def loaddata(data_dir, mode='train', max_len=None):
"""The function for loading data.
This function will load the data, and then turns it into
Lang Object.
Args:
data_dir: A string indicates the location of data set
mode: A string indicates to load train, valid, or test.
Returns:
A list of reading dataset and a dictionary of Langs
"""
data_set = readfile(data_dir + mode + '.json')
if max_len is not None:
data_set = data_set[:max_len]
rt, re, rm, summary = readLang(data_set)
print("Read %s data" % mode)
print("Read %s box score summary" % len(data_set))
print("Embedding size of (r.t, r.e, r.m) and summary:")
print("({}, {}, {}), {}".format(rt.n_words, re.n_words, rm.n_words,
summary.n_words))
langs = {'rt': rt, 're': re, 'rm': rm, 'summary': summary}
return data_set, langs
def runTest(file1, version, model, mode='words'):
"""Gets training or test file for stance detection SemiVal 2016 competition and prints prediction results.
Parameters
----------
file1 : list
a list with text tokens on index (0) and hashtags list on index (1)
istest : Boolean
specifies if the dataset is for test or training
version : int
0: Training dataset, 1: Test dataset, 2:Other domain dataset
mode : str
choose either (words) or (hashtags)
"""
indata = readfile(file1, version)
data = preprocesstweets(indata,
ignoreNONE=False,
version=version,
lowerCase=True)
tfidfAdded = getTfidfRepresentation(data, version, mode)
labels = [d[7] for d in data]
encoder = LabelEncoder()
y = encoder.fit_transform(labels)
print(encoder.classes_)
if version == 0:
x_train, x_test, y_train, y_test = train_test_split(tfidfAdded,
y,
test_size=0.2)
y_test = np_utils.to_categorical(y_test, num_classes=3)
y_train = np_utils.to_categorical(y_train, num_classes=3)
print(x_train.shape[1])
print(model.summary())
model.fit(x_train,
y_train,
epochs=10,
verbose=2,
validation_data=(x_test, y_test))
loss, acc = model.evaluate(x_test, y_test, verbose=0)
ypred = model.predict(x_test)
print('Training Accuracy: %f' % (acc * 100))
print('Training F-Score: ', f1(y_test, ypred) * 100)
if version == 1 or version == 2:
y = np_utils.to_categorical(y, num_classes=3)
loss, acc = model.evaluate(tfidfAdded, y)
ypred = model.predict(tfidfAdded)
otherdomain = ''
if version == 2:
otherdomain = '(other domain)'
print('TEST Accuracy ' + otherdomain + ': %f' % ((acc * 100)))
print('TEST F-Score ' + otherdomain + ': ', (f1(y, ypred) * 100))
def loaddata(data_dir, mode='train', max_len=None, copy_player=COPY_PLAYER):
data_set = readfile(data_dir + mode + '.json', copy_player=copy_player)
if max_len is not None:
data_set = data_set[:max_len]
rt, re, rm, summary = readLang(data_set)
print("Read %s data" % mode)
print("Read %s box score summary" % len(data_set))
print("Embedding size of (r.t, r.e, r.m) and summary:")
print("({}, {}, {}), {}".format(rt.n_words, re.n_words, rm.n_words,
summary.n_words))
langs = {'rt': rt, 're': re, 'rm': rm, 'summary': summary}
return data_set, langs
#External libraries
import numpy as np
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
from sklearn.model_selection import KFold
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import confusion_matrix
#files
from preprocessing import readfile, calc_length, flatten
from preprocessing import padding_data, one_hot_encoding
from preprocessing import CHANNELS
folder = "data"
train_data = readfile(f"{folder}/ae.train", 0)
test_data = readfile(f"{folder}/ae.test", 1)
maxlength = calc_length(test_data, train_data)
train_input, train_output = padding_data(train_data, maxlength, 0, True)
test_input, test_output = padding_data(test_data, maxlength, 1, True)
pca = PCA(n_components=CHANNELS)
train_transformed = pca.fit_transform(train_input)
test_transformed = pca.fit_transform(test_input)
"""
Logistic Regression with PCA
folds: 5, 10, 15
accuracy: 0.3162162162162162
"""
# cv = KFold(n_splits=10, random_state=42, shuffle=True)
# clf = LogisticRegression()
# scores = []
print(x_train.shape[1])
print(model.summary())
model.fit(x_train,
y_train,
epochs=10,
verbose=2,
validation_data=(x_test, y_test))
loss, acc = model.evaluate(x_test, y_test, verbose=0)
ypred = model.predict(x_test)
print('Training Accuracy: %f' % (acc * 100))
print('Training F-Score: ', f1(y_test, ypred) * 100)
if version == 1 or version == 2:
y = np_utils.to_categorical(y, num_classes=3)
loss, acc = model.evaluate(tfidfAdded, y)
ypred = model.predict(tfidfAdded)
otherdomain = ''
if version == 2:
otherdomain = '(other domain)'
print('TEST Accuracy ' + otherdomain + ': %f' % ((acc * 100)))
print('TEST F-Score ' + otherdomain + ': ', (f1(y, ypred) * 100))
indata = readfile('SemEval2016-Task6-subtaskA-traindata-gold.csv', 0)
data = preprocesstweets(indata, ignoreNONE=False, version=0, lowerCase=True)
tfidfAdded = getTfidfRepresentation(data, 0, 'words')
model = createModel(tfidfAdded.shape[1])
runTest('SemEval2016-Task6-subtaskA-traindata-gold.csv', 0, model, 'words')
runTest('SemEval2016-Task6-subtaskA-testdata-gold.txt', 1, model, 'words')
runTest('stance.csv', 2, model, 'words')
validation_data=(Xtest, y_test), verbose=2)
print('History', history.history)
# evaluate
print('Predicting (training)..')
ypred = model.predict(Xtest)
print('Accuracy (TRAIN): %f' % (model.evaluate(Xtest,y_test)[0]*100))
print('FScore (TRAIN): %f' % (f1(y_test, ypred)*100))
print('Predicting (testing)..')
#ypred = model.predict(XtestGroup)
#print('Accuracy (TEST): %f' % (model.evaluate(XtestGroup,y_testGroup)[0]*100))
#print('FScore (TEST): %f' % (f1(y_testGroup,ypred)*100))
indata = readfile('SemEval2016-Task6-subtaskA-traindata-gold.csv', False)
data = preprocesstweets(indata,ignoreNONE=False, version =0)
tweets = [' '.join(d[0]) for d in data]
stances = [d[7] for d in data]
encoder = LabelEncoder()
stances = encoder.fit_transform(stances)
indata = readfile('SemEval2016-Task6-subtaskA-testdata-gold.txt', True)
data = preprocesstweets(indata,ignoreNONE=False, version =1)
tweets2 = [' '.join(d[0]) for d in data]
stances2 = [d[7] for d in data]
stances2 = encoder.fit_transform(stances2)
convModel(tweets, stances, tweets2, stances2)
#The model is not working currently, last edits caused a problem. Reported results were from previous stage