def process_tweets(classifier_name='svc',
train_data_src='tweets_and_news',
keyword_classifier_name='svc'):
## News Data
news_data = datasets.get_news_data(
folder_name='keyword_data', file_name='annotator_data_dump_with_text')
news_fields = ['text', 'category']
## Tweets Data
training_tweet_data = datasets.get_tweet_data(file_type='txt',
file_name='tweet_truth.txt')
unlabeled_tweet_data = datasets.get_tweet_data(
file_type='json', file_name='intermed_dict.json')
tweet_fields = ['tweet_cmplt', 'class_annotated']
X_test = unlabeled_tweet_data[tweet_fields[0]]
if classifier_name == 'news':
train_data = news_data
X_train, Y_train = train_data[news_fields[0]], train_data[
news_fields[1]]
if classifier_name == 'tweets':
train_data = training_tweet_data
X_train, Y_train = train_data[tweet_fields[0]], train_data[
tweet_fields[1]]
## Classify Below
if train_data_src == 'tweets_and_news':
X_train_news, Y_train_news = news_data[news_fields[0]], news_data[
news_fields[1]]
X_train_tweets, Y_train_tweets = training_tweet_data[
tweet_fields[0]], training_tweet_data[tweet_fields[1]]
X_train = X_train_tweets.append(X_train_news)
Y_train = Y_train_tweets.append(Y_train_news)
train_acc = []
if classifier_name in ['svc', 'lr', 'ada_boost']:
Y_predicted, curr_train_acc = classifier.classify(
classifier_name, X_train, Y_train, X_test)
train_acc.append(curr_train_acc)
return Y_predicted
elif classifier_name == 'cosineSim':
keywords = keyword_generator.keyword_driver(keyword_classifier_name,
X_train,
Y_train,
num_of_keywords=50)
Y_predicted_pos, Y_predicted_both = classifier.classify(
classifier_name, X_test, keywords=keywords)
return Y_predicted_both, Y_predicted_pos
if classifier_name == 'svc':
print('SVC train Acc : ', mean(train_acc))
def validation_model():
## Classifier Name
classifier_name = 'svc'
keyword_classifier_name = 'svc'
train_data_src = 'tweets_and_news'
num_splits = 5
## News Data
news_data = datasets.get_news_data(
folder_name='keyword_data', file_name='annotator_data_dump_with_text')
## Tweet Data
tweet_data = datasets.get_tweet_data(file_type='txt',
file_name='tweet_truth.txt')
if train_data_src == 'news':
data = news_data
field_names = ['text', 'category']
elif train_data_src == 'tweets':
data = tweet_data
field_names = ['tweet_cmplt', 'class_annotated']
elif train_data_src == 'tweets_and_news':
data = tweet_data
data_extra = news_data
field_names = ['tweet_cmplt', 'class_annotated']
field_names_extra = ['text', 'category']
kf = KFold(n_splits=num_splits)
kf.get_n_splits(data)
train_acc = []
test_acc = []
pos_f_measure = []
both_f_measure = []
pos_acc_list = []
both_acc_list = []
for train_index, test_index in kf.split(data):
X_train = data[field_names[0]].iloc[train_index]
Y_train = data[field_names[1]].iloc[train_index]
X_test = data[field_names[0]].iloc[test_index]
Y_test = data[field_names[1]].iloc[test_index]
if train_data_src == 'tweets_and_news':
X_extra = data_extra[field_names_extra[0]]
Y_extra = data_extra[field_names_extra[1]]
X_train = X_train.append(X_extra)
Y_train = Y_train.append(Y_extra)
if classifier_name in ['svc', 'lr', 'ada_boost']:
Y_predicted, curr_train_acc, curr_test_acc = classifier.classify(
classifier_name, X_train, Y_train, X_test, Y_test)
train_acc.append(curr_train_acc)
test_acc.append(curr_test_acc)
elif classifier_name == 'cosineSim':
keywords = keyword_generator.keyword_driver(
keyword_classifier_name, X_train, Y_train, num_of_keywords=50)
Y_predicted_pos, Y_predicted_both = classifier.classify(
classifier_name, X_test, keywords=keywords)
Y_test_list = []
Y_pred_both_list = []
Y_pred_pos_list = []
for i in Y_test.keys():
Y_test_list.append(Y_test.get_value(i))
Y_pred_pos_list.append(Y_predicted_pos[i])
Y_pred_both_list.append(Y_predicted_both[i])
Y_test_binary = utility.binarize_data(Y_test_list)
Y_pred_pos_binary = utility.binarize_data(Y_pred_pos_list)
Y_pred_both_binary = utility.binarize_data(Y_pred_both_list)
both_acc_list.append(
ca.calculate_accuracy(Y_predicted_both, Y_test))
both_f_measure.append(
cf.calculate_f_measure(Y_test_binary[1],
Y_pred_both_binary[1]))
pos_acc_list.append(ca.calculate_accuracy(Y_predicted_pos, Y_test))
pos_f_measure.append(
cf.calculate_f_measure(Y_test_binary[1], Y_pred_pos_binary[1]))
if classifier_name == 'svc':
print('SVC train Acc : ', mean(train_acc))
print('SVC test Acc : ', mean(test_acc))
elif classifier_name == 'cosineSim':
print('cosineSim POS Acc : ', mean(pos_acc_list))
print('cosineSim BOTH Acc : ', mean(both_acc_list))
print('cosineSim POS F : ', mean(pos_f_measure))
print('cosineSim BOTH F : ', mean(both_f_measure))
print('############ CLASSIFICATION COMPLETE ##################')
return classified_tweets
#################### Main ############################
if __name__=='__main__':
from lrf.utilities import datasets
file_type = 'txt'
file_name = 'tweet'
raw_tweets = datasets.get_tweet_data(file_type='txt', file_name='tweet_truth.txt')
raw_news = datasets.get_news_data(folder_name='keyword_data',file_name='annotator_data_dump_with_text')
model_dump_path = '../../classifier_data_n_model/'
output_path = '../../classifier_data_n_model/'
## Training the Model
# train_model(raw_tweets,raw_news,model_dump_path,classifier_type='unsupervised')
## Classify new tweets data
classified_tweets = classify_tweets(raw_tweets['tweet_cmplt'],model_dump_path, output_path,classifier_type='unsupervised')
## Dumping the results
from sklearn.externals import joblib
result = joblib.load(os.path.join(output_path,'classified_tweets.pkl'))
import numpy as np
from lrf.keyword_generator import keyword_generator
from lrf.configs import lrf_config
import os
import json
locations = lrf_config.get_locations()
risk_cat_file = os.path.join(locations['INTERMED_DATA_PATH'],
'risk_category_file.json')
news_data = datasets.get_news_data('keyword_data',
'annotator_data_dump_with_text')
train_data_news, test_data_news = utility.split_data(news_data)
field_names_news = ['text', 'category']
tweet_data = datasets.get_tweet_data('txt', 'tweet_truth.txt')
train_data_tweets, test_data_tweets = utility.split_data(tweet_data)
field_names_tweets = ['tweet_cmplt', 'class_annotated']
X_train_data = np.append(train_data_news[field_names_news[0]].values,
train_data_tweets[field_names_tweets[0]].values)
Y_train_data = np.append(train_data_news[field_names_news[1]].values,
train_data_tweets[field_names_tweets[1]].values)
category_keywords = keyword_generator.keyword_driver('svc',
X_train_data,
Y_train_data,
num_of_keywords=50)
for elem in category_keywords:
show_wordcloud(category_keywords[elem]['pos'])
def main():
news_dict = datasets.get_news_data(
folder_name='keyword_data', file_name='annotator_data_dump_with_text')
category_names = ['tweet_cmplt', 'class_annotated']
category_names_news = ['text', 'category']
twitter_dict = datasets.get_tweet_data('txt', 'tweet_truth.txt')
kf = KFold(n_splits=5)
kf.get_n_splits(twitter_dict)
some_dict = {}
train_acc = []
test_acc = []
acc_both = []
f_both = []
acc_pos = []
f_pos = []
ada_test_list = []
ada_train_list = []
news_train = news_dict['text']
news_class = news_dict['category']
for train_index, test_index in kf.split(twitter_dict):
print('%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%')
X_train = twitter_dict['tweet_cmplt'].iloc[train_index]
Y_train = twitter_dict['class_annotated'].iloc[train_index]
X_test = twitter_dict['tweet_cmplt'].iloc[test_index]
Y_test = twitter_dict['class_annotated'].iloc[test_index]
some_dict['tweet_cmplt'] = X_train.append(news_train)
some_dict['class_annotated'] = Y_train.append(news_class)
ada_predicted, ada_train_acc, ada_test_acc = classify(
'ada_boost', some_dict['tweet_cmplt'],
some_dict['class_annotated'], X_test, Y_test)
ada_train_list.append(ada_train_acc)
ada_test_list.append(ada_test_acc)
exit(0)
print('ada_train_list : ', ada_train_list)
print('ada_test_list : ', ada_test_list)
keywords = keyword_generator.keyword_driver(
'svc',
some_dict['tweet_cmplt'],
some_dict['class_annotated'],
num_of_keywords=50)
for item in keywords:
print(item, ' : ', keywords[item])
predicted, curr_train_acc, curr_test_acc = classify(
'svc', some_dict['tweet_cmplt'], some_dict['class_annotated'],
X_test, Y_test)
train_acc.append(curr_train_acc)
test_acc.append(curr_test_acc)
print('train_acc SVC: ', train_acc)
print('test_acc SVC: ', test_acc)
# Y_pred_pos, Y_pred_both = classify('cosineSim', X_test ,keywords = keywords)
Y_test_list = []
Y_pred_both_list = []
Y_pred_pos_list = []
for i in Y_test.keys():
Y_test_list.append(Y_test.get_value(i))
Y_pred_pos_list.append(Y_pred_pos[i])
Y_pred_both_list.append(Y_pred_both[i])
Y_test_binary = utility.binarize_data(Y_test_list)
Y_pred_pos_binary = utility.binarize_data(Y_pred_pos_list)
Y_pred_both_binary = utility.binarize_data(Y_pred_both_list)
acc_both.append(ca.calculate_accuracy(Y_pred_both, Y_test))
f_both.append(
cf.calculate_f_measure(Y_test_binary[1], Y_pred_both_binary[1]))
acc_pos.append(ca.calculate_accuracy(Y_pred_pos, Y_test))
f_pos.append(
cf.calculate_f_measure(Y_test_binary[1], Y_pred_pos_binary[1]))
print('################################ BOTH')
print('acc_both : ', mean(acc_both))
print('f_both : ', mean(f_both))
print('################################ POS')
print('acc_pos : ', mean(acc_pos))
print('f_pos : ', mean(f_pos))
print('############################### SVC')
print('Train_Accuracy : ', mean(train_acc))
print('Test_Accuracy : ', mean(test_acc))
print('############################### ADA_Boost')
print('Train_Accuracy : ', mean(ada_train_list))
print('Test_Accuracy : ', mean(ada_test_list))
exit(0)
# TWEET DATA
twitter_dict = datasets.get_tweet_data('txt', 'tweet_truth.txt')
train_data, test_data = utility.split_data(twitter_dict)
category_names = ['tweet_cmplt', 'class_annotated']
#category_names_tweet = ['tweet_word_list', 'class_annotated']
predicted_data, train_acc, test_acc = classify(
'lr', train_data[category_names[0]], train_data[category_names[1]],
test_data[category_names[0]], test_data[category_names[1]])
#predicted_data, train_acc, test_acc = classify('svc', news_dict[category_names_news[0]], news_dict[category_names_news[1]],
# twitter_dict[category_names_tweet[0]], twitter_dict[category_names_tweet[1]])
# print(predicted_data)
print('train_acc : ', train_acc)
print('test_acc : ', test_acc)