def test_hpf_train_large(tmp_path):
algo = hpf.HPF(20)
ratings = lktu.ml_pandas.renamed.ratings
ratings = ratings.assign(rating=ratings.rating + 0.5)
algo.fit(ratings)
assert algo.n_users == ratings.user.nunique()
assert algo.n_items == ratings.item.nunique()
mfile = tmp_path / 'hpf.dat'
with mfile.open('wb') as mf:
pickle.dump(algo, mf)
with mfile.open('rb') as mf:
a2 = pickle.load(mf)
assert a2.n_users == algo.n_users
assert a2.n_items == algo.n_items
csel = basic.UnratedItemCandidateSelector()
csel.fit(ratings)
rec = basic.TopN(algo, csel)
for u in np.random.choice(ratings.user.unique(), size=50, replace=False):
recs = rec.recommend(u, 50)
assert len(recs) == 50
assert recs.item.nunique() == 50
def test_unrated_selector():
sel = basic.UnratedItemCandidateSelector()
s2 = sel.fit(simple_df)
assert s2 is sel
assert set(sel.candidates(10)) == set([3])
assert set(sel.candidates(12)) == set([3, 2])
assert set(sel.candidates(11)) == set([1, 2, 3])
def test_unrated_big():
ratings = lktu.ml_test.ratings
users = ratings.user.unique()
items = ratings.item.unique()
user_items = ratings.set_index('user').item
sel = basic.UnratedItemCandidateSelector()
s2 = sel.fit(ratings)
assert s2 is sel
# test 100 random users
for u in np.random.choice(users, 100, False):
candidates = sel.candidates(u)
candidates = pd.Series(candidates)
uis = user_items.loc[u]
assert len(uis) + len(candidates) == len(items)
assert candidates.nunique() == len(candidates)
assert all(~candidates.isin(uis))
def test_unrated_override():
sel = basic.UnratedItemCandidateSelector()
sel.fit(simple_df)
assert set(sel.candidates(10, [2])) == set([1, 3])
