def learn_lr_classifier(training_corpus):
D = get_vocabulary_size()
labels = get_labels()
w = [0] * (D + 1)
norm = 1.0
num_iters = 0
while norm > convergence_threshold:
num_iters += 1
if num_iters > max_iters:
break
old_w = list(w)
shuffled = list(training_corpus)
shuffle(shuffled)
for vector in shuffled:
label = 1.0 if float(vector[0]) == labels[0] else 0.0
prediction = get_prediction(vector[1:], w)
delta = label - prediction
update_weights(vector[1:], w, delta)
norm = get_norm(w,old_w)
return w
def test_logistic_regression(w):
if not generate_labels():
return
fp = FileProcessor(testing_data_filepath, ' ')
rows = fp.parse_input_file()
output = []
expected = []
labels = get_labels()
for row in rows:
expected.append(row[0])
row = row[1:]
sum_val = w[0]
for feature in row:
feature_id = int(feature.split(':')[0])
sum_val += w[feature_id]
if sigmoid(sum_val) >= 0.5:
output.append(labels[0])
else:
output.append(labels[1])
if fp.generate_output(output_filepath, output):
print 'Successfully generated predictions.lr'
