def score(y, predictions):
predicted_values = np.round(predictions)
accuracy = np.sum((y - predicted_values) == 0) / len(y)
# precision = np.sum(np.logical_and(y == 1, predicted_values == 1)) / np.sum(predicted_values == 1)
precision = np.sum(
(y + predicted_values) == 2) / np.sum(predicted_values == 1)
recall = np.sum((y + predicted_values) == 2) / np.sum(y == 1)
f_value = 2 * recall * precision / (recall + precision)
return accuracy, precision, recall, f_value
if __name__ == '__main__':
features_file = get_rel_path_from_working_directory(
__file__, '../data/features.txt')
source_file = get_rel_path_from_working_directory(__file__,
'../data/sentiment.txt')
theta_file = get_rel_path_from_working_directory(__file__,
'../data/theta.npy')
with open(features_file) as f:
features = f.read().split()
with open(source_file) as f:
sentences = f.readlines()
theta = np.load(theta_file)
X, Y = create_x_y_data(sentences, features)
predictions = sigmoid(X, theta)
accuracy, precision, recall, f_value = score(Y, predictions)
print('accuracy: %f \tprecision: %f \trecall: %f \tFvalue: %f' %
(accuracy, precision, recall, f_value))
size)]
removed_features = remove_no_appear_features(features,
training_sentences)
train_x, train_y = create_x_y_data(training_sentences,
removed_features)
theta = np.random.rand(len(removed_features) + 1)
for i in range(EPOCH):
print('epoch' + str(i), end='\r')
if i % 500 == 0 or i == EPOCH - 1:
obj, grad = object_and_grad(train_x, train_y, theta)
max_update_value = np.max(np.absolute(ETA * grad))
print(
'epoch%d: objective function value %f max update value %.5e'
% (i, obj, max_update_value))
theta = update_theta(ETA, train_x, train_y, theta)
test_x, test_y = create_x_y_data(testing_sentences, removed_features)
predictions = sigmoid(test_x, theta)
result.append(score(test_y, predictions))
print('tested by testing setntences.\n')
accuracy = sum((e[0] for e in result)) / N
precision = sum((e[1] for e in result)) / N
recall = sum((e[2] for e in result)) / N
f_value = sum((e[3] for e in result)) / N
print('accuracy: %f \tprecision: %f \trecall: %f \tFvalue: %f' %
(accuracy, precision, recall, f_value))
"""結果
accuracy: 0.750328 precision: 0.750446 recall: 0.750421 Fvalue: 0.750312
"""
from mymodule.path_helpers import get_rel_path_from_working_directory
import numpy as np
from stemming.porter2 import stem
from nlp100_73 import sigmoid
def create_x_data(sentences, features):
words = sentence.split()
stemmed_words = [stem(w) for w in words]
x_line = [1] + [1 if f in stemmed_words else 0 for f in features]
x = np.array(x_line)
return x
if __name__ == '__main__':
features_file = get_rel_path_from_working_directory(
__file__, '../data/features.txt')
theta_file = get_rel_path_from_working_directory(__file__,
'../data/theta.npy')
with open(features_file) as f:
features = f.read().split()
theta = np.load(theta_file)
while True:
sentence = input('sentence: ')
x = create_x_data(sentence, features)
h = sigmoid(x, theta)
if h > 0.5:
print('label: +1 (prediction: %f)' % h)
else:
print('label: -1 (prediction: %f)' % (1 - h))