Spanish-NLP/SVM.py at master · vansimonsen/Spanish-NLP · GitHub

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
#!/usr/bin/env python
# -*- coding: utf-8 -*-

import json
import unicodedata
import re
import pandas as pd
pd.set_option('max_colwidth',1000)
from lxml import objectify
import numpy as np


from string import punctuation
import nltk
from nltk.corpus import stopwords
from nltk import word_tokenize
from nltk.stem import SnowballStemmer

from string import punctuation

from sklearn.grid_search import GridSearchCV
from sklearn.cross_validation import cross_val_score
from sklearn.svm import LinearSVC
from sklearn.pipeline import Pipeline
from sklearn.feature_extraction.text import CountVectorizer


#nltk.download('punkt')
#nltk.download('stopwords')


def tr_data(training_path):
	#Data will be a CSV or XML file
	try:
		general_tweets_corpus_train = pd.read_csv(training_path, encoding='utf-8')
	except:
		xml = objectify.parse(open(training_path))
		#sample tweet object
		root = xml.getroot()
		general_tweets_corpus_train = pd.DataFrame(columns=('content', 'polarity', 'agreement'))
		tweets = root.getchildren()
		for i in range(0,len(tweets)):
		    tweet = tweets[i]
		    row = dict(zip(['content', 'polarity', 'agreement'],
		                   [tweet.content.text, tweet.sentiments.polarity.value.text,
		                    tweet.sentiments.polarity.type.text]))
		    row_s = pd.Series(row)
		    row_s.name = i
		    general_tweets_corpus_train = general_tweets_corpus_train.append(row_s)

	tweets_corpus = pd.concat([general_tweets_corpus_train])
	tweets_corpus = tweets_corpus.query('agreement != "DISAGREEMENT" and polarity != "NONE"')
	train_data = tweets_corpus[-tweets_corpus.content.str.contains('^http.*$')]
	return train_data

def ts_data(test_path):
	try:
		test_data = pd.read_csv(test_path, encoding='utf-8')
	except:
		xml = objectify.parse(open(test_path))
		#sample tweet object
		root = xml.getroot()
		test_data = pd.DataFrame(columns=('content', 'polarity'))
		tweets = root.getchildren()
		for i in range(0,len(tweets)):
		    tweet = tweets[i]
		    row = dict(zip(['content'], [tweet.content.text]))
		    row_s = pd.Series(row)
		    if row_s[0] != None:
		        row_s.name = i
		        test_data = test_data.append(row_s)


	test_data = pd.concat([test_data])


	return test_data


def tweets_classification(train_data, test_data, json=False, csv=True, result_path=None):
	#Stem: Cut word in root (wait: wait, waited: wait, waiting: wait)
	def stem_tokens(tokens, stemmer):
	    stemmed = []
	    for item in tokens:
	        stemmed.append(stemmer.stem(item))
	    return stemmed

	#Each word is a token
	def tokenize(text):
	    text = ''.join([c for c in text if c not in non_words])
	    tokens =  word_tokenize(text)

	    # stem
	    try:
	        stems = stem_tokens(tokens, stemmer)
	    except Exception as e:
	        print(e)
	        print(text)
	        stems = ['']
	    return stems

	#Stopwords: Empty word (i.e articles)

	spanish_stopwords = stopwords.words('spanish')
	stemmer = SnowballStemmer('spanish')


	#Non Words: Symbols and Numbers
	non_words = list(punctuation)
	non_words.extend(['¿', '¡'])
	non_words.extend(map(str,range(10)))


	#Binarizing

	train_data['polarity_bin'] = 0
	index = train_data.polarity.isin(['P', 'P+'])
	train_data.polarity_bin.loc[index] = 1


	### BEST PARAMS

	best_params = {'vect__ngram_range': (1, 2), 'cls__loss': 'hinge', 'vect__max_df': 0.5
	 , 'cls__max_iter': 1000, 'vect__min_df': 10, 'vect__max_features': 1000
	 , 'cls__C': 0.2}

	best_pipe = Pipeline([
	    ('vect', CountVectorizer(
	            analyzer = 'word',
	            tokenizer = tokenize,
	            lowercase = True,
	            stop_words = spanish_stopwords,
	            min_df = 10,
	            max_df = 0.5,
	            ngram_range=(1, 2),
	            max_features=1000
	            )),
	    ('cls', LinearSVC(C=.2, loss='hinge',max_iter=1000,multi_class='ovr',
	             random_state=None,
	             penalty='l2',
	             tol=0.0001
	             )),
	])

	best_pipe.fit(train_data.content, train_data.polarity_bin)


	test_data['polarity'] = best_pipe.predict(test_data.content)

	if csv:
		test_data.to_csv(result_path, encoding ='utf-8')
	elif json:
		test_data.to_json(result_path)


	return test_data
#tr, ts = load_data('TASS/csv/general-tweets-train-tagged.csv','Datasets/reformas/csv/RT.csv')

#result = tweets_classification(tr,ts, csv=False, json= True, result_path=0)