def test_als_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)
als1 = ALS(n_user,
n_item,
n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat,
converge=1e-2)
als1.fit(ratings, n_iters=10)
rmse_1 = RMSE(als1.predict(ratings[:, :2]), ratings[:, 2])
als2 = ALS(n_user,
n_item,
n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat,
converge=1e-1)
als2.fit(ratings, n_iters=10)
rmse_2 = RMSE(als2.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse_1 < rmse_2)
def test_als_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)
als1 = ALS(n_user, n_item, n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat,
converge=1e-2)
als1.fit(ratings, n_iters=10)
rmse_1 = RMSE(als1.predict(ratings[:, :2]), ratings[:, 2])
als2 = ALS(n_user, n_item, n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat,
converge=1e-1)
als2.fit(ratings, n_iters=10)
rmse_2 = RMSE(als2.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse_1 < rmse_2)
def test_als_with_random_data(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)
als1 = ALS(n_user,
n_item,
n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat)
als1.fit(ratings, n_iters=1)
rmse_1 = RMSE(als1.predict(ratings[:, :2]), ratings[:, 2])
als2 = ALS(n_user,
n_item,
n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat)
als2.fit(ratings, n_iters=3)
rmse_2 = RMSE(als2.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse_1 > rmse_2)
def test_als_with_missing_data(self):
n_user = 10
n_item = 20
n_feature = self.n_feature
ratings = make_ratings(n_user - 1,
n_item - 1,
5,
10,
self.rating_choices,
seed=self.seed)
als1 = ALS(n_user,
n_item,
n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat)
unuse_user_f_before = als1.user_features_[n_user - 1, :]
unuse_item_f_before = als1.item_features_[n_item - 1, :]
als1.fit(ratings, n_iters=1)
unuse_user_f_after = als1.user_features_[n_user - 1, :]
unuse_item_f_after = als1.item_features_[n_item - 1, :]
# last user/item feature should be
# unchanged since no rating data on them
assert_array_equal(unuse_user_f_before, unuse_user_f_after)
assert_array_equal(unuse_item_f_before, unuse_item_f_after)
def test_als_with_random_data(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)
als1 = ALS(n_user, n_item, n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat)
als1.fit(ratings, n_iters=1)
rmse_1 = RMSE(als1.predict(ratings[:, :2]), ratings[:, 2])
als2 = ALS(n_user, n_item, n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat)
als2.fit(ratings, n_iters=3)
rmse_2 = RMSE(als2.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse_1 > rmse_2)
def test_als_with_ml_100k_rating(self):
n_user = 943
n_item = 1682
n_feature = 10
ratings = self.ratings
als = ALS(n_user, n_item, n_feature,
reg=1e-2,
max_rating=5.,
min_rating=1.,
seed=self.seed)
als.fit(ratings, n_iters=5)
rmse = RMSE(als.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse < 0.8)
def test_als_with_ml_100k_rating(self):
n_user = 943
n_item = 1682
n_feature = 10
ratings = self.ratings
als = ALS(n_user,
n_item,
n_feature,
reg=1e-2,
max_rating=5.,
min_rating=1.,
seed=self.seed)
als.fit(ratings, n_iters=5)
rmse = RMSE(als.predict(ratings[:, :2]), ratings[:, 2])
self.assertTrue(rmse < 0.8)
def test_als_with_missing_data(self):
n_user = 10
n_item = 20
n_feature = self.n_feature
ratings = make_ratings(
n_user - 1, n_item - 1, 5, 10, self.rating_choices, seed=self.seed)
als1 = ALS(n_user, n_item, n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat)
unuse_user_f_before = als1.user_features_[n_user - 1, :]
unuse_item_f_before = als1.item_features_[n_item - 1, :]
als1.fit(ratings, n_iters=1)
unuse_user_f_after = als1.user_features_[n_user - 1, :]
unuse_item_f_after = als1.item_features_[n_item - 1, :]
# last user/item feature should be
# unchanged since no rating data on them
assert_array_equal(unuse_user_f_before, unuse_user_f_after)
assert_array_equal(unuse_item_f_before, unuse_item_f_after)
def test_als_seed(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)
# seed 0
als1 = ALS(n_user, n_item, n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat)
als1.fit(ratings, n_iters=3)
als2 = ALS(n_user, n_item, n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat)
als2.fit(ratings, n_iters=3)
assert_array_equal(als1.user_features_, als2.user_features_)
assert_array_equal(als1.item_features_, als2.item_features_)
# seed 1
als3 = ALS(n_user, n_item, n_feature,
reg=1e-2,
seed=1,
max_rating=self.max_rat,
min_rating=self.min_rat)
als3.fit(ratings, n_iters=3)
assert_raises(AssertionError, assert_array_equal,
als1.user_features_, als3.user_features_)
assert_raises(AssertionError, assert_array_equal,
als1.item_features_, als3.item_features_)
def test_als_seed(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)
# seed 0
als1 = ALS(n_user,
n_item,
n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat)
als1.fit(ratings, n_iters=3)
als2 = ALS(n_user,
n_item,
n_feature,
reg=1e-2,
seed=0,
max_rating=self.max_rat,
min_rating=self.min_rat)
als2.fit(ratings, n_iters=3)
assert_array_equal(als1.user_features_, als2.user_features_)
assert_array_equal(als1.item_features_, als2.item_features_)
# seed 1
als3 = ALS(n_user,
n_item,
n_feature,
reg=1e-2,
seed=1,
max_rating=self.max_rat,
min_rating=self.min_rat)
als3.fit(ratings, n_iters=3)
assert_raises(AssertionError, assert_array_equal, als1.user_features_,
als3.user_features_)
assert_raises(AssertionError, assert_array_equal, als1.item_features_,
als3.item_features_)
validation = ratings[train_size:]
print(train.shape, validation.shape)
print(train[0, :], train[1, :])
# plt.imshow(train, cmap='jet', interpolation='nearest')
# plt.show()
# plt.imshow(validation, cmap='jet', interpolation='nearest')
# plt.show()
# models settings
n_feature = 10
eval_iters = 10
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]))
als = ALS(n_user=n_user,
n_item=n_item,
n_feature=n_feature,
reg=5e-2,
max_rating=5.,
min_rating=1.,
seed=0)
als.fit(train, n_iters=eval_iters)
train_preds = als.predict(train[:, :2])
train_rmse = RMSE(train_preds, train[:, 2])
val_preds = als.predict(validation[:, :2])
val_rmse = RMSE(val_preds, validation[:, 2])
print("after %d iterations, train RMSE: %.6f, validation RMSE: %.6f" % \
(eval_iters, train_rmse, val_rmse))
rating_file = ml_1m_download(ML_1M_FOLDER, file_size=ML_1M_ZIP_SIZE)
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 = 10
eval_iters = 10
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]))
als = ALS(n_user=n_user, n_item=n_item, n_feature=n_feature,
reg=5e-2, max_rating=5., min_rating=1., seed=0)
als.fit(train, n_iters=eval_iters)
train_preds = als.predict(train[:, :2])
train_rmse = RMSE(train_preds, train[:, 2])
val_preds = als.predict(validation[:, :2])
val_rmse = RMSE(val_preds, validation[:, 2])
print("after %d iterations, train RMSE: %.6f, validation RMSE: %.6f" % \
(eval_iters, train_rmse, val_rmse))