def main(input_filename, output_filename):
reviews = pyrecoutils.load_ratings(input_filename)
training_set, validation_set, test_set = pyrecoutils.split_sets(reviews)
model = UserBias()
model.fit(training_set)
model.select(validation_set)
with pyrecoutils.PredictionWriter(output_filename) as out:
training_prediction = [model.predict(u) for u, i, r in training_set]
out.dump(training_set, training_prediction)
training_rmse = pyrecoutils.rmse(training_set, training_prediction)
out.space_line()
validation_prediction = [
model.predict(u) for u, i, r in validation_set
]
out.dump(validation_set, validation_prediction)
validation_rmse = pyrecoutils.rmse(validation_set,
validation_prediction)
out.space_line()
test_prediction = [model.predict(u) for u, i, r in test_set]
out.dump(test_set, test_prediction)
test_rmse = pyrecoutils.rmse(test_set, test_prediction)
logging.info(u"RMSE: %.4f %.4f %.4f", training_rmse, validation_rmse,
test_rmse)
def main(input_filename, output_filename, components, epochs, learning_rate,
decay_rate, user_l2, item_l2, seed):
reviews = pyrecoutils.load_ratings(input_filename)
training_set, validation_set, test_set = pyrecoutils.split_sets(reviews)
model = MatrixFactorization(components, epochs, learning_rate, decay_rate,
user_l2, item_l2, seed)
model.fit(training_set, validation_set, test_set)
model.select(validation_set)
with pyrecoutils.PredictionWriter(output_filename) as out:
training_prediction = [model.predict(u, i) for u, i, r in training_set]
out.dump(training_set, training_prediction)
training_rmse = pyrecoutils.rmse(training_set, training_prediction)
out.space_line()
validation_prediction = [
model.predict(u, i) for u, i, r in validation_set
]
out.dump(validation_set, validation_prediction)
validation_rmse = pyrecoutils.rmse(validation_set,
validation_prediction)
out.space_line()
test_prediction = [model.predict(u, i) for u, i, r in test_set]
out.dump(test_set, test_prediction)
test_rmse = pyrecoutils.rmse(test_set, test_prediction)
logging.info(u"RMSE: %.4f %.4f %.4f", training_rmse, validation_rmse,
test_rmse)
def fit(self, training_reviews, validation_reviews, test_reviews):
self._initialize(training_reviews)
for epoch in range(self.epochs):
for u, i, r in training_reviews:
b_u, b_i = self.user_biases[u], self.item_biases[i]
p_u, q_i = self.user_latent[u], self.item_latent[i]
delta = self.overall + b_u + b_i + np.dot(p_u, q_i) - r
self.user_biases[u] -= self.learning_rate * (
delta + self.user_l2 * b_u)
self.user_latent[u] -= self.learning_rate * (
delta * q_i + self.user_l2 * p_u)
self.item_biases[i] -= self.learning_rate * (
delta + self.item_l2 * b_i)
self.item_latent[i] -= self.learning_rate * (
delta * p_u + self.item_l2 * q_i)
self.learning_rate *= self.decay_rate
training_rmse = pyrecoutils.rmse(
training_reviews,
[self.predict(u, i) for u, i, r in training_reviews])
validation_rmse = pyrecoutils.rmse(
validation_reviews,
[self.predict(u, i) for u, i, r in validation_reviews])
test_rmse = pyrecoutils.rmse(
test_reviews, [self.predict(u, i) for u, i, r in test_reviews])
logging.info(u"Epoch % 4d - RMSE %f %f %f", epoch, training_rmse,
validation_rmse, test_rmse)
def select(self, reviews):
logging.info(u"Selecting threshold on %d reviews", len(reviews))
rmses = []
for t in range(10):
self.threshold = t
predictions = [self.predict(u) for u, i, r in reviews]
rmses.append(pyrecoutils.rmse(reviews, predictions))
threshold = np.argmin(rmses)
logging.info(u"Selected threshold: %f", threshold)
self.threshold = threshold
def main(input_filename, output_filename):
reviews = pyrecoutils.load_ratings(input_filename)
training_set, validation_set, test_set = pyrecoutils.split_sets(reviews)
model = OverallBias()
model.fit(training_set)
with pyrecoutils.PredictionWriter(output_filename) as out:
out.dump(training_set, [model.bias] * len(training_set))
training_rmse = pyrecoutils.rmse(training_set,
[model.bias] * len(training_set))
out.space_line()
out.dump(validation_set, [model.bias] * len(validation_set))
validation_rmse = pyrecoutils.rmse(validation_set,
[model.bias] * len(validation_set))
out.space_line()
out.dump(test_set, [model.bias] * len(test_set))
test_rmse = pyrecoutils.rmse(test_set, [model.bias] * len(test_set))
logging.info(u"RMSE: %.4f %.4f %.4f", training_rmse, validation_rmse,
test_rmse)
def select(self, reviews):
logging.info(u"Selecting threshold on %d reviews", len(reviews))
rmses = []
for t_u in range(10):
self.user_threshold = t_u
for t_i in range(10):
self.item_threshold = t_i
predictions = [self.predict(u, i) for u, i, r in reviews]
rmses.append(pyrecoutils.rmse(reviews, predictions))
threshold = np.argmin(rmses)
user_threshold = threshold / 10
item_threshold = threshold % 10
logging.info(u"Selected user threshold: %f", user_threshold)
self.user_threshold = user_threshold
logging.info(u"Selected item threshold: %f", item_threshold)
self.item_threshold = item_threshold