def prepare_data(data, clean_char=True):
'''
Preparing data for classification. SM Data will be a list of tuples with the format
(text, id, label, user, time), EF data is simply (text, label)
Shuffle, then extract paired lists of samples and labels
if clean_char is True, remove control characters and replace all emojis and emoticons with ' E'
'''
# Each tup is one sample paired with label/tag, no matter length
for tup in data:
# Get text, as list of word tokens
if clean_char:
sample = strip_emoticons(strip_cl_chars(tup[0]))
else:
sample = tup[0]
words = word_tokenize(sample)
# Get tag (label)
if len(tup) > 2:
tag = tup[2] # Twitter + Reddit data
else:
tag = tup[1] # EFcamdat data
# Always map words and tag to corresponding indices
# Format of input we need is:
# sample: [word_idx, word_idx, word_idx...]
# tag/lable: label_idx
yield ([ w2i[word] for word in words ], t2i[tag])
def shuffle_prepare_data(data, clean_char=True, num_labs=6):
'''
Preparing data for classification. Data will be a list of tuples with the format
(text, id, label, user, time)
Shuffle, then extract paired lists of samples and labels
if clean_char is True (default), remove control characters and replace all emojis and emoticons with '<emo> '
num_labs is number of labels / classes to use. Default is 6, i.e. unmapped from data. Alternatives: 3 and 4
'''
lab3_mapping = {
'a1': 'A',
'a2': 'A',
'b1': 'B',
'b2': 'B',
'c1': 'C',
'c2': 'C'
}
lab4_mapping = {
'a1': 'a',
'a2': 'a',
'b1': 'b',
'b2': 'b',
'c1': 'c1',
'c2': 'c2'
}
# Shuffling data
np.random.shuffle(data)
# # Discriminating solely between C1 and C2
# samples, labels = [],[]
# for tup in data:
# if tup[2] in {'c1','c2'}:
# samples.append(strip_emoticons(strip_cl_chars(tup[0])))
# labels.append(tup[2])
if clean_char:
samples = [strip_emoticons(strip_cl_chars(tup[0])) for tup in data]
else:
samples = [tup[0] for tup in data]
if num_labs == 6:
labels = [tup[2] for tup in data]
elif num_labs == 4:
labels = [lab4_mapping[tup[2]] for tup in data]
elif num_labs == 3:
labels = [lab3_mapping[tup[2]] for tup in data]
return samples, labels
def shuffle_prepare_data(data, clean_char=True):
'''
Preparing data for classification. Data will be a list of tuples with the format
(text, id, label, user, time)
Shuffle, then extract paired lists of samples and labels
if clean_char is True (default), remove control characters and replace all emojis and emoticons with '<emo> '
'''
np.random.shuffle(data)
if clean_char:
samples = [strip_emoticons(strip_cl_chars(tup[0])) for tup in data]
else:
samples = [tup[0] for tup in data]
# samples = [ tup[0] for tup in data]
labels = []
for tup in data:
if len(tup) > 2:
labels.append(tup[2]) # Twitter + Reddit data
else:
labels.append(tup[1]) # Efcamdat data
# labels = [ tup[2] for tup in data]
return samples, labels