 

'Sturges' 

.. math:: n_h = \log _{2}n+1 

 

The number of bins is the base 2 log of ``a.size``. This 

estimator assumes normality of data and is too conservative for 

larger, non-normal datasets. This is the default method in R's 

``hist`` method. 

 

'Doane' 

.. math:: n_h = 1 + \log_{2}(n) + 

\log_{2}(1 + \frac{|g_1|}{\sigma_{g_1}}) 

 

g_1 = mean[(\frac{x - \mu}{\sigma})^3] 

 

\sigma_{g_1} = \sqrt{\frac{6(n - 2)}{(n + 1)(n + 3)}} 

 

An improved version of Sturges' formula that produces better 

estimates for non-normal datasets. This estimator attempts to 

account for the skew of the data. 

 

'Sqrt' 

.. math:: n_h = \sqrt n 

The simplest and fastest estimator. Only takes into account the 

data size. 

 

Examples 

-------- 

>>> arr = np.array([0, 0, 0, 1, 2, 3, 3, 4, 5]) 

>>> np.histogram_bin_edges(arr, bins='auto', range=(0, 1)) 

array([0. , 0.25, 0.5 , 0.75, 1. ]) 

>>> np.histogram_bin_edges(arr, bins=2) 

array([0. , 2.5, 5. ]) 

 

For consistency with histogram, an array of pre-computed bins is 

passed through unmodified: 

 

>>> np.histogram_bin_edges(arr, [1, 2]) 

array([1, 2]) 

 

This function allows one set of bins to be computed, and reused across 

multiple histograms: 

 

>>> shared_bins = np.histogram_bin_edges(arr, bins='auto') 

>>> shared_bins 

array([0., 1., 2., 3., 4., 5.]) 

 

>>> group_id = np.array([0, 1, 1, 0, 1, 1, 0, 1, 1]) 

>>> hist_0, _ = np.histogram(arr[group_id == 0], bins=shared_bins) 

>>> hist_1, _ = np.histogram(arr[group_id == 1], bins=shared_bins) 

 

>>> hist_0; hist_1 

array([1, 1, 0, 1, 0]) 

array([2, 0, 1, 1, 2]) 

 

Which gives more easily comparable results than using separate bins for 

each histogram: 

 

>>> hist_0, bins_0 = np.histogram(arr[group_id == 0], bins='auto') 

>>> hist_1, bins_1 = np.histogram(arr[group_id == 1], bins='auto') 

>>> hist_0; hist1 

array([1, 1, 1]) 

array([2, 1, 1, 2]) 

>>> bins_0; bins_1 

array([0., 1., 2., 3.]) 

array([0. , 1.25, 2.5 , 3.75, 5. ]) 

 

""" 

a, weights = _ravel_and_check_weights(a, weights) 

bin_edges, _ = _get_bin_edges(a, bins, range, weights) 

return bin_edges 

 

 

def _histogram_dispatcher( 

a, bins=None, range=None, normed=None, weights=None, density=None): 

return (a, bins, weights) 

 

 

@array_function_dispatch(_histogram_dispatcher) 

def histogram(a, bins=10, range=None, normed=None, weights=None, 

density=None): 

r""" 

Compute the histogram of a set of data. 

 

Parameters 

---------- 

a : array_like 

Input data. The histogram is computed over the flattened array. 

bins : int or sequence of scalars or str, optional 

If `bins` is an int, it defines the number of equal-width 

bins in the given range (10, by default). If `bins` is a 

sequence, it defines a monotonically increasing array of bin edges, 

including the rightmost edge, allowing for non-uniform bin widths. 

 

.. versionadded:: 1.11.0 

 

If `bins` is a string, it defines the method used to calculate the 

optimal bin width, as defined by `histogram_bin_edges`. 

 

range : (float, float), optional 

The lower and upper range of the bins. If not provided, range 

is simply ``(a.min(), a.max())``. Values outside the range are 

ignored. The first element of the range must be less than or 

equal to the second. `range` affects the automatic bin 

computation as well. While bin width is computed to be optimal 

based on the actual data within `range`, the bin count will fill 

the entire range including portions containing no data. 

normed : bool, optional 

 

.. deprecated:: 1.6.0 

 

This is equivalent to the `density` argument, but produces incorrect 

results for unequal bin widths. It should not be used. 

 

.. versionchanged:: 1.15.0 

DeprecationWarnings are actually emitted. 

 

weights : array_like, optional 

An array of weights, of the same shape as `a`. Each value in 

`a` only contributes its associated weight towards the bin count 

(instead of 1). If `density` is True, the weights are 

normalized, so that the integral of the density over the range 

remains 1. 

density : bool, optional 

If ``False``, the result will contain the number of samples in 

each bin. If ``True``, the result is the value of the 

probability *density* function at the bin, normalized such that 

the *integral* over the range is 1. Note that the sum of the 

histogram values will not be equal to 1 unless bins of unity 

width are chosen; it is not a probability *mass* function. 

 

Overrides the ``normed`` keyword if given. 

 

Returns 

------- 

hist : array 

The values of the histogram. See `density` and `weights` for a 

description of the possible semantics. 

bin_edges : array of dtype float 

Return the bin edges ``(length(hist)+1)``. 

 

 

See Also 

-------- 

histogramdd, bincount, searchsorted, digitize, histogram_bin_edges 

 

Notes 

----- 

All but the last (righthand-most) bin is half-open. In other words, 

if `bins` is:: 

 

[1, 2, 3, 4] 

 

then the first bin is ``[1, 2)`` (including 1, but excluding 2) and 

the second ``[2, 3)``. The last bin, however, is ``[3, 4]``, which 

*includes* 4. 

 

 

Examples 

-------- 

>>> np.histogram([1, 2, 1], bins=[0, 1, 2, 3]) 

(array([0, 2, 1]), array([0, 1, 2, 3])) 

>>> np.histogram(np.arange(4), bins=np.arange(5), density=True) 

(array([ 0.25, 0.25, 0.25, 0.25]), array([0, 1, 2, 3, 4])) 

>>> np.histogram([[1, 2, 1], [1, 0, 1]], bins=[0,1,2,3]) 

(array([1, 4, 1]), array([0, 1, 2, 3])) 

 

>>> a = np.arange(5) 

>>> hist, bin_edges = np.histogram(a, density=True) 

>>> hist 

array([ 0.5, 0. , 0.5, 0. , 0. , 0.5, 0. , 0.5, 0. , 0.5]) 

>>> hist.sum() 

2.4999999999999996 

>>> np.sum(hist * np.diff(bin_edges)) 

1.0 

 

.. versionadded:: 1.11.0 

 

Automated Bin Selection Methods example, using 2 peak random data 

with 2000 points: 

 

>>> import matplotlib.pyplot as plt 

>>> rng = np.random.RandomState(10) # deterministic random data 

>>> a = np.hstack((rng.normal(size=1000), 

... rng.normal(loc=5, scale=2, size=1000))) 

>>> plt.hist(a, bins='auto') # arguments are passed to np.histogram 

>>> plt.title("Histogram with 'auto' bins") 

>>> plt.show() 

 

""" 

a, weights = _ravel_and_check_weights(a, weights) 

 

bin_edges, uniform_bins = _get_bin_edges(a, bins, range, weights) 

 

# Histogram is an integer or a float array depending on the weights. 

if weights is None: 

ntype = np.dtype(np.intp) 

else: 

ntype = weights.dtype 

 

# We set a block size, as this allows us to iterate over chunks when 

# computing histograms, to minimize memory usage. 

BLOCK = 65536 

 

# The fast path uses bincount, but that only works for certain types 

# of weight 

simple_weights = ( 

weights is None or 

np.can_cast(weights.dtype, np.double) or 

np.can_cast(weights.dtype, complex) 

) 

 

if uniform_bins is not None and simple_weights: 

# Fast algorithm for equal bins 

# We now convert values of a to bin indices, under the assumption of 

# equal bin widths (which is valid here). 

first_edge, last_edge, n_equal_bins = uniform_bins 

 

# Initialize empty histogram 

n = np.zeros(n_equal_bins, ntype) 

 

# Pre-compute histogram scaling factor 

norm = n_equal_bins / _unsigned_subtract(last_edge, first_edge) 

 

# We iterate over blocks here for two reasons: the first is that for 

# large arrays, it is actually faster (for example for a 10^8 array it 

# is 2x as fast) and it results in a memory footprint 3x lower in the 

# limit of large arrays. 

for i in _range(0, len(a), BLOCK): 

tmp_a = a[i:i+BLOCK] 

if weights is None: 

tmp_w = None 

else: 

tmp_w = weights[i:i + BLOCK] 

 

# Only include values in the right range 

keep = (tmp_a >= first_edge) 

keep &= (tmp_a <= last_edge) 

if not np.logical_and.reduce(keep): 

tmp_a = tmp_a[keep] 

if tmp_w is not None: 

tmp_w = tmp_w[keep] 

 

# This cast ensures no type promotions occur below, which gh-10322 

# make unpredictable. Getting it wrong leads to precision errors 

# like gh-8123. 

tmp_a = tmp_a.astype(bin_edges.dtype, copy=False) 

 

# Compute the bin indices, and for values that lie exactly on 

# last_edge we need to subtract one 

f_indices = _unsigned_subtract(tmp_a, first_edge) * norm 

indices = f_indices.astype(np.intp) 

indices[indices == n_equal_bins] -= 1 

 

# The index computation is not guaranteed to give exactly 

# consistent results within ~1 ULP of the bin edges. 

decrement = tmp_a < bin_edges[indices] 

indices[decrement] -= 1 

# The last bin includes the right edge. The other bins do not. 

increment = ((tmp_a >= bin_edges[indices + 1]) 

& (indices != n_equal_bins - 1)) 

indices[increment] += 1 

 

# We now compute the histogram using bincount 

if ntype.kind == 'c': 

n.real += np.bincount(indices, weights=tmp_w.real, 

minlength=n_equal_bins) 

n.imag += np.bincount(indices, weights=tmp_w.imag, 

minlength=n_equal_bins) 

else: 

n += np.bincount(indices, weights=tmp_w, 

minlength=n_equal_bins).astype(ntype) 

else: 

# Compute via cumulative histogram 

cum_n = np.zeros(bin_edges.shape, ntype) 

if weights is None: 

for i in _range(0, len(a), BLOCK): 

sa = np.sort(a[i:i+BLOCK]) 

cum_n += _search_sorted_inclusive(sa, bin_edges) 

else: 

zero = np.zeros(1, dtype=ntype) 

for i in _range(0, len(a), BLOCK): 

tmp_a = a[i:i+BLOCK] 

tmp_w = weights[i:i+BLOCK] 

sorting_index = np.argsort(tmp_a) 

sa = tmp_a[sorting_index] 

sw = tmp_w[sorting_index] 

cw = np.concatenate((zero, sw.cumsum())) 

bin_index = _search_sorted_inclusive(sa, bin_edges) 

cum_n += cw[bin_index] 

 

n = np.diff(cum_n) 

 

# density overrides the normed keyword 

if density is not None: 

if normed is not None: 

# 2018-06-13, numpy 1.15.0 (this was not noisily deprecated in 1.6) 

warnings.warn( 

"The normed argument is ignored when density is provided. " 

"In future passing both will result in an error.", 

DeprecationWarning, stacklevel=2) 

normed = None 

 

if density: 

db = np.array(np.diff(bin_edges), float) 

return n/db/n.sum(), bin_edges 

elif normed: 

# 2018-06-13, numpy 1.15.0 (this was not noisily deprecated in 1.6) 

warnings.warn( 

"Passing `normed=True` on non-uniform bins has always been " 

"broken, and computes neither the probability density " 

"function nor the probability mass function. " 

"The result is only correct if the bins are uniform, when " 

"density=True will produce the same result anyway. " 

"The argument will be removed in a future version of " 

"numpy.", 

np.VisibleDeprecationWarning, stacklevel=2) 

 

# this normalization is incorrect, but 

db = np.array(np.diff(bin_edges), float) 

return n/(n*db).sum(), bin_edges 

else: 

if normed is not None: 

# 2018-06-13, numpy 1.15.0 (this was not noisily deprecated in 1.6) 

warnings.warn( 

"Passing normed=False is deprecated, and has no effect. " 

"Consider passing the density argument instead.", 

DeprecationWarning, stacklevel=2) 

return n, bin_edges 

 

 

def _histogramdd_dispatcher(sample, bins=None, range=None, normed=None, 

weights=None, density=None): 

return (sample, bins, weights) 

 

 

@array_function_dispatch(_histogramdd_dispatcher) 

def histogramdd(sample, bins=10, range=None, normed=None, weights=None, 

density=None): 

""" 

Compute the multidimensional histogram of some data. 

 

Parameters 

---------- 

sample : (N, D) array, or (D, N) array_like 

The data to be histogrammed. 

 

Note the unusual interpretation of sample when an array_like: 

 

* When an array, each row is a coordinate in a D-dimensional space - 

such as ``histogramgramdd(np.array([p1, p2, p3]))``. 

* When an array_like, each element is the list of values for single 

coordinate - such as ``histogramgramdd((X, Y, Z))``. 

 

The first form should be preferred. 

 

bins : sequence or int, optional 

The bin specification: 

 

* A sequence of arrays describing the monotonically increasing bin 

edges along each dimension. 

* The number of bins for each dimension (nx, ny, ... =bins) 

* The number of bins for all dimensions (nx=ny=...=bins). 

 

range : sequence, optional 

A sequence of length D, each an optional (lower, upper) tuple giving 

the outer bin edges to be used if the edges are not given explicitly in 

`bins`. 

An entry of None in the sequence results in the minimum and maximum 

values being used for the corresponding dimension. 

The default, None, is equivalent to passing a tuple of D None values. 

density : bool, optional 

If False, the default, returns the number of samples in each bin. 

If True, returns the probability *density* function at the bin, 

``bin_count / sample_count / bin_volume``. 

normed : bool, optional 

An alias for the density argument that behaves identically. To avoid 

confusion with the broken normed argument to `histogram`, `density` 

should be preferred. 

weights : (N,) array_like, optional 

An array of values `w_i` weighing each sample `(x_i, y_i, z_i, ...)`. 

Weights are normalized to 1 if normed is True. If normed is False, 

the values of the returned histogram are equal to the sum of the 

weights belonging to the samples falling into each bin. 

 

Returns 

------- 

H : ndarray 

The multidimensional histogram of sample x. See normed and weights 

for the different possible semantics. 

edges : list 

A list of D arrays describing the bin edges for each dimension. 

 

See Also 

-------- 

histogram: 1-D histogram 

histogram2d: 2-D histogram 

 

Examples 

-------- 

>>> r = np.random.randn(100,3) 

>>> H, edges = np.histogramdd(r, bins = (5, 8, 4)) 

>>> H.shape, edges[0].size, edges[1].size, edges[2].size 

((5, 8, 4), 6, 9, 5) 

 

""" 

 

try: 

# Sample is an ND-array. 

N, D = sample.shape 

except (AttributeError, ValueError): 

# Sample is a sequence of 1D arrays. 

sample = np.atleast_2d(sample).T 

N, D = sample.shape 

 

nbin = np.empty(D, int) 

edges = D*[None] 

dedges = D*[None] 

if weights is not None: 

weights = np.asarray(weights) 

 

try: 

M = len(bins) 

if M != D: 

raise ValueError( 

'The dimension of bins must be equal to the dimension of the ' 

' sample x.') 

except TypeError: 

# bins is an integer 

bins = D*[bins] 

 

# normalize the range argument 

if range is None: 

range = (None,) * D 

elif len(range) != D: 

raise ValueError('range argument must have one entry per dimension') 

 

# Create edge arrays 

for i in _range(D): 

if np.ndim(bins[i]) == 0: 

if bins[i] < 1: 

raise ValueError( 

'`bins[{}]` must be positive, when an integer'.format(i)) 

smin, smax = _get_outer_edges(sample[:,i], range[i]) 

edges[i] = np.linspace(smin, smax, bins[i] + 1) 

elif np.ndim(bins[i]) == 1: 

edges[i] = np.asarray(bins[i]) 

if np.any(edges[i][:-1] > edges[i][1:]): 

raise ValueError( 

'`bins[{}]` must be monotonically increasing, when an array' 

.format(i)) 

else: 

raise ValueError( 

'`bins[{}]` must be a scalar or 1d array'.format(i)) 

 

nbin[i] = len(edges[i]) + 1 # includes an outlier on each end 

dedges[i] = np.diff(edges[i]) 

 

# Compute the bin number each sample falls into. 

Ncount = tuple( 

# avoid np.digitize to work around gh-11022 

np.searchsorted(edges[i], sample[:, i], side='right') 

for i in _range(D) 

) 

 

# Using digitize, values that fall on an edge are put in the right bin. 

# For the rightmost bin, we want values equal to the right edge to be 

# counted in the last bin, and not as an outlier. 

for i in _range(D): 

# Find which points are on the rightmost edge. 

on_edge = (sample[:, i] == edges[i][-1]) 

# Shift these points one bin to the left. 

Ncount[i][on_edge] -= 1 

 

# Compute the sample indices in the flattened histogram matrix. 

# This raises an error if the array is too large. 

xy = np.ravel_multi_index(Ncount, nbin) 

 

# Compute the number of repetitions in xy and assign it to the 

# flattened histmat. 

hist = np.bincount(xy, weights, minlength=nbin.prod()) 

 

# Shape into a proper matrix 

hist = hist.reshape(nbin) 

 

# This preserves the (bad) behavior observed in gh-7845, for now. 

hist = hist.astype(float, casting='safe') 

 

# Remove outliers (indices 0 and -1 for each dimension). 

core = D*(slice(1, -1),) 

hist = hist[core] 

 

# handle the aliasing normed argument 

if normed is None: 

if density is None: 

density = False 

elif density is None: 

# an explicit normed argument was passed, alias it to the new name 

density = normed 

else: 

raise TypeError("Cannot specify both 'normed' and 'density'") 

 

if density: 

# calculate the probability density function 

s = hist.sum() 

for i in _range(D): 

shape = np.ones(D, int) 

shape[i] = nbin[i] - 2 

hist = hist / dedges[i].reshape(shape) 

hist /= s 

 

if (hist.shape != nbin - 2).any(): 

raise RuntimeError( 

"Internal Shape Error") 

return hist, edges