def test_Interactions_with_0_ratings(interactions_pandas, df_for_interactions_with_0_ratings):
with pytest.warns(UserWarning):
interactions_with_0s = Interactions(users=df_for_interactions_with_0_ratings['user_id'],
items=df_for_interactions_with_0_ratings['item_id'],
ratings=df_for_interactions_with_0_ratings['ratings'],
check_num_negative_samples_is_valid=False)
assert np.array_equal(interactions_pandas.toarray(), interactions_with_0s.toarray())
def test_items_length_bad(self, df_for_interactions_with_missing_ids):
with pytest.raises(ValueError):
Interactions(users=df_for_interactions_with_missing_ids['user_id'],
items=df_for_interactions_with_missing_ids['item_id'][:-1],
ratings=df_for_interactions_with_missing_ids['ratings'],
allow_missing_ids=True,
check_num_negative_samples_is_valid=False)
def movielens_implicit_interactions(movielens_implicit_df):
return Interactions(users=movielens_implicit_df['user_id'],
items=movielens_implicit_df['item_id'],
ratings=movielens_implicit_df['rating'],
num_negative_samples=10,
max_number_of_samples_to_consider=200,
allow_missing_ids=True)
def test_implicit_interactions():
return Interactions(
users=[0, 0, 0, 1, 1, 1, 2, 2],
items=[0, 1, 2, 1, 2, 3, 0, 2],
ratings=[1, 1, 1, 1, 1, 1, 1, 1],
check_num_negative_samples_is_valid=False,
)
def df_to_interactions(df: pd.DataFrame,
user_col: str = 'user_id',
item_col: str = 'item_id',
ratings_col: Optional[str] = 'rating',
**kwargs) -> Interactions:
"""
Helper function to convert a DataFrame to an ``Interactions`` object.
Parameters
-------------
df: pd.DataFrame
Dataframe with columns for user IDs, item IDs, and (optionally) ratings
user_col: str
Column name for the user IDs
item_col: str
Column name for the item IDs
ratings_col: str
Column name for the ratings column. If ``None``, will default to ratings of all 1s
**kwargs
Keyword arguments to pass to ``Interactions``
Returns
-------------
interactions: collie_recs.interactions.Interactions
"""
ratings = df[ratings_col] if ratings_col is not None else None
return Interactions(users=df[user_col],
items=df[item_col],
ratings=ratings,
**kwargs)
def test_duplicate_user_item_pairs(self,
interactions_pandas,
df_for_interactions_with_duplicates):
duplicated_interactions = Interactions(users=df_for_interactions_with_duplicates['user_id'],
items=df_for_interactions_with_duplicates['item_id'],
check_num_negative_samples_is_valid=False)
assert duplicated_interactions.mat.getnnz() == interactions_pandas.mat.getnnz()
def test_Interactions_exact_negative_samples_num_negative_samples_too_large(
self,
ratings_matrix_for_interactions,
):
with pytest.raises(AssertionError):
Interactions(mat=ratings_matrix_for_interactions,
max_number_of_samples_to_consider=200,
num_negative_samples=8)
def test_Interactions_with_missing_ids_raises_error(
self,
df_for_interactions_with_missing_ids,
ratings_matrix_for_interactions_with_missing_ids,
sparse_ratings_matrix_for_interactions_with_missing_ids
):
with pytest.raises(ValueError):
Interactions(mat=ratings_matrix_for_interactions_with_missing_ids,
check_num_negative_samples_is_valid=False)
with pytest.raises(ValueError):
Interactions(mat=sparse_ratings_matrix_for_interactions_with_missing_ids,
check_num_negative_samples_is_valid=False)
with pytest.raises(ValueError):
Interactions(users=df_for_interactions_with_missing_ids['user_id'],
items=df_for_interactions_with_missing_ids['item_id'],
ratings=df_for_interactions_with_missing_ids['ratings'],
check_num_negative_samples_is_valid=False)
def test_Interactions_with_missing_ids(
self,
df_for_interactions_with_missing_ids,
ratings_matrix_for_interactions_with_missing_ids,
sparse_ratings_matrix_for_interactions_with_missing_ids
):
Interactions(mat=ratings_matrix_for_interactions_with_missing_ids,
allow_missing_ids=True,
check_num_negative_samples_is_valid=False)
Interactions(mat=sparse_ratings_matrix_for_interactions_with_missing_ids,
allow_missing_ids=True,
check_num_negative_samples_is_valid=False)
Interactions(users=df_for_interactions_with_missing_ids['user_id'],
items=df_for_interactions_with_missing_ids['item_id'],
ratings=df_for_interactions_with_missing_ids['ratings'],
allow_missing_ids=True,
check_num_negative_samples_is_valid=False)
def test_Interactions_approximate_negative_samples(self, ratings_matrix_for_interactions):
interactions = Interactions(mat=ratings_matrix_for_interactions,
num_negative_samples=NUM_NEGATIVE_SAMPLES,
max_number_of_samples_to_consider=0,
seed=42)
assert interactions.positive_items == {}
for _ in range(3):
_, negative_samples = interactions[0]
assert len(negative_samples) == 3
def test_Interactions(ratings_matrix_for_interactions,
sparse_ratings_matrix_for_interactions,
df_for_interactions):
interactions_1 = Interactions(mat=ratings_matrix_for_interactions,
check_num_negative_samples_is_valid=False)
interactions_2 = Interactions(mat=sparse_ratings_matrix_for_interactions,
check_num_negative_samples_is_valid=False)
interactions_3 = Interactions(users=df_for_interactions['user_id'],
items=df_for_interactions['item_id'],
ratings=df_for_interactions['rating'],
check_num_negative_samples_is_valid=False)
np.testing.assert_equal(interactions_1.toarray(), interactions_2.toarray())
np.testing.assert_equal(interactions_1.toarray(), interactions_3.toarray())
assert interactions_1.num_users == interactions_2.num_users == interactions_3.num_users
assert interactions_1.num_items == interactions_2.num_items == interactions_3.num_items
assert (interactions_1.num_interactions == interactions_2.num_interactions
== interactions_3.num_interactions)
def test_Interactions_approximate_negative_samples_partway_through(
ratings_matrix_for_interactions):
with pytest.warns(UserWarning):
interactions = Interactions(mat=ratings_matrix_for_interactions,
num_negative_samples=NUM_NEGATIVE_SAMPLES,
max_number_of_samples_to_consider=1,
seed=42)
assert interactions.positive_items != {}
for _ in range(3):
_, negative_samples = interactions[0]
assert len(negative_samples) == 3
def interactions_to_split_with_a_user_with_only_one_interaction():
df = pd.DataFrame(
data={
'user_id':
[0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 4, 4, 4, 4],
'item_id':
[0, 1, 2, 3, 4, 5, 6, 7, 8, 1, 2, 3, 4, 1, 2, 3, 4, 1, 1, 2, 4, 5],
'rating':
[1, 2, 3, 4, 5, 4, 3, 2, 1, 1, 2, 3, 4, 2, 3, 4, 5, 1, 2, 3, 5, 4]
})
return Interactions(users=df['user_id'],
items=df['item_id'],
ratings=df['rating'],
check_num_negative_samples_is_valid=False)
def test_Interactions_approximate_negative_samples_many_users(
ratings_matrix_for_interactions):
interactions = Interactions(mat=ratings_matrix_for_interactions,
num_negative_samples=NUM_NEGATIVE_SAMPLES,
max_number_of_samples_to_consider=0,
seed=42)
assert interactions.positive_items == {}
for _ in range(3):
_, negative_samples = interactions[list(range(NUM_USERS_TO_GENERATE))]
assert len(negative_samples) == NUM_USERS_TO_GENERATE
for negative_sample in negative_samples:
assert len(negative_sample) == NUM_NEGATIVE_SAMPLES
def _subset_interactions(interactions: Interactions, idxs: Iterable[int]) -> Interactions:
idxs = np.array(idxs)
coo_mat = coo_matrix(
(interactions.mat.data[idxs], (interactions.mat.row[idxs],
interactions.mat.col[idxs])),
shape=(interactions.num_users, interactions.num_items)
)
return Interactions(
mat=coo_mat,
num_negative_samples=interactions.num_negative_samples,
allow_missing_ids=True,
num_users=interactions.num_users,
num_items=interactions.num_items,
check_num_negative_samples_is_valid=False,
max_number_of_samples_to_consider=interactions.max_number_of_samples_to_consider,
seed=interactions.seed,
)
def test_Interactions_exact_negative_samples(self, ratings_matrix_for_interactions):
interactions = Interactions(mat=ratings_matrix_for_interactions,
num_negative_samples=NUM_NEGATIVE_SAMPLES,
max_number_of_samples_to_consider=200,
seed=42)
assert interactions.positive_items != {}
all_negative_samples = list()
for _ in range(10):
_, negative_samples = interactions[0]
assert len(negative_samples) == NUM_NEGATIVE_SAMPLES
for negative_sample in negative_samples:
assert negative_sample.item() not in ratings_matrix_for_interactions[0].nonzero()[0]
all_negative_samples += negative_samples.tolist()
assert len(set(all_negative_samples)) > NUM_NEGATIVE_SAMPLES
def test_Interactions_exact_negative_samples_many_users(self, ratings_matrix_for_interactions):
interactions = Interactions(mat=ratings_matrix_for_interactions,
num_negative_samples=NUM_NEGATIVE_SAMPLES,
max_number_of_samples_to_consider=200,
seed=42)
assert interactions.positive_items != {}
for _ in range(10):
(user_ids, _), negative_samples = interactions[list(range(NUM_USERS_TO_GENERATE))]
assert len(negative_samples) == NUM_USERS_TO_GENERATE
for idx, user_id in enumerate(user_ids):
assert len(negative_samples[idx]) == NUM_NEGATIVE_SAMPLES
for negative_sample in negative_samples[idx]:
assert (
negative_sample.item()
not in ratings_matrix_for_interactions[user_id].nonzero()[0]
)
def test_stratified_split(interactions_to_split):
train_expected_df = pd.DataFrame(
data={
'user_id': [0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 3, 4, 4],
'item_id': [1, 2, 3, 4, 6, 8, 1, 2, 3, 4, 2, 4, 5],
'rating': [2, 3, 4, 5, 3, 1, 1, 2, 4, 5, 5, 5, 4],
})
train_expected = Interactions(
mat=coo_matrix(
(
train_expected_df['rating'],
(train_expected_df['user_id'], train_expected_df['item_id']),
),
shape=(interactions_to_split.num_users,
interactions_to_split.num_items),
),
allow_missing_ids=True,
check_num_negative_samples_is_valid=False,
)
validate_expected_df = pd.DataFrame(
data={
'user_id': [0, 1, 2, 3, 4],
'item_id': [7, 3, 2, 1, 2],
'rating': [2, 3, 3, 1, 3],
})
validate_expected = Interactions(
mat=coo_matrix(
(
validate_expected_df['rating'],
(validate_expected_df['user_id'],
validate_expected_df['item_id']),
),
shape=(interactions_to_split.num_users,
interactions_to_split.num_items),
),
allow_missing_ids=True,
check_num_negative_samples_is_valid=False,
)
test_expected_df = pd.DataFrame(
data={
'user_id': [0, 0, 1, 2, 3, 4],
'item_id': [0, 5, 4, 1, 4, 1],
'rating': [1, 4, 4, 2, 4, 2],
})
test_expected = Interactions(
mat=coo_matrix(
(
test_expected_df['rating'],
(test_expected_df['user_id'], test_expected_df['item_id']),
),
shape=(interactions_to_split.num_users,
interactions_to_split.num_items),
),
allow_missing_ids=True,
check_num_negative_samples_is_valid=False,
)
(train_actual, validate_actual,
test_actual) = stratified_split(interactions=interactions_to_split,
val_p=0.1,
test_p=0.2,
seed=46)
np.testing.assert_array_equal(train_actual.toarray(),
train_expected.toarray())
np.testing.assert_array_equal(validate_actual.toarray(),
validate_expected.toarray())
np.testing.assert_array_equal(test_actual.toarray(),
test_expected.toarray())
assert (train_actual.num_users == train_expected.num_users ==
validate_actual.num_users == validate_expected.num_users ==
test_actual.num_users == test_expected.num_users)
assert (train_actual.num_items == train_expected.num_items ==
validate_actual.num_items == validate_expected.num_items ==
test_actual.num_items == test_expected.num_items)
def interactions_sparse_matrix(sparse_ratings_matrix_for_interactions):
return Interactions(mat=sparse_ratings_matrix_for_interactions,
check_num_negative_samples_is_valid=False)
def interactions_pandas(df_for_interactions):
return Interactions(users=df_for_interactions['user_id'],
items=df_for_interactions['item_id'],
ratings=df_for_interactions['ratings'],
check_num_negative_samples_is_valid=False)
def test_ratings_None_but_its_okay(self, df_for_interactions):
Interactions(users=df_for_interactions['user_id'],
items=df_for_interactions['item_id'],
ratings=None,
check_num_negative_samples_is_valid=False)
def test_users_None(self, df_for_interactions):
with pytest.raises(AssertionError):
Interactions(users=None,
items=df_for_interactions['item_id'],
ratings=df_for_interactions['ratings'],
check_num_negative_samples_is_valid=False)
def run_movielens_example(epochs: int = 20, gpus: int = 0) -> None:
"""
Retrieve and split data, train and evaluate a model, and save it.
From the terminal, you can run this script with:
.. code-block:: bash
python collie_recs/movielens/run.py --epochs 20
Parameters
-------------
epochs: int
Number of epochs for model training
gpus: int
Number of gpus to train on
"""
t = Timer()
t.timecheck(' 1.0 - retrieving MovieLens 100K dataset')
df = read_movielens_df(decrement_ids=True)
t.timecheck(' 1.0 complete')
t.timecheck(' 2.0 - splitting data')
df_imp = convert_to_implicit(df)
interactions = Interactions(users=df_imp['user_id'],
items=df_imp['item_id'],
allow_missing_ids=True)
train, val, test = stratified_split(interactions, val_p=0.1, test_p=0.1)
train_loader = InteractionsDataLoader(train, batch_size=1024, shuffle=True)
val_loader = InteractionsDataLoader(val, batch_size=1024, shuffle=False)
t.timecheck(' 2.0 complete')
t.timecheck(' 3.0 - training the model')
model = MatrixFactorizationModel(train=train_loader,
val=val_loader,
dropout_p=0.05,
loss='adaptive',
lr=5e-2,
embedding_dim=10,
optimizer='adam',
weight_decay=1e-7)
trainer = CollieTrainer(
model=model,
gpus=gpus,
max_epochs=epochs,
deterministic=True,
logger=False,
checkpoint_callback=False,
callbacks=[EarlyStopping(monitor='val_loss_epoch', mode='min')],
weights_summary='full',
terminate_on_nan=True)
trainer.fit(model)
model.eval()
t.timecheck('\n 3.0 complete')
t.timecheck(' 4.0 - evaluating model')
auc_score, mrr_score, mapk_score = evaluate_in_batches([auc, mrr, mapk],
test,
model,
k=10)
print(f'AUC: {auc_score}')
print(f'MRR: {mrr_score}')
print(f'MAP@10: {mapk_score}')
t.timecheck(' 4.0 complete')
t.timecheck(' 5.0 - saving model')
absolute_data_path = DATA_PATH / 'fitted_model'
model.save_model(absolute_data_path)
t.timecheck(' 5.0 complete')
