Merge branch 'master' of ssh://the-cake-is-a-lie.net:20022/jonas/NLP-LAB

This commit is contained in:
Jonas Weinz 2018-04-29 19:52:11 +02:00
commit d5c17c09a8
4 changed files with 1381 additions and 16 deletions

View File

@ -0,0 +1,511 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Exercise 1"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"import nltk\n",
"from nltk import word_tokenize, pos_tag"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Classifiers\n",
"note: for model1 and model3 you can try different classifiers: Hidden Markov Model, Logistic Regression, Maximum Entropy Markov Models, Decision Trees, Naive Bayes, etc.. __choose one!__"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"from sklearn.tree import DecisionTreeClassifier\n",
"from sklearn.feature_extraction import DictVectorizer\n",
"from sklearn.pipeline import Pipeline"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 1. model1 = your POS tagger model (english)"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'word': 'bims', 'length': 4, 'is_capitalized': False, 'prefix-1': 'b', 'suffix-1': 's', 'prev_word': 'i', 'next_word': 'der'}\n"
]
}
],
"source": [
"def features(sentence, index):\n",
" return {\n",
" 'word': sentence[index],\n",
" 'length': len(sentence[index]),\n",
" 'is_capitalized': sentence[index][0].upper() == sentence[index][0],\n",
" 'prefix-1': sentence[index][0],\n",
" 'suffix-1': sentence[index][-1],\n",
" 'prev_word': '' if index == 0 else sentence[index - 1],\n",
" 'next_word': '' if index == len(sentence) - 1 else sentence[index + 1]\n",
" }\n",
"\n",
"print(features(\"halli hallo i bims der Programmierer\".strip().split(\" \"), 3))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 2. model2 = pre-trained POS tagger model using NLTK (maxentropy english)\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 3. model3.x = rule-based classifiers (x = 1 to 5)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 4. model4 = your POS tagger model (not english)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 5. model5 = pre-trained POS tagger model using RDRPOSTagger 1 or TreeTagger 2 (not english)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Corpora\n",
"note: data split for training/test = 0.8/0.2 (sequencial)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### 1. X1 = nltk.corpus.treebank (english)"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[nltk_data] Downloading package treebank to\n",
"[nltk_data] /Users/Carsten/nltk_data...\n",
"[nltk_data] Package treebank is already up-to-date!\n"
]
}
],
"source": [
"nltk.download('treebank')\n",
"x1 = nltk.corpus.treebank"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### 2. X2 = nltk.corpus.brown (english)"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[nltk_data] Downloading package brown to /Users/Carsten/nltk_data...\n",
"[nltk_data] Package brown is already up-to-date!\n"
]
}
],
"source": [
"nltk.download('brown')\n",
"x2 = nltk.corpus.brown"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### 3. X3 = other language (not english)"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#nltk.download('brown')\n",
"#x3 = other language"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Task 1\n",
"* get results for english (plot a graph with all classifiers x results)\n",
" * performance 1.1 = model1 in X1"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"##### Generate Training and Testdata\n",
"1. split annotaed sentences into training and testdata\n",
"2. split trainingdata into input data and teacherdata\n",
" *input is the feature vector of each word\n",
" *output is a list of POS tags for each word and sentences"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"got 3131 training sentences and 783 test sentences\n"
]
}
],
"source": [
"#to generate trainingsdata, delete the assigned tags as a function\n",
"def untag(tagged_sentence):\n",
" return [w for w, t in tagged_sentence]\n",
"\n",
"#object including the annotated sentences\n",
"annotated_sent = nltk.corpus.treebank.tagged_sents()\n",
"\n",
"#to split the data, calculate the borders for ratio\n",
"cutoff = int(.8 * len(annotated_sent))\n",
"training_sentences = annotated_sent[:cutoff]\n",
"test_sentences = annotated_sent[cutoff:]\n",
"\n",
"#show the amount of sentences\n",
"print(\"got \",len(training_sentences),\" training sentences and \", len(test_sentences), \" test sentences\")\n",
"\n",
"#for training split sentences with its tags into y (for a sentences its resulting tags for each word) and transform sentences and x as a list of the features extracet for echt word in the sentences\n",
"def transform_to_dataset(tagged_sentences):\n",
" X, y = [], []\n",
" for tagged_sentence in tagged_sentences:\n",
" for index in range(len(tagged_sentence)):\n",
" X.append(features(untag(tagged_sentence), index))\n",
" y.append(tagged_sentence[index][1]) \n",
" return X, y\n",
"\n",
"#trainings inputset X and training teacher set y\n",
"X, y = transform_to_dataset(training_sentences)"
]
},
{
"cell_type": "markdown",
"metadata": {
"collapsed": true
},
"source": [
"#### Implementing a classifier\n",
"relevant imports\n",
"* decision tree as the AI for classfing\n",
"* dict vercorizer transforms the feature dictionary into a vector as the input for the tree"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"from sklearn.tree import DecisionTreeClassifier\n",
"from sklearn.feature_extraction import DictVectorizer\n",
"from sklearn.pipeline import Pipeline"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Pipeline manages vectorizer and classifier"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"clf = Pipeline([\n",
" ('vectorizer', DictVectorizer(sparse=False)),\n",
" ('classifier', DecisionTreeClassifier(criterion='entropy'))\n",
"])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"##### Calculate performance 1.1 \n",
"* fit the decision tree for a limited amount (size) of training \n",
"* test data and compare with score function on testdata"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"training OK\n",
"Accuracy: 0.880832376865\n"
]
}
],
"source": [
"size=10000\n",
"clf.fit(X[:size], y[:size])\n",
" \n",
"print('training OK')\n",
" \n",
"X_test, y_test = transform_to_dataset(test_sentences)\n",
"\n",
"performance1_1 = clf.score(X_test, y_test)\n",
"\n",
"print(\"Accuracy:\", performance1_1)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"##### Calculate other performances"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"performance1_2 = 0\n",
"performance1_3 = 0\n",
"performance1_4 = 0\n",
"performance1_5 = 0\n",
"performance1_6 = 0"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### Using the classifier\n",
"for results the link of pos_tags:\n",
"https://www.ling.upenn.edu/courses/Fall_2003/ling001/penn_treebank_pos.html"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"3.6.3\n",
"checking...\n",
"[('Hello', 'NNP'), ('world', 'VBD'), (',', ','), ('lets', 'NNS'), ('do', 'VB'), ('something', 'VBG'), ('awesome', 'NN'), ('today', 'NN'), ('!', 'CD')]\n"
]
}
],
"source": [
"def pos_tag(sentence):\n",
" print('checking...')\n",
" tagged_sentence = []\n",
" tags = clf.predict([features(sentence, index) for index in range(len(sentence))])\n",
" return zip(sentence, tags)\n",
"\n",
"import platform\n",
"print(platform.python_version())\n",
"\n",
"print(list(pos_tag(word_tokenize('Hello world, lets do something awesome today!'))))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Results for Task 1\n",
"* get results for english (plot a graph with all classifiers x results)\n",
" * performance 1.1 = model1 in X1\n",
" * performance 1.2 = model2 in X1\n",
" * performance 1.3.x = model3.x in X1\n",
" * performance 1.4 = model1 in X2\n",
" * performance 1.5 = model2 in X2\n",
" * performance 1.6.x = model3.x in X2"
]
},
{
"cell_type": "code",
"execution_count": 22,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"High five! You successfully sent some data to your account on plotly. View your plot in your browser at https://plot.ly/~carsten95/0 or inside your plot.ly account where it is named 'basic-bar'\n"
]
},
{
"data": {
"text/html": [
"<iframe id=\"igraph\" scrolling=\"no\" style=\"border:none;\" seamless=\"seamless\" src=\"https://plot.ly/~carsten95/0.embed\" height=\"525px\" width=\"100%\"></iframe>"
],
"text/plain": [
"<plotly.tools.PlotlyDisplay object>"
]
},
"execution_count": 22,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"import plotly\n",
"plotly.tools.set_credentials_file(username='carsten95', api_key='vElf5IOxiFheQdjTxjXW')\n",
"plotly.__version__\n",
"import plotly.plotly as py\n",
"import plotly.graph_objs as go\n",
"\n",
"data = [go.Bar(\n",
" x=['performance 1.1', 'performance 1.2', 'performance 1.3', 'performance 1.4', 'performance 1.5' , 'performance 1.6'],\n",
" y=[performance1_1, performance1_2, performance1_3, performance1_4, performance1_5, performance1_6]\n",
" )]\n",
"\n",
"py.iplot(data, filename='basic-bar')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.6.3"
}
},
"nbformat": 4,
"nbformat_minor": 2
}

File diff suppressed because one or more lines are too long

View File

@ -0,0 +1,545 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# NLP Lab Task 1 SoSe 18\n",
"\n",
"## POS Tagger\n",
"due to 08.05.2018\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"import nltk\n",
"from nltk import word_tokenize, pos_tag"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[('Hi', 'NNP'), (',', ','), ('welcome', 'NN'), ('to', 'TO'), ('the', 'DT'), ('NLP', 'NNP'), ('lab', 'NN'), ('!', '.')]\n"
]
}
],
"source": [
"tokens = word_tokenize(\"Hi, welcome to the NLP lab!\")\n",
"print(pos_tag(tokens))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Exploring the Penn TreeBank (PTB) Corpus"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[nltk_data] Downloading package treebank to\n",
"[nltk_data] /Users/Carsten/nltk_data...\n",
"[nltk_data] Package treebank is already up-to-date!\n",
"[('Pierre', 'NNP'), ('Vinken', 'NNP'), (',', ','), ('61', 'CD'), ('years', 'NNS'), ('old', 'JJ'), (',', ','), ('will', 'MD'), ('join', 'VB'), ('the', 'DT'), ('board', 'NN'), ('as', 'IN'), ('a', 'DT'), ('nonexecutive', 'JJ'), ('director', 'NN'), ('Nov.', 'NNP'), ('29', 'CD'), ('.', '.')]\n",
"Tagged sentences: 3914\n",
"Tagged words: 100676\n",
"[nltk_data] Downloading package tagsets to /Users/Carsten/nltk_data...\n",
"[nltk_data] Package tagsets is already up-to-date!\n",
"$: dollar\n",
" $ -$ --$ A$ C$ HK$ M$ NZ$ S$ U.S.$ US$\n",
"'': closing quotation mark\n",
" ' ''\n",
"(: opening parenthesis\n",
" ( [ {\n",
"): closing parenthesis\n",
" ) ] }\n",
",: comma\n",
" ,\n",
"--: dash\n",
" --\n",
".: sentence terminator\n",
" . ! ?\n",
":: colon or ellipsis\n",
" : ; ...\n",
"CC: conjunction, coordinating\n",
" & 'n and both but either et for less minus neither nor or plus so\n",
" therefore times v. versus vs. whether yet\n",
"CD: numeral, cardinal\n",
" mid-1890 nine-thirty forty-two one-tenth ten million 0.5 one forty-\n",
" seven 1987 twenty '79 zero two 78-degrees eighty-four IX '60s .025\n",
" fifteen 271,124 dozen quintillion DM2,000 ...\n",
"DT: determiner\n",
" all an another any both del each either every half la many much nary\n",
" neither no some such that the them these this those\n",
"EX: existential there\n",
" there\n",
"FW: foreign word\n",
" gemeinschaft hund ich jeux habeas Haementeria Herr K'ang-si vous\n",
" lutihaw alai je jour objets salutaris fille quibusdam pas trop Monte\n",
" terram fiche oui corporis ...\n",
"IN: preposition or conjunction, subordinating\n",
" astride among uppon whether out inside pro despite on by throughout\n",
" below within for towards near behind atop around if like until below\n",
" next into if beside ...\n",
"JJ: adjective or numeral, ordinal\n",
" third ill-mannered pre-war regrettable oiled calamitous first separable\n",
" ectoplasmic battery-powered participatory fourth still-to-be-named\n",
" multilingual multi-disciplinary ...\n",
"JJR: adjective, comparative\n",
" bleaker braver breezier briefer brighter brisker broader bumper busier\n",
" calmer cheaper choosier cleaner clearer closer colder commoner costlier\n",
" cozier creamier crunchier cuter ...\n",
"JJS: adjective, superlative\n",
" calmest cheapest choicest classiest cleanest clearest closest commonest\n",
" corniest costliest crassest creepiest crudest cutest darkest deadliest\n",
" dearest deepest densest dinkiest ...\n",
"LS: list item marker\n",
" A A. B B. C C. D E F First G H I J K One SP-44001 SP-44002 SP-44005\n",
" SP-44007 Second Third Three Two * a b c d first five four one six three\n",
" two\n",
"MD: modal auxiliary\n",
" can cannot could couldn't dare may might must need ought shall should\n",
" shouldn't will would\n",
"NN: noun, common, singular or mass\n",
" common-carrier cabbage knuckle-duster Casino afghan shed thermostat\n",
" investment slide humour falloff slick wind hyena override subhumanity\n",
" machinist ...\n",
"NNP: noun, proper, singular\n",
" Motown Venneboerger Czestochwa Ranzer Conchita Trumplane Christos\n",
" Oceanside Escobar Kreisler Sawyer Cougar Yvette Ervin ODI Darryl CTCA\n",
" Shannon A.K.C. Meltex Liverpool ...\n",
"NNPS: noun, proper, plural\n",
" Americans Americas Amharas Amityvilles Amusements Anarcho-Syndicalists\n",
" Andalusians Andes Andruses Angels Animals Anthony Antilles Antiques\n",
" Apache Apaches Apocrypha ...\n",
"NNS: noun, common, plural\n",
" undergraduates scotches bric-a-brac products bodyguards facets coasts\n",
" divestitures storehouses designs clubs fragrances averages\n",
" subjectivists apprehensions muses factory-jobs ...\n",
"PDT: pre-determiner\n",
" all both half many quite such sure this\n",
"POS: genitive marker\n",
" ' 's\n",
"PRP: pronoun, personal\n",
" hers herself him himself hisself it itself me myself one oneself ours\n",
" ourselves ownself self she thee theirs them themselves they thou thy us\n",
"PRP$: pronoun, possessive\n",
" her his mine my our ours their thy your\n",
"RB: adverb\n",
" occasionally unabatingly maddeningly adventurously professedly\n",
" stirringly prominently technologically magisterially predominately\n",
" swiftly fiscally pitilessly ...\n",
"RBR: adverb, comparative\n",
" further gloomier grander graver greater grimmer harder harsher\n",
" healthier heavier higher however larger later leaner lengthier less-\n",
" perfectly lesser lonelier longer louder lower more ...\n",
"RBS: adverb, superlative\n",
" best biggest bluntest earliest farthest first furthest hardest\n",
" heartiest highest largest least less most nearest second tightest worst\n",
"RP: particle\n",
" aboard about across along apart around aside at away back before behind\n",
" by crop down ever fast for forth from go high i.e. in into just later\n",
" low more off on open out over per pie raising start teeth that through\n",
" under unto up up-pp upon whole with you\n",
"SYM: symbol\n",
" % & ' '' ''. ) ). * + ,. < = > @ A[fj] U.S U.S.S.R * ** ***\n",
"TO: \"to\" as preposition or infinitive marker\n",
" to\n",
"UH: interjection\n",
" Goodbye Goody Gosh Wow Jeepers Jee-sus Hubba Hey Kee-reist Oops amen\n",
" huh howdy uh dammit whammo shucks heck anyways whodunnit honey golly\n",
" man baby diddle hush sonuvabitch ...\n",
"VB: verb, base form\n",
" ask assemble assess assign assume atone attention avoid bake balkanize\n",
" bank begin behold believe bend benefit bevel beware bless boil bomb\n",
" boost brace break bring broil brush build ...\n",
"VBD: verb, past tense\n",
" dipped pleaded swiped regummed soaked tidied convened halted registered\n",
" cushioned exacted snubbed strode aimed adopted belied figgered\n",
" speculated wore appreciated contemplated ...\n",
"VBG: verb, present participle or gerund\n",
" telegraphing stirring focusing angering judging stalling lactating\n",
" hankerin' alleging veering capping approaching traveling besieging\n",
" encrypting interrupting erasing wincing ...\n",
"VBN: verb, past participle\n",
" multihulled dilapidated aerosolized chaired languished panelized used\n",
" experimented flourished imitated reunifed factored condensed sheared\n",
" unsettled primed dubbed desired ...\n",
"VBP: verb, present tense, not 3rd person singular\n",
" predominate wrap resort sue twist spill cure lengthen brush terminate\n",
" appear tend stray glisten obtain comprise detest tease attract\n",
" emphasize mold postpone sever return wag ...\n",
"VBZ: verb, present tense, 3rd person singular\n",
" bases reconstructs marks mixes displeases seals carps weaves snatches\n",
" slumps stretches authorizes smolders pictures emerges stockpiles\n",
" seduces fizzes uses bolsters slaps speaks pleads ...\n",
"WDT: WH-determiner\n",
" that what whatever which whichever\n",
"WP: WH-pronoun\n",
" that what whatever whatsoever which who whom whosoever\n",
"WP$: WH-pronoun, possessive\n",
" whose\n",
"WRB: Wh-adverb\n",
" how however whence whenever where whereby whereever wherein whereof why\n",
"``: opening quotation mark\n",
" ` ``\n",
"None\n"
]
}
],
"source": [
"nltk.download('treebank')\n",
"annotated_sent = nltk.corpus.treebank.tagged_sents()\n",
" \n",
"print(annotated_sent[0])\n",
"print(\"Tagged sentences: \", len(annotated_sent))\n",
"print(\"Tagged words:\", len(nltk.corpus.treebank.tagged_words()))\n",
"\n",
"# tagsets\n",
"nltk.download('tagsets')\n",
"print(nltk.help.upenn_tagset())"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Training"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'is_capitalized': False,\n",
" 'next_word': 'sentence',\n",
" 'prefix-1': 'a',\n",
" 'prev_word': 'is',\n",
" 'suffix-1': 'a',\n",
" 'word': 'a'}\n"
]
}
],
"source": [
"# TODO: improve this feature extraction function\n",
" \n",
"def features(sentence, index):\n",
" return {\n",
" 'word': sentence[index],\n",
" 'is_capitalized': sentence[index][0].upper() == sentence[index][0],\n",
" 'prefix-1': sentence[index][0],\n",
" 'suffix-1': sentence[index][-1],\n",
" 'prev_word': '' if index == 0 else sentence[index - 1],\n",
" 'next_word': '' if index == len(sentence) - 1 else sentence[index + 1]\n",
" }\n",
"import pprint \n",
"pprint.pprint(features(['This', 'is', 'a', 'sentence'], 2))"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"def untag(tagged_sentence):\n",
" return [w for w, t in tagged_sentence]"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"2935\n",
"979\n"
]
}
],
"source": [
"cutoff = int(.75 * len(annotated_sent))\n",
"training_sentences = annotated_sent[:cutoff]\n",
"test_sentences = annotated_sent[cutoff:]\n",
" \n",
"print(len(training_sentences))\n",
"print(len(test_sentences))\n",
" \n",
"def transform_to_dataset(tagged_sentences):\n",
" X, y = [], []\n",
" for tagged in tagged_sentences:\n",
" for index in range(len(tagged)):\n",
" X.append(features(untag(tagged), index))\n",
" y.append(tagged[index][1])\n",
" \n",
" return X, y\n",
" \n",
"X, y = transform_to_dataset(training_sentences)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Implementing a classifier"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"training OK\n",
"Accuracy: 0.878515185602\n"
]
}
],
"source": [
"from sklearn.tree import DecisionTreeClassifier\n",
"from sklearn.feature_extraction import DictVectorizer\n",
"from sklearn.pipeline import Pipeline\n",
"\n",
"size=10000\n",
"\n",
"clf = Pipeline([\n",
" ('vectorizer', DictVectorizer(sparse=False)),\n",
" ('classifier', DecisionTreeClassifier(criterion='entropy'))\n",
"])\n",
"clf.fit(X[:size], y[:size])\n",
" \n",
"print('training OK')\n",
" \n",
"X_test, y_test = transform_to_dataset(test_sentences)\n",
" \n",
"print(\"Accuracy:\", clf.score(X_test, y_test))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Using the classifier"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"3.6.3\n",
"checking...\n",
"[('Hello', 'NN'), ('world', 'NN'), (',', ','), ('lets', 'NNS'), ('do', 'VB'), ('something', 'VBG'), ('awesome', 'NN'), ('today', 'NN'), ('!', 'NNP')]\n"
]
}
],
"source": [
"def pos_tag(sentence):\n",
" print('checking...')\n",
" tagged_sentence = []\n",
" tags = clf.predict([features(sentence, index) for index in range(len(sentence))])\n",
" return zip(sentence, tags)\n",
"\n",
"import platform\n",
"print(platform.python_version())\n",
"\n",
"print(list(pos_tag(word_tokenize('Hello world, lets do something awesome today!'))))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Rule-based POS taggers\n",
"1. DefaultTagger that simply tags everything with the same tag\n",
"2. RegexpTagger that applies tags according to a set of regular expressions\n",
"3. N-Gram (n-gram tagger is a generalization of a unigram tagger whose context is the current word together with the part-of-speech tags of the n-1 preceding token)\n",
" * UnigramTagger\n",
" * BigramTagger\n",
" * TrigramTagger"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [
{
"ename": "NameError",
"evalue": "name 'brown_tagged_sents' is not defined",
"output_type": "error",
"traceback": [
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[0;31mNameError\u001b[0m Traceback (most recent call last)",
"\u001b[0;32m<ipython-input-9-cac1441958dc>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m 7\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0mnltk\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mTrigramTagger\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0mtg\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m 8\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 9\u001b[0;31m \u001b[0msize\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbrown_tagged_sents\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0;36m0.9\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m 10\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m 11\u001b[0m patterns = [(r'.*ing$', 'VBG'), (r'.*ed$', 'VBD'), (r'.*es$', 'VBZ'), (r'.*ould$', 'MD'), (r'.*\\'s$', 'NN$'), \n",
"\u001b[0;31mNameError\u001b[0m: name 'brown_tagged_sents' is not defined"
]
}
],
"source": [
"#nltk.download('brown')\n",
"\n",
"from nltk.corpus import brown\n",
"from nltk import DefaultTagger as df\n",
"from nltk import UnigramTagger as ut\n",
"from nltk import BigramTagger as bt\n",
"from nltk import TrigramTagger as tg\n",
"\n",
"size = int(len(brown_tagged_sents) * 0.9)\n",
"\n",
"patterns = [(r'.*ing$', 'VBG'), (r'.*ed$', 'VBD'), (r'.*es$', 'VBZ'), (r'.*ould$', 'MD'), (r'.*\\'s$', 'NN$'), \n",
" (r'.*s$', 'NNS'), (r'^-?[0-9]+(.[0-9]+)?$', 'CD'), (r'.*', 'NN')]\n",
"\n",
"brown_tagged_sents = brown.tagged_sents(categories='news')\n",
"brown_sents = brown.sents(categories='news')\n",
"\n",
"train_sents = brown_tagged_sents[:size]\n",
"test_sents = brown_tagged_sents[size:]\n",
"\n",
"def_model = nltk.DefaultTagger('NN')\n",
"uni_model = nltk.UnigramTagger(train_sents)\n",
"bi_model = nltk.BigramTagger(train_sents)\n",
"tri_model = nltk.TrigramTagger(train_sents)\n",
"regexp_model = nltk.RegexpTagger(patterns)\n",
"\n",
"# performance of Default Tagger\n",
"print(def_model.evaluate(train_sents))\n",
"print(def_model.evaluate(test_sents))\n",
"print()\n",
"# performance of Unigram Tagger\n",
"print(uni_model.evaluate(train_sents))\n",
"print(uni_model.evaluate(test_sents))\n",
"print()\n",
"# performance of Bigram Tagger\n",
"print(bi_model.evaluate(train_sents))\n",
"print(bi_model.evaluate(test_sents))\n",
"print()\n",
"# performance of Trigram Tagger\n",
"print(tri_model.evaluate(train_sents))\n",
"print(tri_model.evaluate(test_sents))\n",
"print()\n",
"# performance of Regex Tagger\n",
"print(regexp_model.evaluate(train_sents))\n",
"print(regexp_model.evaluate(test_sents))\n",
"print()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Exercise 01\n",
"### In this lab you will learn how to train your own POS tagger classifier and test it against some pre-trained models\n",
"__Pleases implement your code and upload it to git using (jupyter notebook format)__\n",
"#### Classifiers\n",
"1. model1 = your POS tagger model (english)\n",
"2. model2 = pre-trained POS tagger model using NLTK (maxentropy english)\n",
"3. model3.x = rule-based classifiers (x = 1 to 5)\n",
"4. model4 = your POS tagger model (not english)\n",
"5. model5 = pre-trained POS tagger model using RDRPOSTagger 1 or TreeTagger 2 (not english)\n",
"\n",
"note: for model1 and model3 you can try different classifiers: Hidden Markov Model, Logistic Regression, Maximum Entropy Markov Models, Decision Trees, Naive Bayes, etc..__choose one!__\n",
"#### Corpora\n",
"1. X1 = nltk.corpus.treebank (english)\n",
"2. X2 = nltk.corpus.brown (english)\n",
"3. X3 = other language (not english)\n",
"note: data split for training/test = 0.8/0.2 (sequencial)\n",
"#### Task 1\n",
"* get results for english (plot a graph with all classifiers x results)\n",
" * performance 1.1 = model1 in X1\n",
" * performance 1.2 = model2 in X1\n",
" * performance 1.3.x = model3.x in X1\n",
" * performance 1.4 = model1 in X2\n",
" * performance 1.5 = model2 in X2\n",
" * performance 1.6.x = model3.x in X2\n",
"#### Task 2\n",
"* train your model with standard features (plot a graph with all classifiers x results)\n",
" * performance 2.1 = model4 in X3\n",
" * performance 2.2 = model5 in X3\n",
"### notes:\n",
"1. you can save your trained models using pickle (import pickle)\n",
"2. please upload your jupyter file to git\n",
"3. this script just gives a general idea, please organize and comment your code accordingly\n",
"4. you have to make sure the language you choose is supported for one of the classifiers suggested (see above) AND you are able to find a corpus in that language (example: Tiger Corpus for German). You can also search the Web in order to try to find a pre-trained classifier in your language. If that is not possible, just choose one existing. Please also make sure the language you have choosen does not overlap with other students.\n",
"5. If you are able to find an annotated corpus and format is CoNLL, you can easly read it using the following method in NLTK:\n",
"corp = nltk.corpus.ConllCorpusReader()\n",
"6. a nice library to create charts: https://plot.ly/python/bar-charts/"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.6.3"
}
},
"nbformat": 4,
"nbformat_minor": 2
}

View File

@ -2,8 +2,10 @@
"cells": [ "cells": [
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 18, "execution_count": 1,
"metadata": {}, "metadata": {
"collapsed": true
},
"outputs": [], "outputs": [],
"source": [ "source": [
"import nltk" "import nltk"
@ -18,15 +20,15 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 13, "execution_count": 2,
"metadata": {}, "metadata": {},
"outputs": [ "outputs": [
{ {
"name": "stdout", "name": "stdout",
"output_type": "stream", "output_type": "stream",
"text": [ "text": [
"[nltk_data] Downloading package punkt to /home/jonas/nltk_data...\n", "[nltk_data] Downloading package punkt to /Users/Carsten/nltk_data...\n",
"[nltk_data] Package punkt is already up-to-date!\n" "[nltk_data] Unzipping tokenizers/punkt.zip.\n"
] ]
}, },
{ {
@ -35,7 +37,7 @@
"True" "True"
] ]
}, },
"execution_count": 13, "execution_count": 2,
"metadata": {}, "metadata": {},
"output_type": "execute_result" "output_type": "execute_result"
} }
@ -46,7 +48,7 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 19, "execution_count": 3,
"metadata": {}, "metadata": {},
"outputs": [ "outputs": [
{ {
@ -54,9 +56,8 @@
"output_type": "stream", "output_type": "stream",
"text": [ "text": [
"[nltk_data] Downloading package averaged_perceptron_tagger to\n", "[nltk_data] Downloading package averaged_perceptron_tagger to\n",
"[nltk_data] /home/jonas/nltk_data...\n", "[nltk_data] /Users/Carsten/nltk_data...\n",
"[nltk_data] Package averaged_perceptron_tagger is already up-to-\n", "[nltk_data] Unzipping taggers/averaged_perceptron_tagger.zip.\n"
"[nltk_data] date!\n"
] ]
}, },
{ {
@ -65,7 +66,7 @@
"True" "True"
] ]
}, },
"execution_count": 19, "execution_count": 3,
"metadata": {}, "metadata": {},
"output_type": "execute_result" "output_type": "execute_result"
} }
@ -83,8 +84,10 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 6, "execution_count": 4,
"metadata": {}, "metadata": {
"collapsed": true
},
"outputs": [], "outputs": [],
"source": [ "source": [
"sentence = \"This is a test sentence.\"" "sentence = \"This is a test sentence.\""
@ -92,7 +95,7 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 16, "execution_count": 5,
"metadata": {}, "metadata": {},
"outputs": [ "outputs": [
{ {
@ -112,7 +115,7 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 17, "execution_count": 6,
"metadata": {}, "metadata": {},
"outputs": [ "outputs": [
{ {
@ -138,7 +141,9 @@
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": null, "execution_count": null,
"metadata": {}, "metadata": {
"collapsed": true
},
"outputs": [], "outputs": [],
"source": [] "source": []
} }