Hide keyboard shortcuts

Hot-keys on this page

r m x p   toggle line displays

j k   next/prev highlighted chunk

0   (zero) top of page

1   (one) first highlighted chunk

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

#!/usr/bin/env python3 

# -*- coding: utf-8 -*- 

 

"""Preprocessing. 

 

This module contains functions for various preprocessing steps 

provided by `DARIAH-DE`_. 

 

.. _DARIAH-DE: 

https://de.dariah.eu 

https://github.com/DARIAH-DE 

""" 

 

__author__ = "DARIAH-DE" 

__authors__ = "Steffen Pielstroem, Philip Duerholt, Sina Bock, Severin Simmler" 

__email__ = "pielstroem@biozentrum.uni-wuerzburg.de" 

 

from collections import Counter, defaultdict 

import csv 

import glob 

from itertools import chain 

import logging 

from lxml import etree 

import numpy as np 

import os 

import pandas as pd 

from pathlib import Path 

from itertools import zip_longest 

from abc import abstractmethod, abstractproperty 

from copy import deepcopy 

import regex 

 

 

log = logging.getLogger('preprocessing') 

log.addHandler(logging.NullHandler()) 

logging.basicConfig(level=logging.INFO, 

format='%(levelname)s %(name)s: %(message)s') 

 

regular_expression = r'\p{Letter}+\p{Punctuation}?\p{Letter}+' 

 

 

class BaseDocList: 

""" 

 

""" 

 

def __init__(self, basepath): 

self.basepath = Path(basepath) 

self._segment_counts = None 

 

def copy(self): 

return deepcopy(self) 

 

def full_path(self, document, as_str=False): 

""" 

Constructs a full path for the given document. 

 

Args: 

document: this is one document in the way the subclass chooses to 

represent documents. 

as_str (bool): if True, the result is a str, otherwise it is a `Path` 

 

Notes: 

The default implementation passed document on to `Path()`. 

Implementers will most probably want to override this. 

""" 

path = Path(self.basepath, document) 

if as_str: 

path = str(path) 

return path 

 

@abstractmethod 

def get_docs(self): 

""" 

Returns a sequence of documents, in the form the implementing class 

chooses. 

 

Note: 

Subclasses may implement a method `_get_item(self, index)`, with 

index being integer or slice, to speed access up. 

""" 

pass 

 

def full_paths(self, as_str=False): 

""" 

Returns a list of full paths. Calls full_path. 

""" 

return [self.full_path(doc, as_str) for doc in self.get_docs()] 

 

@abstractmethod 

def label(self, document): 

""" 

Returns a label suitable for the document. 

""" 

pass 

 

def __iter__(self): 

""" 

When used as an iterable, this object looks like an iterable of full paths. 

""" 

return iter(self.full_paths(as_str=True)) 

 

def __len__(self): 

""" 

When used as a sequence, this object looks like a sequence of full paths. 

""" 

return len(self.get_docs()) 

 

def __getitem__(self, index): 

""" 

When used as a sequence, this object looks like a sequence of full paths. 

""" 

try: 

selection = self._getitem(index) 

except AttributeError: 

selection = self.get_docs()[index] 

 

if isinstance(index, slice): 

return [self.full_path(doc, as_str=True) for doc in selection] 

else: 

return self.full_path(selection, as_str=True) 

 

def labels(self): 

""" 

Returns a list of (human-readable) labels for all documents. 

""" 

return [self.label(doc) for doc in self.get_docs()] 

 

def flatten_segments(self, segmented_docs): 

""" 

Records and flattens segment counts according to the stream of documents. 

 

Assume you have three documents 

 

A : I am an example document 

B : Me too 

C : All examples reference themselves 

 

docs = SimpleDocList('.', filenames=['A','B','C']) 

 

Now, you have an (external) segmenter function that segments each document 

into segments each being at most two tokens long. The data structure your 

segmenter will produce looks similar to the following: 

 

segmented_corpus = \ 

[[['I', 'am'], ['an', 'example'], ['document']], 

[['Me', 'too']], 

[['All', 'examples'], ['reference', 'themselves']]] 

 

Now, if you run docs.flatten_segments(self), it will do two things: it will 

record how many segments each document has (A: 3, B: 1, C: 2), and it will 

return a structure flattened by one level as in the following: 

 

[['I', 'am'], ['an', 'example'], ['document'], ['Me', 'too'], 

['All', 'examples'], ['reference', 'themselves']] 

 

I.e. the result will look like a corpus of six shorter documents. This 

matches with the iteration you get when you call docs.segments(). 

 

Args: 

segmented_docs: Iterable of documents, each document being an 

iterable of segments. 

 

Returns: 

Iterable of segments. 

 

Notes: 

Instead of lists you will receive generators, but you can iterate 

over these as well. 

""" 

segment_counts = [] 

self._segment_counts = segment_counts 

for doc in segmented_docs: 

segment_counts.append(0) 

for segment in doc: 

segment_counts[-1] += 1 

yield segment 

 

def segment_counts(self): 

""" 

Returns an iterable of the number of each segments for each document. 

""" 

return self._segment_counts 

 

def segments(self): 

""" 

Yields a tuple (document, segment_no) for each segment, with document 

being the internal representation of each document and segment_count an 

integer starting at 0 

""" 

for document, segment_count in zip_longest(self.get_docs(), 

self.segment_counts()): 

if segment_count is None: 

yield (document, None) 

else: 

for segment_no in range(segment_count): 

yield (document, segment_no) 

 

def segment_filenames(self, 

format="{path.stem}.{segment:0{maxwidth}d}{path.suffix}", 

basepath=None, 

as_str=False): 

""" 

 

Args: 

pattern (str): A `strings.Formatter` pattern that describes how 

to form each filename. The following formatter variables are 

available: 

 

path (Path): original file path 

segment (int): current segment number 

maxwidth (int): number of digits required for the largest 

segment number overall 

Raises: 

ValueError: if no segments 

""" 

segment_counts = self.segment_counts() 

if segment_counts is None: 

raise ValueError("No segments recorded.") 

maxwidth = len(str(max(segment_counts))) 

if basepath is None: 

basepath = self.basepath 

 

for document, segment_no in self.segments(): 

filename = format.format(path=document, maxwidth=maxwidth, 

segment=segment_no) 

segment_path = Path(basepath, filename) 

if as_str: 

yield str(segment_path) 

else: 

yield segment_path 

 

 

 

 

 

class PathDocList(BaseDocList): 

""" 

Document list based on a list of Paths. 

""" 

 

def __init__(self, basepath, glob_pattern='*', filenames=None): 

""" 

Creates a new document list either from the given file names 

or by looking for files matching the glob_pattern in the basepath. 

 

Args: 

basepath (Path or str): Root directory where your corpus resides 

glob_pattern (str): A file glob pattern matching the files to 

include. 

filenames (list): An iterable of paths or file names relative to 

basepath. If `None`, look for files on the file system. 

""" 

self.basepath = Path(basepath) 

self._segment_counts = None 

if filenames is None: 

self._files = [p.relative_to(self.basepath) 

for p in self.basepath.glob(glob_pattern)] 

else: 

paths = (Path(name) for name in filenames) 

if glob_pattern is not None: 

paths = (path for path in paths if path.match(glob_pattern)) 

self._files = list(paths) 

 

def get_docs(self): 

return self._files 

 

def label(self, document): 

return document.stem 

 

def with_segment_files(self, basepath=None, **kwargs): 

""" 

Returns a copy of this list which has the recorded segment numbers 

incorporated into the file names. I.e., this version does not know 

anymore about segments but rather has a file name for each segment. 

 

Args: 

pattern (str): A `strings.Formatter` pattern that describes how 

to form each filename. The following formatter variables are 

available: 

 

path (Path): original file path 

segment (int): current segment number 

maxwidth (int): number of digits required for the largest 

segment number overall 

Raises: 

ValueError: if no segments 

""" 

segment_counts = self.segment_counts() 

if segment_counts is None: 

raise ValueError("No segments recorded.") 

if basepath is None: 

basepath = self.basepath 

result = self.copy() 

result._segment_counts = 0 

result.basepath = basepath 

result._files = list(self.segment_filenames(basepath='', **kwargs)) 

return result 

 

 

def read_from_txt(doclist): 

"""Opens TXT files using file paths. 

 

Description: 

With this function you can read plain text files. Commit a list of 

full paths or one single path as argument. 

Use the function `create_document_list()` to create a list of your text 

files. 

 

Args: 

doclist Union(list[str], str): List of all documents in the corpus 

or single path to TXT file. 

 

Yields: 

Document. 

 

Todo: 

* Separate metadata (author, header) 

 

Example: 

>>> list(read_from_txt('corpus_txt/Doyle_AScandalinBohemia.txt'))[0][:20] 

'A SCANDAL IN BOHEMIA' 

""" 

log.info("Accessing TXT documents ...") 

if isinstance(doclist, str): 

with open(doclist, 'r', encoding='utf-8') as f: 

yield f.read() 

elif isinstance(doclist, list): 

for file in doclist: 

with open(file, 'r', encoding='utf-8') as f: 

yield f.read() 

 

 

def read_from_tei(doclist): 

"""Opens TEI XML files using file paths. 

 

Description: 

With this function you can read TEI encoded XML files. Commit a list of 

full paths or one single path as argument. 

Use the function `create_document_list()` to create a list of your XML 

files. 

 

Args: 

doclist Union(list[str], str): List of all documents in the corpus 

or single path to TEI XML file. 

 

Yields: 

Document. 

 

Todo: 

* Seperate metadata (author, header)? 

 

Example: 

>>> list(read_from_tei('corpus_tei/Schnitzler_Amerika.xml'))[0][142:159] 

'Arthur Schnitzler' 

""" 

log.info("Accessing TEI XML documents ...") 

if not isinstance(doclist, list): 

doclist = [doclist] 

ns = dict(tei='http://www.tei-c.org/ns/1.0') 

for file in doclist: 

tree = etree.parse(file) 

text_el = tree.xpath('//tei:text', namespaces=ns)[0] 

yield "".join(text_el.xpath('.//text()')) 

 

 

def read_from_csv(doclist, columns=['ParagraphId', 'TokenId', 'Lemma', 'CPOS', 'NamedEntity']): 

"""Opens CSV files using file paths. 

 

Description: 

With this function you can read CSV files generated by `DARIAH-DKPro-Wrapper`_, 

a tool for natural language processing. Commit a list of full paths or 

one single path as argument. You also have the ability to select certain 

columns. 

Use the function `create_document_list()` to create a list of your CSV 

files. 

.. _DARIAH-DKPro-Wrapper: 

https://github.com/DARIAH-DE/DARIAH-DKPro-Wrapper 

 

Args: 

doclist Union(list[str], str): List of all documents in the corpus 

or single path to CSV file. 

columns (list[str]): List of CSV column names. 

Defaults to '['ParagraphId', 'TokenId', 'Lemma', 'CPOS', 'NamedEntity']'. 

 

Yields: 

Document. 

 

Todo: 

* Seperate metadata (author, header)? 

 

Example: 

>>> list(read_from_csv('corpus_csv/Doyle_AScandalinBohemia.txt.csv'))[0][:4] # doctest: +NORMALIZE_WHITESPACE 

ParagraphId TokenId Lemma CPOS NamedEntity 

0 0 0 a ART _ 

1 0 1 scandal NP _ 

2 0 2 in PP _ 

3 0 3 bohemia NP _ 

""" 

log.info("Accessing CSV documents ...") 

if isinstance(doclist, str): 

doclist = [doclist] 

for file in doclist: 

df = pd.read_csv(file, sep='\t', quoting=csv.QUOTE_NONE) 

yield df[columns] 

 

 

def tokenize(doc_txt, expression=regular_expression, lower=True, simple=False): 

"""Tokenizes with Unicode Regular Expressions. 

 

Description: 

With this function you can tokenize a document with a regular expression. 

You also have the ability to commit your own regular expression. The default 

expression is '\p{Letter}+\p{Punctuation}?\p{Letter}+', which means one or 

more letters, followed by one or no punctuation, followed by one or more 

letters. So one letter words won't match. 

In case you want to lower alls tokens, set the argument `lower` to True (it 

is by default). 

If you want a very simple and primitive tokenization, set the argument 

`simple` to True. 

Use the functions `read_from_txt()`, `read_from_tei()` or `read_from_csv()` 

to read your text files. 

 

Args: 

doc_txt (str): Document as string. 

expression (str): Regular expression to find tokens. 

lower (boolean): If True, lowers all words. Defaults to True. 

simple (boolean): Uses simple regular expression (r'\w+'). Defaults to False. 

If set to True, argument `expression` will be ignored. 

 

Yields: 

Tokens 

 

Example: 

>>> list(tokenize("This is one example text.")) 

['this', 'is', 'one', 'example', 'text'] 

""" 

if lower: 

doc_txt = doc_txt.lower() 

if simple: 

pattern = regex.compile(r'\w+') 

else: 

pattern = regex.compile(expression) 

doc_txt = regex.sub("\.", "", doc_txt) 

doc_txt = regex.sub("‒", " ", doc_txt) 

doc_txt = regex.sub("–", " ", doc_txt) 

doc_txt = regex.sub("—", " ", doc_txt) 

doc_txt = regex.sub("―", " ", doc_txt) 

tokens = pattern.finditer(doc_txt) 

for match in tokens: 

yield match.group() 

 

 

def filter_pos_tags(doc_csv, pos_tags=['ADJ', 'V', 'NN']): 

"""Gets lemmas by selected POS-tags from DARIAH-DKPro-Wrapper output. 

 

Description: 

With this function you can select certain columns of a CSV file 

generated by `DARIAH-DKPro-Wrapper`_, a tool for natural language processing. 

Use the function `read_from_csv()` to read CSV files. 

.. _DARIAH-DKPro-Wrapper: 

https://github.com/DARIAH-DE/DARIAH-DKPro-Wrapper 

 

Args: 

doc_csv (DataFrame): DataFrame containing DARIAH-DKPro-Wrapper output. 

pos_tags (list[str]): List of DKPro POS-tags that should be selected. 

Defaults to '['ADJ', 'V', 'NN']'. 

 

Yields: 

Lemma. 

 

Example: 

>>> df = pd.DataFrame({'CPOS': ['CARD', 'ADJ', 'NN', 'NN'], 

... 'Lemma': ['one', 'more', 'example', 'text']}) 

>>> list(filter_pos_tags(df))[0] # doctest: +NORMALIZE_WHITESPACE 

1 more 

2 example 

3 text 

Name: Lemma, dtype: object 

""" 

log.info("Accessing %s ...", pos_tags) 

doc_csv = doc_csv[doc_csv['CPOS'].isin(pos_tags)] 

yield doc_csv['Lemma'] 

 

 

def split_paragraphs(doc_txt, sep=regex.compile('\n')): 

"""Splits the given document by paragraphs. 

 

Description: 

With this function you can split a document by paragraphs. You also have 

the ability to select a certain regular expression to split the document. 

Use the functions `read_from_txt()`, `read_from_tei()` or `read_from_csv()` 

to read your text files. 

 

Args: 

doc_txt (str): Document text. 

sep (regex.Regex): Separator indicating a paragraph. 

 

Returns: 

List of paragraphs. 

 

Example: 

>>> split_paragraphs("This test contains \\n paragraphs.") 

['This test contains ', ' paragraphs.'] 

""" 

if not hasattr(sep, 'match'): 

sep = regex.compile(sep) 

return sep.split(doc_txt) 

 

 

def segment_fuzzy(document, segment_size=5000, tolerance=0.05): 

"""Segments a document, tolerating existing chunks (like paragraphs). 

 

Description: 

Consider you have a document. You wish to split the document into 

segments of about 1000 tokens, but you prefer to keep paragraphs together 

if this does not increase or decrease the token size by more than 5%. 

 

Args: 

document: The document to process. This is an Iterable of chunks, each 

of which is an iterable of tokens. 

segment_size (int): The target length of each segment in tokens. 

tolerance (Number): How much may the actual segment size differ from 

the segment_size? If 0 < tolerance < 1, this is interpreted as a 

fraction of the segment_size, otherwise it is interpreted as an 

absolute number. If tolerance < 0, chunks are never split apart. 

 

Yields: 

Segments. Each segment is a list of chunks, each chunk is a list of 

tokens. 

 

Example: 

>>> list(segment_fuzzy([['This', 'test', 'is', 'very', 'clear'], 

... ['and', 'contains', 'chunks']], 2)) # doctest: +NORMALIZE_WHITESPACE 

[[['This', 'test']], 

[['is', 'very']], 

[['clear'], ['and']], 

[['contains', 'chunks']]] 

""" 

if tolerance > 0 and tolerance < 1: 

tolerance = round(segment_size * tolerance) 

 

current_segment = [] 

current_size = 0 

carry = None 

doc_iter = iter(document) 

 

try: 

while True: 

chunk = list(carry if carry else next(doc_iter)) 

carry = None 

current_segment.append(chunk) 

current_size += len(chunk) 

 

if current_size >= segment_size: 

too_long = current_size - segment_size 

too_short = segment_size - (current_size - len(chunk)) 

 

if tolerance >= 0 and min(too_long, too_short) > tolerance: 

chunk_part0 = chunk[:-too_long] 

carry = chunk[-too_long:] 

current_segment[-1] = chunk_part0 

elif too_long >= too_short: 

carry = current_segment.pop() 

yield current_segment 

current_segment = [] 

current_size = 0 

except StopIteration: 

pass 

 

# handle leftovers 

if current_segment: 

yield current_segment 

 

 

def segment(document, segment_size=1000, tolerance=0, chunker=None, 

tokenizer=None, flatten_chunks=False, materialize=False): 

"""Segments a document into segments of about `segment_size` tokens, respecting existing chunks. 

 

Description: 

Consider you have a document. You wish to split the document into 

segments of about 1000 tokens, but you prefer to keep paragraphs together 

if this does not increase or decrease the token size by more than 5%. 

This is a convenience wrapper around `segment_fuzzy()`. 

 

Args: 

segment_size (int): The target size of each segment, in tokens. 

tolerance (Number): see `segment_fuzzy` 

chunker (callable): a one-argument function that cuts the document into 

chunks. If this is present, it is called on the given document. 

tokenizer (callable): a one-argument function that tokenizes each chunk. 

flatten_chunks (bool): if True, undo the effect of the chunker by 

chaining the chunks in each segment, thus each segment consists of 

tokens. This can also be a one-argument function in order to 

customize the un-chunking. 

 

Example: 

>>> list(segment([['This', 'test', 'is', 'very', 'clear'], 

... ['and', 'contains', 'chunks']], 2)) # doctest: +NORMALIZE_WHITESPACE 

[[['This', 'test']], 

[['is', 'very']], 

[['clear'], ['and']], 

[['contains', 'chunks']]] 

""" 

if chunker is not None: 

document = chunker(document) 

if tokenizer is not None: 

document = map(tokenizer, document) 

 

segments = segment_fuzzy(document, segment_size, tolerance) 

 

if flatten_chunks: 

if not callable(flatten_chunks): 

def flatten_chunks(segment): 

return list(chain.from_iterable(segment)) 

segments = map(flatten_chunks, segments) 

if materialize: 

segments = list(segments) 

 

return segments 

 

def remove_features_from_file(doc_token_list, features_to_be_removed): 

"""Removes features using feature list. 

 

Description: 

With this function you can remove features from ppreprocessed files. 

Commit a list of features. 

Use the function `tokenize()` to access your files. 

 

Args: 

doc_token_list Union(list[str], str): List of all documents in the corpus 

and their tokens. 

features_to_be_removed list[str]: List of features that should be 

removed 

Yields: 

cleaned token array 

 

Todo: 

 

Example: 

>>> doc_tokens = [['short', 'example', 'example', 'text', 'text']] 

>>> features_to_be_removed = ['example'] 

>>> test = remove_features_from_file(doc_tokens, features_to_be_removed) 

>>> list(test) 

[['short', 'text', 'text']] 

""" 

#log.info("Removing features ...") 

doc_token_array = np.array(doc_token_list) 

feature_array = np.array(features_to_be_removed) 

#get indices of features that should be deleted 

indices = np.where(np.in1d(doc_token_array, feature_array,)) 

doc_token_array = np.delete(doc_token_array, indices) 

yield doc_token_array.tolist() 

 

def create_mallet_import(doc_tokens_cleaned, doc_labels, outpath = os.path.join('tutorial_supplementals', 'mallet_input')): 

"""Creates files for mallet import. 

 

Description: 

With this function you can create preprocessed plain text files. 

Commit a list of full paths or one single path as argument. 

Use the function `remove_features_from_file()` to create a list of tokens 

per document. 

 

Args: 

doc_tokens_cleaned Union(list[str], str): List of tokens per document 

doc_labels list[str]: List of documents labels. 

 

Todo: 

 

Example: 

>>> doc_labels = ['examplefile'] 

>>> doc_tokens_cleaned = [['short', 'example', 'text']] 

>>> create_mallet_import(doc_tokens_cleaned, doc_labels) 

>>> outpath = os.path.join('tutorial_supplementals', 'mallet_input') 

>>> os.path.isfile(os.path.join(outpath, 'examplefile.txt')) 

True 

""" 

#log.info("Generating mallet input files ...") 

if not os.path.exists(outpath): 

os.makedirs(outpath) 

 

for tokens, label in zip(doc_tokens_cleaned, doc_labels): 

with open(os.path.join(outpath,label+'.txt'), 'w', encoding="utf-8") as f: 

f.write(str(tokens)) 

 

 

def create_dictionary(tokens): 

"""Creates a dictionary of unique tokens with identifier. 

 

Description: 

With this function you can create a dictionary of unique tokens as key 

and an identifier as value. 

Use the function `tokenize()` to tokenize your text files. 

 

Args: 

tokens (list): List of tokens. 

 

Returns: 

Dictionary. 

 

Example: 

>>> create_dictionary(['example']) 

{'example': 1} 

""" 

if all(isinstance(element, list) for element in tokens): 

tokens = {token for element in tokens for token in element} 

return {token: id_ for id_, token in enumerate(set(tokens), 1)} 

 

 

def _create_large_counter(doc_labels, doc_tokens, type_dictionary): 

"""Creates a dictionary of dictionaries. 

 

Description: 

Only the function `create_sparse_bow()` uses this private function to 

create a dictionary of dictionaries. 

The first level consists of the document label as key, and the dictionary 

of counts as value. The second level consists of token ID as key, and the 

count of tokens in document pairs as value. 

 

Args: 

doc_labels (list): List of doc labels. 

doc_tokens (list): List of tokens. 

type_dictionary (dict): Dictionary of {token: id}. 

 

Returns: 

Dictionary of dictionaries. 

 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'example', 'text', 'text']] 

>>> type_dictionary = {'short': 1, 'example': 2, 'text': 3} 

>>> isinstance(_create_large_counter(doc_labels, doc_tokens, type_dictionary), defaultdict) 

True 

""" 

largecounter = defaultdict(dict) 

for doc, tokens in zip(doc_labels, doc_tokens): 

largecounter[doc] = Counter( 

[type_dictionary[token] for token in tokens]) 

return largecounter 

 

 

def _create_sparse_index(largecounter): 

"""Creates a sparse index for pandas DataFrame. 

 

Description: 

Only the function `create_sparse_bow()` uses this private function to 

create a pandas multiindex out of tuples. 

The multiindex represents document ID to token IDs relations. 

 

Args: 

largecounter (dict): Dictionary of {document: {token: frequency}}. 

 

Returns: 

Pandas MultiIndex. 

 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'example', 'text', 'text']] 

>>> type_dictionary = {'short': 1, 'example': 2, 'text': 3} 

>>> largecounter = _create_large_counter(doc_labels, doc_tokens, type_dictionary) 

>>> isinstance(_create_sparse_index(largecounter), pd.MultiIndex) 

True 

""" 

tuples = [] 

for key in range(1, len(largecounter) + 1): 

if len(largecounter[key]) == 0: 

tuples.append((key, 0)) 

for value in largecounter[key]: 

tuples.append((key, value)) 

sparse_index = pd.MultiIndex.from_tuples( 

tuples, names=['doc_id', 'token_id']) 

return sparse_index 

 

 

def create_sparse_bow(doc_labels, doc_tokens, type_dictionary, doc_dictionary): 

"""Creates sparse matrix for bag-of-words model. 

 

Description: 

This function creates a sparse DataFrame ('bow' means `bag-of-words`_) 

containing document and type identifier as multiindex and type 

frequencies as values representing the counts of tokens for each token 

in each document. 

It is also the main function that incorporates the private functions 

`_create_large_counter()` and `_create_sparse_index()``. 

Use the function `get_labels()` for `doc_labels`, `tokenize()` for 

`doc_tokens`, and `create_dictionary()` for `type_dictionary` as well 

as for `doc_ids`. 

Use the function `create_dictionary()` to generate the dictionaries 

`type_dictionary` and `doc_dictionary`. 

.. _bag-of-words: 

https://en.wikipedia.org/wiki/Bag-of-words_model 

 

Args: 

doc_labels (list[str]): List of doc labels as string. 

doc_tokens (list[str]): List of tokens as string. 

type_dictionary (dict[str]): Dictionary with {token: id}. 

doc_ids (dict[str]): Dictionary with {document label: id}. 

 

Returns: 

Multiindexed Pandas DataFrame. 

 

ToDo: 

* Test if it's necessary to build sparse_df_filled with int8 zeroes instead of int64. 

* Avoid saving sparse bow as .mm file to ingest into gensim. 

 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'text']] 

>>> type_dictionary = {'short': 1, 'example': 2, 'text': 3} 

>>> doc_ids = {'exampletext': 1} 

>>> len(create_sparse_bow(doc_labels, doc_tokens, type_dictionary, doc_ids)) 

3 

""" 

temp_counter = _create_large_counter( 

doc_labels, doc_tokens, type_dictionary) 

largecounter = {doc_dictionary[key]: value for key, value in temp_counter.items()} 

sparse_index = _create_sparse_index(largecounter) 

sparse_bow_filled = pd.DataFrame( 

np.zeros((len(sparse_index), 1), dtype=int), index=sparse_index) 

index_iterator = sparse_index.groupby( 

sparse_index.get_level_values('doc_id')) 

 

for doc_id in range(1, len(sparse_index.levels[0]) + 1): 

for token_id in [val[1] for val in index_iterator[doc_id]]: 

sparse_bow_filled.set_value( 

(doc_id, token_id), 0, int(largecounter[doc_id][token_id])) 

return sparse_bow_filled 

 

 

def save_sparse_bow(sparse_bow, output): 

"""Saves sparse matrix for bag-of-words model. 

 

Description: 

With this function you can save the sparse matrix as `.mm file`_. 

.. _.mm file: http://math.nist.gov/MatrixMarket/formats.html#MMformat 

 

Args: 

sparse_bow (DataFrame): DataFrame with term and term frequency by document. 

output (str): Path to output file without extension, e.g. /tmp/sparsebow. 

 

Returns: 

None. 

 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'text']] 

>>> type_dictionary = {'short': 1, 'example': 2, 'text': 3} 

>>> doc_ids = {'exampletext': 1} 

>>> sparse_bow = create_sparse_bow(doc_labels, doc_tokens, type_dictionary, doc_ids) 

>>> save_sparse_bow(sparse_bow, 'sparsebow') 

>>> import os.path 

>>> os.path.isfile('sparsebow.mm') 

True 

""" 

num_docs = sparse_bow.index.get_level_values("doc_id").max() 

num_types = sparse_bow.index.get_level_values("token_id").max() 

sum_counts = sparse_bow[0].sum() 

 

header_string = str(num_docs) + " " + str(num_types) + \ 

" " + str(sum_counts) + "\n" 

 

with open('.'.join([output, 'mm']), 'w', encoding="utf-8") as f: 

f.write("%%MatrixMarket matrix coordinate real general\n") 

f.write(header_string) 

sparse_bow.to_csv(f, sep=' ', header=None) 

 

 

def find_stopwords(sparse_bow, id_types, mfw=200): 

"""Creates a stopword list. 

 

Description: 

With this function you can determine most frequent words, also known as 

stopwords. First, you have to translate your corpus into the bag-of-words 

model using the function `create_sparse_matrix()` and create an dictionary 

containing types and identifier using `create_dictionary()`. 

 

Args: 

sparse_bow (DataFrame): DataFrame with term and term frequency by document. 

id_types (dict[str]): Dictionary with {token: id}. 

mfw (int): Target size of most frequent words to be considered. 

 

Returns: 

Most frequent words in a list. 

 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'short', 'example', 'text']] 

>>> id_types = {'short': 1, 'example': 2, 'text': 3} 

>>> doc_ids = {'exampletext': 1} 

>>> sparse_bow = create_sparse_bow(doc_labels, doc_tokens, id_types, doc_ids) 

>>> find_stopwords(sparse_bow, id_types, 1) 

['short'] 

""" 

log.info("Finding stopwords ...") 

type2id = {value: key for key, value in id_types.items()} 

sparse_bow_collapsed = sparse_bow.groupby( 

sparse_bow.index.get_level_values('token_id')).sum() 

sparse_bow_stopwords = sparse_bow_collapsed[0].nlargest(mfw) 

stopwords = [type2id[key] 

for key in sparse_bow_stopwords.index.get_level_values('token_id')] 

return stopwords 

 

 

def find_hapax(sparse_bow, id_types): 

"""Creates a list with hapax legommena. 

 

Description: 

With this function you can determine hapax legomena for each document. 

First, you have to translate your corpus into the bag-of-words 

model using the function `create_sparse_matrix()` and create an dictionary 

containing types and identifier using `create_dictionary()`. 

 

Args: 

sparse_bow (DataFrame): DataFrame with term and term frequency by document. 

id_types (dict[str]): Dictionary with {token: id}. 

 

Returns: 

Hapax legomena in a list. 

 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'example', 'text', 'text']] 

>>> id_types = {'short': 1, 'example': 2, 'text': 3} 

>>> doc_ids = {'exampletext': 1} 

>>> sparse_bow = create_sparse_bow(doc_labels, doc_tokens, id_types, doc_ids) 

>>> find_hapax(sparse_bow, id_types) 

['short'] 

""" 

log.info("Finding hapax legomena ...") 

type2id = {value: key for key, value in id_types.items()} 

sparse_bow_collapsed = sparse_bow.groupby( 

sparse_bow.index.get_level_values('token_id')).sum() 

sparse_bow_hapax = sparse_bow_collapsed.loc[sparse_bow_collapsed[0] == 1] 

hapax = [type2id[key] 

for key in sparse_bow_hapax.index.get_level_values('token_id')] 

return hapax 

 

 

def remove_features(sparse_bow, id_types, features): 

"""Removes features based on a list of words (types). 

 

Description: 

With this function you can clean your corpus from stopwords and hapax 

legomena. 

First, you have to translate your corpus into the bag-of-words 

model using the function `create_sparse_bow()` and create a dictionary 

containing types and identifier using `create_dictionary()`. 

Use the functions `find_stopwords()` and `find_hapax()` to generate a 

feature list. 

 

Args: 

sparse_bow (DataFrame): DataFrame with term and term frequency by document. 

features Union(set, list): Set or list containing features to remove. 

(not included) features (str): Text as iterable. 

 

Returns: 

Clean corpus. 

 

ToDo: 

* Adapt function to work with mm-corpus format. 

 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'example', 'text', 'text']] 

>>> id_types = {'short': 1, 'example': 2, 'text': 3} 

>>> doc_ids = {'exampletext': 1} 

>>> sparse_bow = create_sparse_bow(doc_labels, doc_tokens, id_types, doc_ids) 

>>> features = ['short'] 

>>> len(remove_features(sparse_bow, id_types, features)) 

2 

""" 

log.info("Removing features ...") 

if isinstance(features, list): 

features = set(features) 

stoplist_applied = [word for word in set(id_types.keys()) if word in features] 

clean_sparse_bow = sparse_bow.drop([id_types[word] for word in stoplist_applied], level="token_id") 

log.debug("%s features removed.", len(features)) 

return clean_sparse_bow 

 

 

def make_doc2bow_list(sparse_bow): 

"""Creates doc2bow_list for gensim. 

 

Description: 

With this function you can create a doc2bow_list as input for the gensim 

function `get_document_topics()` to show topics for each document. 

 

Args: 

sparse_bow (DataFrame): DataFrame with term and term frequency by document. 

 

Returns: 

List of lists containing tuples. 

 

Example: 

>>> doc_labels = ['exampletext1', 'exampletext2'] 

>>> doc_tokens = [['test', 'corpus'], ['for', 'testing']] 

>>> type_dictionary = {'test': 1, 'corpus': 2, 'for': 3, 'testing': 4} 

>>> doc_dictionary = {'exampletext1': 1, 'exampletext2': 2} 

>>> sparse_bow = create_sparse_bow(doc_labels, doc_tokens, type_dictionary, doc_dictionary) 

>>> from gensim.models import LdaModel 

>>> from gensim.corpora import Dictionary 

>>> corpus = [['test', 'corpus'], ['for', 'testing']] 

>>> dictionary = Dictionary(corpus) 

>>> documents = [dictionary.doc2bow(document) for document in corpus] 

>>> model = LdaModel(corpus=documents, id2word=dictionary, iterations=1, passes=1, num_topics=1) 

>>> make_doc2bow_list(sparse_bow) 

[[(1, 1), (2, 1)], [(3, 1), (4, 1)]] 

""" 

doc2bow_list = [] 

for doc in sparse_bow.index.groupby(sparse_bow.index.get_level_values('doc_id')): 

temp = [(token, count) for token, count in zip( 

sparse_bow.loc[doc].index, sparse_bow.loc[doc][0])] 

doc2bow_list.append(temp) 

return doc2bow_list 

 

 

def gensim2dataframe(model, num_keys=10): 

"""Converts gensim output to DataFrame. 

 

Description: 

With this function you can convert gensim output (usually a list of 

tuples) to a DataFrame, a more convenient datastructure. 

 

Args: 

model: Gensim LDA model. 

num_keys (int): Number of top keywords for topic. 

 

Returns: 

DataFrame. 

 

ToDo: 

 

Example: 

>>> from gensim.models import LdaModel 

>>> from gensim.corpora import Dictionary 

>>> corpus = [['test', 'corpus'], ['for', 'testing']] 

>>> dictionary = Dictionary(corpus) 

>>> documents = [dictionary.doc2bow(document) for document in corpus] 

>>> model = LdaModel(corpus=documents, id2word=dictionary, iterations=1, passes=1, num_topics=1) 

>>> isinstance(gensim2dataframe(model, 4), pd.DataFrame) 

True 

""" 

# num_topics = model.num_topics 

# topics_df = pd.DataFrame(index=range(num_topics), columns=range(num_keys)) 

# topics = model.show_topics( 

# num_topics=num_topics, log=False, formatted=False) 

# for topic in topics: 

# idx = topic[0] 

# temp = topic[1] 

# topics_df.loc[idx] = temp 

 

num_topics = model.num_topics 

topics_df = pd.DataFrame(index = range(num_topics), 

columns= range(num_keys)) 

 

topics = model.show_topics(num_topics = model.num_topics, formatted=False) 

 

for topic, values in topics: 

keyword = [value[0] for value in values] 

topics_df.loc[topic] = keyword 

 

return topics_df 

 

 

def doctopic2dataframe(model, doc2bow_list, doc2id): 

"""Use only for testing purposes, not working properly 

 

Note: 

 

Args: 

 

Returns: 

 

ToDo: make it work 

""" 

df = pd.DataFrame() 

for idx, doc in enumerate(doc2bow_list, 1): 

df[doc2id[idx]] = pd.Series( 

[value[1] for value in model.get_document_topics(doc)]) 

return df.fillna(0)