Newer
Older
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
{
"cells": [
{
"attachments": {},
"cell_type": "markdown",
"metadata": {
"id": "aXLlx8vXQlJw"
},
"source": [
"# Zero Shot Topic Classification with Transformers\n",
"\n",
"https://joeddav.github.io/blog/2020/05/29/ZSL.html\n",
"\n",
"https://colab.research.google.com/github/joeddav/blog/blob/master/_notebooks/2020-05-29-ZSL.ipynb#scrollTo=La_ga8KvSFYd\n",
"\n",
"https://huggingface.co/spaces/joeddav/zero-shot-demo"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {
"id": "3kYI_pq3Q1BT"
},
"source": [
"## 1. Configuration"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {
"id": "P_L0rDhZQ6Fn"
},
"source": [
"### 1.1 Setup colab environment"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "FsAR4CsB3aUc",
"outputId": "e0791012-6858-4ee0-f724-7f33c6985ee8"
},
"outputs": [],
"source": [
"from psutil import virtual_memory\n",
"ram_gb = virtual_memory().total / 1e9\n",
"print('Your runtime has {:.1f} gigabytes of available RAM\\n'.format(ram_gb))\n",
"\n",
"if ram_gb < 20:\n",
" print('Not using a high-RAM runtime')\n",
"else:\n",
" print('You are using a high-RAM runtime!')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "h5MwRwL53aYY",
"outputId": "20a93907-e5df-47b1-9172-d1693ef76dc5"
},
"outputs": [],
"source": [
"from google.colab import drive\n",
"drive.mount('/content/drive')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"output_path = \"drive/MyDrive/Classification-EDdA/\""
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {
"id": "4z78CLYi75kV"
},
"source": [
"### 1.2 Import libraries"
]
},
{
"cell_type": "code",
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
"metadata": {
"id": "bcptSr6o3ac7"
},
"outputs": [],
"source": [
"import pandas as pd\n",
"\n",
"from transformers import BartForSequenceClassification, BartTokenizer\n"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {
"id": "Lc1DRh4b7mto"
},
"source": [
"## 2. Load datasets"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"#### 2.1 Download datasets"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "ybiJYL0h3ahh",
"outputId": "0638f9a2-f9a0-4d96-9760-991ddc5747ca"
},
"outputs": [],
"source": [
"!wget https://geode.liris.cnrs.fr/EDdA-Classification/datasets/EDdA_dataframe_withContent.tsv\n",
"!wget https://geode.liris.cnrs.fr/EDdA-Classification/datasets/training_set.tsv\n",
"!wget https://geode.liris.cnrs.fr/EDdA-Classification/datasets/test_set.tsv"
]
},
{
"cell_type": "code",
"metadata": {},
"outputs": [],
"source": [
"dataset_path = 'EDdA_dataframe_withContent.tsv'\n",
"training_set_path = 'training_set.tsv'\n",
"test_set_path = 'test_set.tsv'\n",
"\n",
"input_path = '/Users/lmoncla/Nextcloud-LIRIS/GEODE/GEODE - Partage consortium/Classification domaines EDdA/datasets/'\n",
"#input_path = ''\n",
"output_path = ''"
]
},
{
"cell_type": "code",
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "LRKJzWmf3pCg",
"outputId": "686c3ef4-8267-4266-95af-7193725aadca"
},
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
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>volume</th>\n",
" <th>numero</th>\n",
" <th>head</th>\n",
" <th>normClass</th>\n",
" <th>classEDdA</th>\n",
" <th>author</th>\n",
" <th>id_enccre</th>\n",
" <th>domaine_enccre</th>\n",
" <th>ensemble_domaine_enccre</th>\n",
" <th>content</th>\n",
" <th>contentWithoutClass</th>\n",
" <th>firstParagraph</th>\n",
" <th>nb_word</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>11</td>\n",
" <td>2973</td>\n",
" <td>ORNIS</td>\n",
" <td>Commerce</td>\n",
" <td>Comm.</td>\n",
" <td>unsigned</td>\n",
" <td>v11-1767-0</td>\n",
" <td>commerce</td>\n",
" <td>Commerce</td>\n",
" <td>ORNIS, s. m. toile des Indes, (Comm.) sortes d...</td>\n",
" <td>ORNIS, s. m. toile des Indes, () sortes de\\nto...</td>\n",
" <td>ORNIS, s. m. toile des Indes, () sortes de\\nto...</td>\n",
" <td>45</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>3</td>\n",
" <td>3525</td>\n",
" <td>COMPRENDRE</td>\n",
" <td>Philosophie</td>\n",
" <td>terme de Philosophie,</td>\n",
" <td>Diderot</td>\n",
" <td>v3-1722-0</td>\n",
" <td>NaN</td>\n",
" <td>NaN</td>\n",
" <td>* COMPRENDRE, v. act. terme de Philosophie,\\nc...</td>\n",
" <td>* COMPRENDRE, v. act. \\nc'est appercevoir la l...</td>\n",
" <td>* COMPRENDRE, v. act. \\nc'est appercevoir la l...</td>\n",
" <td>92</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>1</td>\n",
" <td>2560</td>\n",
" <td>ANCRE</td>\n",
" <td>Marine</td>\n",
" <td>Marine</td>\n",
" <td>d'Alembert & Diderot</td>\n",
" <td>v1-1865-0</td>\n",
" <td>marine</td>\n",
" <td>Marine</td>\n",
" <td>ANCRE, s. f. (Marine.) est un instrument de fe...</td>\n",
" <td>ANCRE, s. f. (.) est un instrument de fer\\nABC...</td>\n",
" <td>ANCRE, s. f. (.) est un instrument de fer\\nABC...</td>\n",
" <td>3327</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>16</td>\n",
" <td>4241</td>\n",
" <td>VAKEBARO</td>\n",
" <td>Géographie moderne</td>\n",
" <td>Géog. mod.</td>\n",
" <td>unsigned</td>\n",
" <td>v16-2587-0</td>\n",
" <td>géographie</td>\n",
" <td>Géographie</td>\n",
" <td>VAKEBARO, (Géog. mod.) vallée du royaume\\nd'Es...</td>\n",
" <td>VAKEBARO, () vallée du royaume\\nd'Espagne dans...</td>\n",
" <td>VAKEBARO, () vallée du royaume\\nd'Espagne dans...</td>\n",
" <td>34</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>8</td>\n",
" <td>3281</td>\n",
" <td>INSPECTEUR</td>\n",
" <td>Histoire ancienne</td>\n",
" <td>Hist. anc.</td>\n",
" <td>unsigned</td>\n",
" <td>v8-2533-0</td>\n",
" <td>histoire</td>\n",
" <td>Histoire</td>\n",
" <td>INSPECTEUR, s. m. inspector ; (Hist. anc.) cel...</td>\n",
" <td>INSPECTEUR, s. m. inspector ; () celui \\nà qui...</td>\n",
" <td>INSPECTEUR, s. m. inspector ; () celui \\nà qui...</td>\n",
" <td>102</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" volume numero head normClass classEDdA \\\n",
"0 11 2973 ORNIS Commerce Comm. \n",
"1 3 3525 COMPRENDRE Philosophie terme de Philosophie, \n",
"2 1 2560 ANCRE Marine Marine \n",
"3 16 4241 VAKEBARO Géographie moderne Géog. mod. \n",
"4 8 3281 INSPECTEUR Histoire ancienne Hist. anc. \n",
"\n",
" author id_enccre domaine_enccre ensemble_domaine_enccre \\\n",
"0 unsigned v11-1767-0 commerce Commerce \n",
"1 Diderot v3-1722-0 NaN NaN \n",
"2 d'Alembert & Diderot v1-1865-0 marine Marine \n",
"3 unsigned v16-2587-0 géographie Géographie \n",
"4 unsigned v8-2533-0 histoire Histoire \n",
"\n",
" content \\\n",
"0 ORNIS, s. m. toile des Indes, (Comm.) sortes d... \n",
"1 * COMPRENDRE, v. act. terme de Philosophie,\\nc... \n",
"2 ANCRE, s. f. (Marine.) est un instrument de fe... \n",
"3 VAKEBARO, (Géog. mod.) vallée du royaume\\nd'Es... \n",
"4 INSPECTEUR, s. m. inspector ; (Hist. anc.) cel... \n",
"\n",
" contentWithoutClass \\\n",
"0 ORNIS, s. m. toile des Indes, () sortes de\\nto... \n",
"1 * COMPRENDRE, v. act. \\nc'est appercevoir la l... \n",
"2 ANCRE, s. f. (.) est un instrument de fer\\nABC... \n",
"3 VAKEBARO, () vallée du royaume\\nd'Espagne dans... \n",
"4 INSPECTEUR, s. m. inspector ; () celui \\nà qui... \n",
"\n",
" firstParagraph nb_word \n",
"0 ORNIS, s. m. toile des Indes, () sortes de\\nto... 45 \n",
"1 * COMPRENDRE, v. act. \\nc'est appercevoir la l... 92 \n",
"2 ANCRE, s. f. (.) est un instrument de fer\\nABC... 3327 \n",
"3 VAKEBARO, () vallée du royaume\\nd'Espagne dans... 34 \n",
"4 INSPECTEUR, s. m. inspector ; () celui \\nà qui... 102 "
]
},
"execution_count": 18,
"metadata": {},
"output_type": "execute_result"
}
],
"df = pd.read_csv(input_path + test_set_path, sep=\"\\t\")\n",
"df.head()"
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"(15854, 13)"
]
},
"execution_count": 19,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df.shape"
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {},
"outputs": [],
"source": [
"#column_text = 'contentWithoutClass'\n",
"column_text = 'content'\n",
"column_class = 'ensemble_domaine_enccre'"
]
},
{
"cell_type": "code",
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
"execution_count": 22,
"metadata": {},
"outputs": [],
"source": [
"df = df.dropna(subset=[column_text, column_class]).reset_index(drop=True)"
]
},
{
"cell_type": "code",
"execution_count": 23,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"(13441, 13)"
]
},
"execution_count": 23,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df.shape"
]
},
{
"cell_type": "code",
"execution_count": 32,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"['Commerce',\n",
" 'Marine',\n",
" 'Géographie',\n",
" 'Histoire',\n",
" 'Belles-lettres - Poésie',\n",
" 'Economie domestique',\n",
" 'Droit - Jurisprudence',\n",
" 'Médecine - Chirurgie',\n",
" 'Militaire (Art) - Guerre - Arme',\n",
" 'Beaux-arts',\n",
" 'Antiquité',\n",
" 'Histoire naturelle',\n",
" 'Grammaire',\n",
" 'Philosophie',\n",
" 'Arts et métiers',\n",
" 'Pharmacie',\n",
" 'Religion',\n",
" 'Pêche',\n",
" 'Anatomie',\n",
" 'Architecture',\n",
" 'Musique',\n",
" 'Jeu',\n",
" 'Caractères',\n",
" 'Métiers',\n",
" 'Physique - [Sciences physico-mathématiques]',\n",
" 'Maréchage - Manège',\n",
" 'Chimie',\n",
" 'Blason',\n",
" 'Chasse',\n",
" 'Mathématiques',\n",
" 'Médailles',\n",
" 'Superstition',\n",
" 'Agriculture - Economie rustique',\n",
" 'Mesure',\n",
" 'Monnaie',\n",
" 'Minéralogie',\n",
" 'Politique',\n",
" 'Spectacle']"
]
},
"execution_count": 32,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"classes = df[column_class].unique().tolist()\n",
"classes"
]
},
{
"cell_type": "code",
"execution_count": 35,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 110
},
"id": "RbsHOiJdNYRL",
"outputId": "bbdafc35-cf09-4a20-c3c0-901b8adce561"
},
"outputs": [
{
"data": {
"text/plain": [
"\"ORNIS, s. m. toile des Indes, (Comm.) sortes de\\ntoiles de coton ou de mousseline, qui se font a Brampour ville de l'Indoustan, entre Surate & Agra. Ces\\ntoiles sont par bandes, moitié coton & moitié or &\\nargent. Il y en a depuis quinze jusqu'à vingt aunes.\""
]
},
"execution_count": 35,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df[column_text].tolist()[0]"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"## 3. Classification"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"The approach, proposed by [Yin et al. (2019)](https://arxiv.org/abs/1909.00161), uses a pre-trained MNLI sequence-pair classifier as an out-of-the-box zero-shot text classifier that actually works pretty well. The idea is to take the sequence we're interested in labeling as the \"premise\" and to turn each candidate label into a \"hypothesis.\" If the NLI model predicts that the premise \"entails\" the hypothesis, we take the label to be true. See the code snippet below which demonstrates how easily this can be done with 🤗 Transformers."
]
},
{
"cell_type": "code",
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
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "e5a45c55993f47019fbdc0aceda84def",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Downloading: 0%| | 0.00/899k [00:00<?, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "a7285f3fe7154920a0bb05fdb921d6f9",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Downloading: 0%| | 0.00/456k [00:00<?, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "0cb38fe0e7934c49bbce246e88dd6e53",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Downloading: 0%| | 0.00/26.0 [00:00<?, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "2c9dd205446a4b25848f1f06beeac8ae",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Downloading: 0%| | 0.00/1.15k [00:00<?, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "032334b5310d4588993b7d85329d916c",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Downloading: 0%| | 0.00/1.63G [00:00<?, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"# load model pretrained on MNLI\n",
"tokenizer = BartTokenizer.from_pretrained('facebook/bart-large-mnli')\n",
"model = BartForSequenceClassification.from_pretrained('facebook/bart-large-mnli')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"''' \n",
"## Example from: https://joeddav.github.io/blog/2020/05/29/ZSL.html\n",
"\n",
"# load model pretrained on MNLI\n",
"tokenizer = BartTokenizer.from_pretrained('facebook/bart-large-mnli')\n",
"model = BartForSequenceClassification.from_pretrained('facebook/bart-large-mnli')\n",
"\n",
"# pose sequence as a NLI premise and label (politics) as a hypothesis\n",
"premise = 'Who are you voting for in 2020?'\n",
"hypothesis = 'This text is about politics.'\n",
"\n",
"# run through model pre-trained on MNLI\n",
"input_ids = tokenizer.encode(premise, hypothesis, return_tensors='pt')\n",
"logits = model(input_ids)[0]\n",
"\n",
"# we throw away \"neutral\" (dim 1) and take the probability of\n",
"# \"entailment\" (2) as the probability of the label being true \n",
"entail_contradiction_logits = logits[:,[0,2]]\n",
"probs = entail_contradiction_logits.softmax(dim=1)\n",
"true_prob = probs[:,1].item() * 100\n",
"print(f'Probability that the label is true: {true_prob:0.2f}%')\n",
"'''"
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
},
{
"cell_type": "code",
"execution_count": 36,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"The hypothesis with the highest score is: \"Commerce\" with a probability of 70.05%\n"
]
}
],
"source": [
"# pose sequence as a NLI premise and label (politics) as a hypothesis\n",
"premise = df[column_text].tolist()[0]\n",
"#hypothesis = 'This text is about politics.'\n",
"hypotheses = classes\n",
"\n",
"# list to store the true probability of each hypothesis\n",
"true_probs = []\n",
"\n",
"# loop through hypotheses\n",
"for hypothesis in hypotheses:\n",
"\n",
" # run through model pre-trained on MNLI\n",
" input_ids = tokenizer.encode(premise, hypothesis, return_tensors='pt')\n",
" logits = model(input_ids)[0]\n",
"\n",
" # we throw away \"neutral\" (dim 1) and take the probability of\n",
" # \"entailment\" (2) as the probability of the label being true \n",
" entail_contradiction_logits = logits[:,[0,2]]\n",
" probs = entail_contradiction_logits.softmax(dim=1)\n",
" true_prob = probs[:,1].item() * 100\n",
"\n",
" # append true probability to list\n",
" true_probs.append(true_prob)\n",
"\n",
"# print the true probability for each hypothesis\n",
"#for i, hypothesis in enumerate(hypotheses):\n",
"# print(f'Probability that hypothesis \"{hypothesis}\" is true: {true_probs[i]:0.2f}%')\n",
"# print(f'Probability that the label is true: {true_prob:0.2f}%')\n",
"\n",
"# get index of hypothesis with highest score\n",
"highest_index = max(range(len(true_probs)), key=lambda i: true_probs[i])\n",
"\n",
"# get hypothesis with highest score\n",
"highest_hypothesis = hypotheses[highest_index]\n",
"\n",
"# get highest probability\n",
"highest_prob = true_probs[highest_index]\n",
"\n",
"# print the results\n",
"print(f'The hypothesis with the highest score is: \"{highest_hypothesis}\" with a probability of {highest_prob:0.2f}%')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"colab": {
"collapsed_sections": [],
"machine_shape": "hm",
"name": "EDdA-Classification_Clustering.ipynb",
"provenance": []
},
"kernelspec": {
"display_name": "geode-classification-py39",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.13"