def main():
# load labelled data
classes = POS | NEU | NEG
train_loc = root+'Data/twitterData/train_alternative.tsv'
dev_loc = root+'Data/twitterData/dev_alternative.tsv'
test_loc = root+'Data/twitterData/test_alternative.tsv'
train, dev, test = get_final_semeval_data(classes, train_loc, dev_loc, test_loc)
# load model
model = svm_pipeline()
# run main routine
for funrun in ('km', 'af', 'cl'):
run(model, train[0], train[1], dev[0], dev[1], mode=['filter'], retrain=5,
token=funrun)
def evaluate_mutator(mutator, threshold, min_percent, latex=True):
labels = (POS, NEU, NEG)
train_loc = root+'Data/twitterData/train_alternative.tsv'
dev_loc = root+'Data/twitterData/dev_alternative.tsv'
test_loc = root+'Data/twitterData/test_alternative.tsv'
train, dev, test = get_final_semeval_data(reduce(lambda x, y: x|y, labels), train_loc, dev_loc, test_loc)
dev_x, dev_y = dev
for label in labels:
pred_y = []
for tweet in dev_x:
pred_y.append(label if mutator.apply_filter(tweet, label) else -1)
if pred_y.count(label) < min_percent*len(pred_y):
yield str(label), ' (%.3f,0.0)' % threshold
else:
yield str(label), ' (%.3f,%.4f)' % (threshold, precision(dev_y, pred_y, label))
