def writer(class_name, table, cls_to_accept):
class_writer = unicodecsv.writer(open(class_name, 'w'), encoding='utf-8')
class_writer.writerow(("tweet_id", "tweet_text"))
for cls, tweet_id, tweet_text in sql_convenience.extract_classifications_and_tweets(args.table):
if cls == cls_to_accept:
# remove carriage returns
tweet_text = tweet_text.replace("\n", " ")
class_writer.writerow((tweet_id, tweet_text))
def writer(class_name, table, cls_to_accept):
class_writer = unicodecsv.writer(open(class_name, 'w'), encoding='utf-8')
class_writer.writerow(("tweet_id", "tweet_text"))
for cls, tweet_id, tweet_text in sql_convenience.extract_classifications_and_tweets(
args.table):
if cls == cls_to_accept:
# remove carriage returns
tweet_text = tweet_text.replace("\n", " ")
class_writer.writerow((tweet_id, tweet_text))
def label_learned_set(vectorizer, clfl, threshold, validation_table):
for row in sql_convenience.extract_classifications_and_tweets(validation_table):
cls, tweet_id, tweet_text = row
spd = vectorizer.transform([tweet_text]).todense()
predicted_cls = clfl.predict(spd)
predicted_class = predicted_cls[0] # turn 1D array of 1 item into 1 item
predicted_proba = clfl.predict_proba(spd)[0][predicted_class]
if predicted_proba < threshold and predicted_class == 1:
predicted_class = 0 # force to out-of-class if we don't trust our answer
sql_convenience.update_class(tweet_id, validation_table, predicted_class)
def label_learned_set(vectorizer, clfl, threshold, validation_table):
for row in sql_convenience.extract_classifications_and_tweets(validation_table):
cls, tweet_id, tweet_text = row
spd = vectorizer.transform([tweet_text]).todense()
predicted_cls = clfl.predict(spd)
predicted_class = predicted_cls[0] # turn 1D array of 1 item into 1 item
predicted_proba = clfl.predict_proba(spd)[0][predicted_class]
if predicted_proba < threshold and predicted_class == 1:
predicted_class = 0 # force to out-of-class if we don't trust our answer
sql_convenience.update_class(tweet_id, validation_table, predicted_class)
import unicodecsv
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Tweet annotator')
parser.add_argument(
'keyword',
help=
'Keyword we wish to disambiguate (determines table name and used to filter tweets)'
)
parser.add_argument(
'--csv',
default=None,
help='CSV filename to write to (e.g. output.csv), defaults to stdout')
args = parser.parse_args()
if args.csv is None:
writer_stream = sys.stdout
else:
writer_stream = open(args.csv, "w")
writer = unicodecsv.writer(writer_stream, encoding='utf-8')
classifications_and_tweets = sql_convenience.extract_classifications_and_tweets(
args.keyword)
for cls, tweet_id, tweet in classifications_and_tweets:
writer.writerow((cls, tweet))
if not writer_stream.isatty():
# close the file (but not stdout if that's what we're using!)
writer_stream.close()
if __name__ == "__main__":
# gold_std table, comparison_table
parser = argparse.ArgumentParser(description='Score results against a gold standard')
parser.add_argument('gold_standard_table', help='Name of the gold standard table (e.g. annotations_apple)')
parser.add_argument('comparison_table', help='Name of the table we will score against the gold_standard_table (e.g. scikit_apple)')
args = parser.parse_args()
# counters for the 4 types of classification
tp = 0 # True Positives (predicted in class and are actually in class)
tn = 0 # True Negatives (predicted out of class and are actually out of class)
fp = 0 # False Positives (predicted in class but are actually out of class)
fn = 0 # False Negatives (predicted out of class but are actually in class)
# for each tweet in comparison table, get tweet_id and cls
classifications_and_tweets = sql_convenience.extract_classifications_and_tweets(args.gold_standard_table)
for gold_class, tweet_id, tweet in classifications_and_tweets:
cls, _, _ = sql_convenience.extract_classification_and_tweet(args.comparison_table, tweet_id)
if gold_class == sql_convenience.CLASS_IN:
if cls == sql_convenience.CLASS_IN:
tp += 1
else:
assert cls == sql_convenience.CLASS_OUT
fn += 1
else:
assert gold_class == sql_convenience.CLASS_OUT
if cls == sql_convenience.CLASS_OUT:
tn += 1
else:
assert cls == sql_convenience.CLASS_IN
fp += 1
"""Annotate tweets by hand to create a gold standard"""
from __future__ import division # 1/2 == 0.5, as in Py3
from __future__ import absolute_import # avoid hiding global modules with locals
from __future__ import print_function # force use of print("hello")
from __future__ import unicode_literals # force unadorned strings "" to be unicode without prepending u""
import argparse
import sys
import sql_convenience
import unicodecsv
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Tweet annotator')
parser.add_argument('keyword', help='Keyword we wish to disambiguate (determines table name and used to filter tweets)')
parser.add_argument('--csv', default=None, help='CSV filename to write to (e.g. output.csv), defaults to stdout')
args = parser.parse_args()
if args.csv is None:
writer_stream = sys.stdout
else:
writer_stream = open(args.csv, "w")
writer = unicodecsv.writer(writer_stream, encoding='utf-8')
classifications_and_tweets = sql_convenience.extract_classifications_and_tweets(args.keyword)
for cls, tweet_id, tweet in classifications_and_tweets:
writer.writerow((cls, tweet))
if not writer_stream.isatty():
# close the file (but not stdout if that's what we're using!)
writer_stream.close()
