def test_python_errors(rating_true, rating_pred):
"""Test raise errors."""
with pytest.raises(ValueError):
rmse(rating_true, rating_true, col_user="******")
with pytest.raises(ValueError):
mae(
rating_pred,
rating_pred,
col_rating=DEFAULT_PREDICTION_COL,
col_user="******",
)
with pytest.raises(ValueError):
rsquared(rating_true, rating_pred, col_item="not_item")
with pytest.raises(ValueError):
exp_var(
rating_pred,
rating_pred,
col_rating=DEFAULT_PREDICTION_COL,
col_item="not_item",
)
with pytest.raises(ValueError):
precision_at_k(rating_true, rating_pred, col_rating="not_rating")
with pytest.raises(ValueError):
recall_at_k(rating_true, rating_pred, col_prediction="not_prediction")
with pytest.raises(ValueError):
ndcg_at_k(rating_true, rating_true, col_user="******")
with pytest.raises(ValueError):
map_at_k(
rating_pred,
rating_pred,
col_rating=DEFAULT_PREDICTION_COL,
col_user="******",
)
def test_python_precision(rating_true, rating_pred, rating_nohit):
assert (
precision_at_k(
rating_true=rating_true,
rating_pred=rating_true,
col_prediction=DEFAULT_RATING_COL,
k=10,
)
== 0.6
)
assert precision_at_k(rating_true, rating_nohit, k=10) == 0.0
assert precision_at_k(rating_true, rating_pred, k=10) == pytest.approx(0.26666, TOL)
# Check normalization
single_user = pd.DataFrame(
{
DEFAULT_USER_COL: [1, 1, 1],
DEFAULT_ITEM_COL: [1, 2, 3],
DEFAULT_RATING_COL: [5, 4, 3],
}
)
assert (
precision_at_k(
rating_true=single_user,
rating_pred=single_user,
col_rating=DEFAULT_RATING_COL,
col_prediction=DEFAULT_RATING_COL,
k=3,
)
== 1
)
same_items = pd.DataFrame(
{
DEFAULT_USER_COL: [1, 1, 1, 2, 2, 2],
DEFAULT_ITEM_COL: [1, 2, 3, 1, 2, 3],
DEFAULT_RATING_COL: [5, 4, 3, 5, 5, 3],
}
)
assert (
precision_at_k(
rating_true=same_items,
rating_pred=same_items,
col_prediction=DEFAULT_RATING_COL,
k=3,
)
== 1
)
# Check that if the sample size is smaller than k, the maximum precision can not be 1
# if we do precision@5 when there is only 3 items, we can get a maximum of 3/5.
assert (
precision_at_k(
rating_true=same_items,
rating_pred=same_items,
col_prediction=DEFAULT_RATING_COL,
k=5,
)
== 0.6
)