def test_bpmf_with_random_data(self):
n_user = 1000
n_item = 2000
ratings = make_ratings(n_user,
n_item,
20,
30,
self.rating_choices,
seed=self.seed)
bpmf1 = BPMF(n_user,
n_item,
self.n_feature,
max_rating=self.max_rat,
min_rating=self.min_rat,
seed=self.seed)
bpmf1.fit(ratings, n_iters=1)
rmse_1 = RMSE(bpmf1.predict(ratings[:, :2]), ratings[:, 2])
bpmf2 = BPMF(n_user,
n_item,
self.n_feature,
max_rating=self.max_rat,
min_rating=self.min_rat,
seed=self.seed)
bpmf2.fit(ratings, n_iters=3)
rmse_2 = RMSE(bpmf2.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse_1 > rmse_2)
def test_bpmf_convergence(self):
n_user = 100
n_item = 200
n_feature = self.n_feature
ratings = make_ratings(n_user,
n_item,
20,
30,
self.rating_choices,
seed=self.seed)
bpmf1 = BPMF(n_user,
n_item,
n_feature,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat,
converge=1e-2)
bpmf1.fit(ratings, n_iters=5)
rmse_1 = RMSE(bpmf1.predict(ratings[:, :2]), ratings[:, 2])
bpmf2 = BPMF(n_user,
n_item,
n_feature,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat,
converge=1e-1)
bpmf2.fit(ratings, n_iters=5)
rmse_2 = RMSE(bpmf2.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse_1 < rmse_2)
def test_bpmf_convergence(self):
n_user = 100
n_item = 200
n_feature = self.n_feature
ratings = make_ratings(
n_user, n_item, 20, 30, self.rating_choices, seed=self.seed)
bpmf1 = BPMF(n_user, n_item, n_feature,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat,
converge=1e-3)
bpmf1.fit(ratings, n_iters=5)
rmse_1 = RMSE(bpmf1.predict(ratings[:, :2]), ratings[:, 2])
bpmf2 = BPMF(n_user, n_item, n_feature,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat,
converge=1e-2)
bpmf2.fit(ratings, n_iters=5)
rmse_2 = RMSE(bpmf2.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse_1 < rmse_2)
def test_bpmf_with_ml_100k_rating(self):
n_user = 943
n_item = 1682
n_feature = 10
ratings = self.ratings
bpmf = BPMF(n_user, n_item, n_feature,
max_rating=5.,
min_rating=1.,
seed=self.seed)
bpmf.fit(ratings, n_iters=30)
rmse = RMSE(bpmf.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse < 0.85)
def test_bpmf_with_ml_100k_rating(self):
n_user = 943
n_item = 1682
n_feature = 10
ratings = self.ratings
bpmf = BPMF(n_user,
n_item,
n_feature,
max_rating=5.,
min_rating=1.,
seed=self.seed)
bpmf.fit(ratings, n_iters=15)
rmse = RMSE(bpmf.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse < 0.85)
def test_bpmf_with_random_data(self):
n_user = 1000
n_item = 2000
ratings = make_ratings(
n_user, n_item, 20, 30, self.rating_choices, seed=self.seed)
bpmf1 = BPMF(n_user, n_item, self.n_feature,
max_rating=self.max_rat,
min_rating=self.min_rat,
seed=self.seed)
bpmf1.fit(ratings, n_iters=1)
rmse_1 = RMSE(bpmf1.predict(ratings[:, :2]), ratings[:, 2])
bpmf2 = BPMF(n_user, n_item, self.n_feature,
max_rating=self.max_rat,
min_rating=self.min_rat,
seed=self.seed)
bpmf2.fit(ratings, n_iters=3)
rmse_2 = RMSE(bpmf2.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse_1 > rmse_2)
# models settings; do now the loop over several n_features.
results = pd.DataFrame(
columns=['Number of features', 'Train RMSE', 'Test RMSE'])
n_features_list = [10, 20, 30, 40, 50, 60, 70, 80, 90, 100]
eval_iters = 50
for n_feature in n_features_list:
print(
"n_user: %d, n_item: %d, n_feature: %d, training size: %d, validation size: %d"
% (n_user, n_item, n_feature, train.shape[0], validation.shape[0]))
bpmf = BPMF(n_user=n_user,
n_item=n_item,
n_feature=n_feature,
max_rating=5.,
min_rating=1.,
seed=0)
train_rmse_list, test_rmse_list = bpmf.fit(train,
validation,
n_iters=eval_iters)
row = pd.DataFrame({
'Number of features': n_feature,
'Train RMSE': train_rmse_list,
'Test RMSE': test_rmse_list
})
results = results.append(row)
results.to_csv("results/1M_movielens_features{}_iterations{}.csv".format(
n_features_list, eval_iters))
ratings = load_movielens_1m_ratings(rating_file)
n_user = max(ratings[:, 0])
n_item = max(ratings[:, 1])
# shift user_id & movie_id by 1. let user_id & movie_id start from 0
ratings[:, (0, 1)] -= 1
# split data to training & testing
train_pct = 0.9
rand_state.shuffle(ratings)
train_size = int(train_pct * ratings.shape[0])
train = ratings[:train_size]
validation = ratings[train_size:]
# models settings
n_feature = 20
eval_iters = 50
print("n_user: %d, n_item: %d, n_feature: %d, training size: %d, validation size: %d" % (
n_user, n_item, n_feature, train.shape[0], validation.shape[0]))
bpmf = BPMF(n_user=n_user, n_item=n_item, n_feature=n_feature,
max_rating=5., min_rating=1., seed=0)
bpmf.fit(train, n_iters=eval_iters)
train_preds = bpmf.predict(train[:, :2])
train_rmse = RMSE(train_preds, train[:, 2])
val_preds = bpmf.predict(validation[:, :2])
val_rmse = RMSE(val_preds, validation[:, 2])
print("after %d iteration, train RMSE: %.6f, validation RMSE: %.6f" %
(eval_iters, train_rmse, val_rmse))