def test_perform(self):
sim = CosineSimilarity()
a = np.array([5, 9, 7, 8, 3, 5, 4, 2, 6, 4])
b = np.array([8, 1, 3, 10, 8, 4, 9, 2, 1, 6])
self.assertAlmostEqual(sim.perform(a, b), 0.7552110293516224)
a = np.array([0, 0, 0])
b = np.array([1, 1, 1])
self.assertEqual(sim.perform(a, b), 0)
a = np.array([1, 1, 1])
b = np.array([0, 0, 0])
self.assertEqual(sim.perform(a, b), 0)
a = np.array([0, 0, 0])
b = np.array([0, 0, 0])
self.assertEqual(sim.perform(a, b), 0)
a = np.array([1, 1, 1])
b = np.array([1, 1, 1])
self.assertEqual(sim.perform(a, b), 1)
a = np.array([1, 1, 1])
b = np.array([-1, -1, -1])
self.assertEqual(sim.perform(a, b), -1)
def test_empty_frame(self):
ratings_only_positive = pd.DataFrame.from_records(
[("A000", "tt0114576", 5, "54654675")],
columns=["from_id", "to_id", "score", "timestamp"])
ratings_only_negative = pd.DataFrame.from_records(
[("A000", "tt0114576", 1, "54654675")],
columns=["from_id", "to_id", "score", "timestamp"])
ratings_item_inexistent = pd.DataFrame.from_records(
[("A000", "not exists", 1, "54654675")],
columns=["from_id", "to_id", "score", "timestamp"])
# ClassifierRecommender returns an empty frame
alg = ClassifierRecommender({'Plot': ['tfidf', 'embedding']},
SkSVC(),
threshold=3)
rs = ContentBasedRS(alg, ratings_only_positive, self.movies_multiple)
result = rs.fit_rank('A000')
self.assertTrue(result.empty)
alg = ClassifierRecommender({'Plot': ['tfidf', 'embedding']},
SkSVC(),
threshold=3)
rs = ContentBasedRS(alg, ratings_only_negative, self.movies_multiple)
result = rs.fit_rank('A000')
self.assertTrue(result.empty)
alg = ClassifierRecommender({'Plot': ['tfidf', 'embedding']},
SkSVC(),
threshold=3)
rs = ContentBasedRS(alg, ratings_item_inexistent, self.movies_multiple)
result = rs.fit_rank('A000')
self.assertTrue(result.empty)
# CentroidVector returns an empty frame
alg = CentroidVector({'Plot': ['tfidf', 'embedding']},
CosineSimilarity(),
threshold=3)
rs = ContentBasedRS(alg, ratings_only_negative, self.movies_multiple)
result = rs.fit_rank('A000')
self.assertTrue(result.empty)
alg = CentroidVector({'Plot': ['tfidf', 'embedding']},
CosineSimilarity(),
threshold=3)
rs = ContentBasedRS(alg, ratings_item_inexistent, self.movies_multiple)
result = rs.fit_rank('A000')
self.assertTrue(result.empty)
def test_rank_multiple_representations(self):
# Multiple representations with auto threshold based on the mean ratings of the user
alg = CentroidVector({'Plot': ['tfidf', 'embedding'],
"Genre": ['tfidf', 'embedding'],
'imdbRating': [0]}, CosineSimilarity())
user_ratings = self.ratings.query('from_id == "A000"')
alg.process_rated(user_ratings, self.movies_dir)
alg.fit()
# rank with filter_list
res_filtered = alg.rank(user_ratings, self.movies_dir, filter_list=self.filter_list)
item_ranked_set = set(res_filtered['to_id'])
self.assertEqual(len(item_ranked_set), len(self.filter_list))
self.assertCountEqual(item_ranked_set, self.filter_list)
# rank without filter_list
res_all_unrated = alg.rank(user_ratings, self.movies_dir)
item_rated_set = set(user_ratings['to_id'])
item_ranked_set = set(res_all_unrated['to_id'])
# We expect this to be empty, since the alg should rank only unrated items (unless in filter list)
rated_in_ranked = item_ranked_set.intersection(item_rated_set)
self.assertEqual(len(rated_in_ranked), 0)
# rank with n_recs specified
n_recs = 5
res_n_recs = alg.rank(user_ratings, self.movies_dir, n_recs)
self.assertEqual(len(res_n_recs), n_recs)
item_rated_set = set(user_ratings['to_id'])
item_ranked_set = set(res_n_recs['to_id'])
# We expect this to be empty, since the alg should rank only unrated items (unless in filter list)
rated_in_ranked = item_ranked_set.intersection(item_rated_set)
self.assertEqual(len(rated_in_ranked), 0)
def test_rank_single_representation(self):
# Single representation
alg = CentroidVector({'Genre': ['embedding']}, CosineSimilarity(), threshold=0)
user_ratings = self.ratings.query('from_id == "A000"')
alg.process_rated(user_ratings, self.movies_dir)
alg.fit()
# rank with filter_list
res_filtered = alg.rank(user_ratings, self.movies_dir, filter_list=self.filter_list)
item_ranked_set = set(res_filtered['to_id'])
self.assertEqual(len(item_ranked_set), len(self.filter_list))
self.assertCountEqual(item_ranked_set, self.filter_list)
# rank without filter_list
res_all_unrated = alg.rank(user_ratings, self.movies_dir)
item_rated_set = set(user_ratings['to_id'])
item_ranked_set = set(res_all_unrated['to_id'])
# We expect this to be empty, since the alg should rank only unrated items (unless in filter list)
rated_in_ranked = item_ranked_set.intersection(item_rated_set)
self.assertEqual(len(rated_in_ranked), 0)
# rank with n_recs specified
n_recs = 5
res_n_recs = alg.rank(user_ratings, self.movies_dir, n_recs)
self.assertEqual(len(res_n_recs), n_recs)
item_rated_set = set(user_ratings['to_id'])
item_ranked_set = set(res_n_recs['to_id'])
# We expect this to be empty, since the alg should rank only unrated items (unless in filter list)
rated_in_ranked = item_ranked_set.intersection(item_rated_set)
self.assertEqual(len(rated_in_ranked), 0)
def test_predict(self):
alg = CentroidVector({'Genre': ['embedding']}, CosineSimilarity(), threshold=0)
user_ratings = self.ratings.query('from_id == "A000"')
alg.process_rated(user_ratings, self.movies_dir)
alg.fit()
# Will raise Exception since it's not a Score Prediction Algorithm
with self.assertRaises(NotPredictionAlg):
alg.predict(user_ratings, self.movies_dir)
def test_fit_cb_w_testrating_methodology(self):
rs = ContentBasedRS(
CentroidVector(
{"Plot": "tfidf"},
CosineSimilarity(),
), ratings, items_dir)
em = EvalModel(rs, KFoldPartitioning(), metric_list=[Precision()])
sys_result, users_result = em.fit()
self.assertIsInstance(sys_result, pd.DataFrame)
self.assertIsInstance(users_result, pd.DataFrame)
def test_raise_errors(self):
# Only negative available
self.ratings = pd.DataFrame.from_records([
("A000", "tt0112281", -1, "54654675")],
columns=["from_id", "to_id", "score", "timestamp"])
alg = CentroidVector({'Plot': 'embedding'}, CosineSimilarity(), 0)
user_ratings = self.ratings.query('from_id == "A000"')
with self.assertRaises(OnlyNegativeItems):
alg.process_rated(user_ratings, self.movies_dir)
# No Item avilable locally
self.ratings = pd.DataFrame.from_records([
("A000", "non existent", 1, "54654675")],
columns=["from_id", "to_id", "score", "timestamp"])
alg = CentroidVector({'Plot': 'embedding'}, CosineSimilarity(), 0)
user_ratings = self.ratings.query('from_id == "A000"')
with self.assertRaises(NoRatedItems):
alg.process_rated(user_ratings, self.movies_dir)
def test_centroid_vector(self):
recs_number = 3
# Test prediction and ranking with the Centroid Vector algorithm
alg = CentroidVector({'Plot': ['tfidf', 'embedding']},
CosineSimilarity())
rs = ContentBasedRS(alg, ratings, self.movies_multiple)
# Prediction should raise error since it's not a ScorePredictionAlg
with self.assertRaises(NotPredictionAlg):
rs.fit_predict('A000')
# Test ranking with the Centroid Vector algorithm on specified items
result_rank_filtered = rs.fit_rank('A000',
filter_list=self.filter_list)
self.assertEqual(len(result_rank_filtered), len(self.filter_list))
# Test top-n ranking with the Centroid Vector algorithm
result_rank_numbered = rs.fit_rank('A000', recs_number=recs_number)
self.assertEqual(len(result_rank_numbered), recs_number)
def test_fit_cb_w_allitems_methodology(self):
rs = ContentBasedRS(
CentroidVector(
{"Plot": "tfidf"},
CosineSimilarity(),
), ratings, items_dir)
items = set([
os.path.splitext(f)[0] for f in os.listdir(items_dir)
if os.path.isfile(os.path.join(items_dir, f)) and f.endswith('xz')
])
em = EvalModel(rs,
KFoldPartitioning(),
metric_list=[Precision()],
methodology=AllItemsMethodology(items))
sys_result, users_result = em.fit()
self.assertIsInstance(sys_result, pd.DataFrame)
self.assertIsInstance(users_result, pd.DataFrame)
def setUp(self) -> None:
# ContentBasedAlgorithm is an abstract class, so we need to instantiate
# a subclass to test its methods. No initialization since we are not testing
# methods that need it
self.alg = CentroidVector({'Plot': 'tfidf'}, CosineSimilarity(), 0)