def test__similar_users__bad():
"""ensure the similar_users() method raises an exception for an user not in training data"""
with pytest.raises(AssertionError):
model = RankFM(factors=2)
model.fit(intx_train_pd_int)
similar = model.similar_users(9, n_users=1)
def test__similar_users__good():
"""test the similar_users() method for a valid [user_id]"""
model = RankFM(factors=2)
model.fit(intx_train_pd_int)
similar = model.similar_users(1, n_users=2)
shape = similar.shape == (2, )
users = np.isin(similar, intx_train_pd_int['user_id'].unique()).all()
assert shape and users
def test__fit__good(interactions, user_features, item_features):
"""assert that the model can be successfully fit on the input data"""
model = RankFM(factors=2)
model.fit(interactions,
user_features,
item_features,
epochs=2,
verbose=True)
assert model.is_fit
def test__similar_items__good():
"""test the similar_items() method for a valid [item_id]"""
model = RankFM(factors=2)
model.fit(intx_train_pd_int)
similar = model.similar_items(1, n_items=3)
shape = similar.shape == (3, )
items = np.isin(similar, intx_train_pd_int['item_id'].unique()).all()
assert shape and items
def test__predict__good__disjoint_drop():
"""test the predict() method on disjoint validation pairs with the cold_start='drop' option"""
model = RankFM(factors=2)
model.fit(intx_train_pd_int)
scores = model.predict(intx_valid_disjoint, cold_start='drop')
shape = scores.shape == (5, )
dtype = scores.dtype == np.float32
nmiss = np.sum(np.isnan(scores).astype(np.int32)) == 0
assert shape and dtype and nmiss
def test__predict__good__train():
"""test the predict() method on the training inputs"""
model = RankFM(factors=2)
model.fit(intx_train_pd_int)
scores = model.predict(intx_train_pd_int)
shape = scores.shape == (9, )
dtype = scores.dtype == np.float32
nmiss = np.sum(np.isnan(scores).astype(np.int32)) == 0
assert shape and dtype and nmiss
def test__recommend__good__train():
"""test the recommend() method on the training users"""
model = RankFM(factors=2)
model.fit(intx_train_pd_int)
recs = model.recommend(train_users, n_items=3)
klass = isinstance(recs, pd.DataFrame)
shape = recs.shape == (3, 3)
index = np.array_equal(recs.index.values, train_users)
items = recs.isin(intx_train_pd_int['item_id'].values).all().all()
assert klass and shape and index and items
def test__recommend__good__valid__nan():
"""test the recommend() method on a disjoint set of validation users"""
model = RankFM(factors=2)
model.fit(intx_train_pd_int)
recs = model.recommend(valid_users, n_items=3, cold_start='nan')
klass = isinstance(recs, pd.DataFrame)
shape = recs.shape == (4, 3)
index = np.array_equal(sorted(recs.index.values), sorted(valid_users))
items = recs.dropna().isin(intx_train_pd_int['item_id'].values).all().all()
new_users = list(set(valid_users) - set(train_users))
nmiss = recs.loc[new_users].isnull().all().all()
assert klass and shape and index and items and nmiss
def test__recommend__good__valid__drop():
"""test the recommend() method on a disjoint set of validation users"""
model = RankFM(factors=2)
model.fit(intx_train_pd_int)
recs = model.recommend(valid_users, n_items=3, cold_start='drop')
klass = isinstance(recs, pd.DataFrame)
shape = recs.shape == (2, 3)
index = np.isin(recs.index.values, valid_users).all()
items = recs.dropna().isin(intx_train_pd_int['item_id'].values).all().all()
same_users = list(set(valid_users) & set(train_users))
match_users = np.array_equal(sorted(same_users), sorted(recs.index.values))
assert klass and shape and index and items and match_users
def test__recommend__good__train__filter():
"""test the recommend() method on the training users but filter previous items"""
model = RankFM(factors=2)
model.fit(intx_train_pd_int)
recs = model.recommend(train_users, n_items=3, filter_previous=True)
klass = isinstance(recs, pd.DataFrame)
shape = recs.shape == (3, 3)
index = np.array_equal(recs.index.values, train_users)
items = recs.isin(intx_train_pd_int['item_id'].values).all().all()
recs_long = recs.stack().reset_index().drop('level_1', axis=1)
recs_long.columns = ['user_id', 'item_id']
intersect = pd.merge(intx_train_pd_int,
recs_long,
on=['user_id', 'item_id'],
how='inner').empty
assert klass and shape and index and items and intersect
def model(interactions, item_features, factors, max_samples, epochs):
"""Builds a Factorization Machine recommendation model.
Args:
interactions: A dataframe of user/item interactions.
item_features: A dataframe of item features matrix.
factors: The latent factor rank (hyperparameter).
max_samples: The maximum number of negative samples to draw for WARP loss (hyperparameter).
epochs: Number of training epochs.
Returns:
model: A RankFM model object.
"""
model = RankFM(factors=factors,
loss='warp',
max_samples=max_samples,
alpha=0.01,
sigma=0.1,
learning_rate=0.1,
learning_schedule='invscaling')
model.fit(interactions, item_features=item_features, epochs=epochs, verbose=True)
return model
def test__fit__bad__if_str_cols():
"""ensure that the [item_features] does not contain string columns"""
with pytest.raises(ValueError):
model = RankFM(factors=2)
model.fit(intx_train_pd_int, item_features=if_str_cols)
def test__fit__bad__if_no_id():
"""ensure that the [item_features] contains a [item_id] column"""
with pytest.raises(KeyError):
model = RankFM(factors=2)
model.fit(intx_train_pd_int, item_features=if_no_id)
def test__fit__bad__rating_col():
"""ensure that having more than 2 columns in the interaction data causes an assertion failure"""
with pytest.raises(AssertionError):
model = RankFM(factors=2)
model.fit(intx_train_pd_rating)