-
Notifications
You must be signed in to change notification settings - Fork 4
/
kaggle.py
218 lines (159 loc) · 6.2 KB
/
kaggle.py
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
import pandas as pd
import numpy as np
import re
import codecs
import time
import os
import itertools
from sklearn import cross_validation
from bs4 import BeautifulSoup
from contractions import contractions
from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score, recall_score, precision_score
from sklearn.naive_bayes import GaussianNB, MultinomialNB
from sklearn.svm import SVC
def grammarContractions(original_text):
text = original_text.lower().split()
reformed = [contractions[word] if word in contractions else word for word in text]
return " ".join(reformed)
def cleaning(original_text):
text = BeautifulSoup(original_text,"lxml").get_text()
# Remove Encodings
text = re.sub(r'\\\\', r'\\', text)
text = re.sub(r'\\x\w{2,2}',' ', text)
text = re.sub(r'\\u\w{4,4}', ' ', text)
text = re.sub(r'\\n', '.', text)
#Whitespace Formatting
text = text.replace('"', ' ')
text = text.replace('\\', ' ')
text = text.replace('_', ' ')
text = text.replace('-', ' ')
text = re.sub(' +',' ', text)
#Remove Unicode characters
text = codecs.decode(text, 'unicode-escape')
text = ''.join([i if ord(i) < 128 else '' for i in text])
#Remove email addresses
text = re.sub(r'[\w\-][\w\-\.]+@[\w\-][\w\-\.]+[a-zA-Z]{1,4}', ' ', text)
#Remove Twitter Usernames
text = re.sub(r"(\A|\s)@(\w+)+[a-zA-Z0-9_\.]", ' ', text)
#Remove urls
text = re.sub(r'\w+:\/\/\S+', ' ', text)
# Word Standardizing (Ex. Looooolll should be Looll)
text = ''.join(''.join(s)[:2] for _, s in itertools.groupby(text))
#Convert words to lower case
text = text.lower().split()
#Remove contractions by expansion of words
text = [contractions[word] if word in contractions else word for word in text]
#Rejoin words
text = " ".join(text)
#Remove non-alphabets
text = re.sub("[^a-z\s]", " ", text)
return " ".join(text.split())
#For Cross Validating and Evaluation of Model
def crossValidationTest(model, vectorizer, features, df):
cv = cross_validation.KFold(n=len(df), shuffle=False, n_folds=10)
features = np.asarray(features)
accuracy_array = []
precision_array = []
recall_array = []
for train_cv, test_cv in cv:
#Use the K-Fold for Training set and Testing set
X_train, X_test = features[train_cv], features[test_cv]
y_train, y_test = df.iloc[train_cv]['Insult'], df.iloc[test_cv]['Insult']
#Training set transformed and fitted
data_features_train = vectorizer.fit_transform(X_train)
data_features_train = data_features_train.toarray()
#Model trained
model.fit(data_features_train, y_train)
#Testing set transformed
data_features_test = vectorizer.transform(X_test)
data_features_test = data_features_test.toarray()
#Predict results
result = model.predict(data_features_test)
#Get accuracy
accuracy = accuracy_score(y_true=y_test, y_pred=result)
recall = recall_score(y_true=y_test, y_pred=result)
precision = precision_score(y_true=y_test, y_pred=result)
accuracy_array.append(accuracy)
recall_array.append(recall)
precision_array.append(precision)
print("AVERAGE ACCURACY: ", round(np.average(accuracy_array)*100, 2))
print("AVERAGE RECALL: ", round(np.average(recall_array)*100, 2))
print("AVERAGE PRECISION: ", round(np.average(precision_array)*100, 2))
#Read in the CSV file
df = pd.read_csv("Kaggle/second dataset.csv")
# One array to store clean comments, the other to store indexes of empty comments
clean_comments = []
empty_comments = []
#Drop Latin comments (Performed by manual Data Analysis)
df.drop(df.index[[4,798,3127,6183,6347]], inplace=True)
df.reset_index(inplace=True, drop=True)
#Split Training and Testing Set
#train_df = df.iloc[:3900]
for i in range(len(df)):
clean_comment = cleaning(df['Comment'][i])
# Add comment if it is not empty
if clean_comment != "":
clean_comments.append(clean_comment)
else:
empty_comments.append(i)
# Drop columns of empty comments
df.drop(df.index[empty_comments], inplace=True)
df.reset_index(inplace=True, drop=True)
# Create a new dataframe
clean_df = pd.DataFrame(data={'Insult':df['Insult'], 'Comment': clean_comments})
print("\n\n",len(clean_df))
# Save the cleaned dataset
clean_df.to_csv('second_clean_dataset.csv', index=False)
#Rest the index of the test set
#Implement Bag-Of-Words/N-grams/TF-IDF
#vectorizer = CountVectorizer(analyzer="word", ngram_range=(3, 3))
#vectorizer = TfidfVectorizer(stop_words='english')
#Implement Classifier here
#model = GaussianNB()
#model = RandomForestClassifier(n_estimators=100)
#model = LinearSVC()
#crossValidationTest(model, vectorizer, clean_comments_train, train_df)
#exit()
#data_features_train = vectorizer.fit_transform(clean_comments_train)
#data_features_train = data_features_train.toarray()
#model.fit(data_features_train, train_df['Insult'])
#Implement SVM Classifier
#model = SVC(kernel='linear')
#model.fit( data_features_train, train_df["Insult"] )
#test_df = df.iloc[3900:]
#test_df.reset_index(inplace=True, drop=True)
#clean_comments_test = []
#for i in range(len(test_df)):
# clean_comment = cleaning(test_df['Comment'][i])
# clean_comments_test.append(clean_comment)
#data_features_test = tf_idf.transform(clean_comments_test)
#data_features_test = vectorizer.transform(clean_comments_test)
#data_features_test = data_features_test.toarray()
#result = model.predict(data_features_test)
#test_labels = test_df.as_matrix(columns=['Insult'])
#accuracy = accuracy_score(y_true=test_df['Insult'], y_pred=result)
#recall = recall_score(y_true=test_df['Insult'], y_pred=result)
#precision = precision_score(y_true=test_df['Insult'], y_pred=result)
#print("TEST ACCURACY: ", round(accuracy*100, 2))
#print("TEST RECALL: ", round(recall*100, 2))
#print("TEST PRECISION: ", round(precision*100, 2))
# ACCURACY SCORE
# ==============
# BoW + RF = 84% (6000 to 589)
# BoW + NB = 66% (6000 to 589)
# BoW + SVM = 85% (6000 to 589)
# TF-IDF + RF = 82% (6000 to 589)
# TF-IDF + NB = 65% (6000 to 589)
# TF-IDF + SVM = 87% (6000 to 589)
# 2-grams + RF = 80% (6000 to 589)
# 2-grams + NB = 72% (6000 to 589)
# 2-grams + SVM = EMERGENCY!
# 3-grams + RF = EMERGENCY!
# 3-grams + NB = EMERGENCY!
# 3-grams + SVM = EMERGENCY
#NETLINGO for Arconyms
#Spellchecker using regex
#Google Distance
#Frequency Words