def test_predict_not_fitted():
model = LightFM()
with pytest.raises(ValueError):
model.predict(np.arange(10), np.arange(10))
with pytest.raises(ValueError):
model.predict_rank(1)
with pytest.raises(ValueError):
model.get_user_representations()
with pytest.raises(ValueError):
model.get_item_representations()
def test_predict_not_fitted():
model = LightFM()
with pytest.raises(ValueError):
model.predict(np.arange(10), np.arange(10))
with pytest.raises(ValueError):
model.predict_rank(1)
with pytest.raises(ValueError):
model.get_user_representations()
with pytest.raises(ValueError):
model.get_item_representations()
def test_get_representations():
model = LightFM(random_state=SEED)
model.fit_partial(train, epochs=10)
num_users, num_items = train.shape
for (item_features, user_features) in (
(None, None),
(
(sp.identity(num_items) + sp.random(num_items, num_items)),
(sp.identity(num_users) + sp.random(num_users, num_users)),
),
):
test_predictions = model.predict(test.row,
test.col,
user_features=user_features,
item_features=item_features)
item_biases, item_latent = model.get_item_representations(
item_features)
user_biases, user_latent = model.get_user_representations(
user_features)
assert item_latent.dtype == np.float32
assert user_latent.dtype == np.float32
predictions = (
(user_latent[test.row] * item_latent[test.col]).sum(axis=1) +
user_biases[test.row] + item_biases[test.col])
assert np.allclose(test_predictions, predictions, atol=0.000001)
def test_get_representations():
model = LightFM(random_state=SEED)
model.fit_partial(train, epochs=10)
num_users, num_items = train.shape
for (item_features, user_features) in (
(None, None),
(
(sp.identity(num_items) + sp.random(num_items, num_items)),
(sp.identity(num_users) + sp.random(num_users, num_users)),
),
):
test_predictions = model.predict(
test.row, test.col, user_features=user_features, item_features=item_features
)
item_biases, item_latent = model.get_item_representations(item_features)
user_biases, user_latent = model.get_user_representations(user_features)
assert item_latent.dtype == np.float32
assert user_latent.dtype == np.float32
predictions = (
(user_latent[test.row] * item_latent[test.col]).sum(axis=1)
+ user_biases[test.row]
+ item_biases[test.col]
)
assert np.allclose(test_predictions, predictions, atol=0.000001)
