def test_vocab_size():
X = ["One", "Two", "Three"]
vocab_size = 1
keras_vectorizer = KerasVectorizer(vocab_size=vocab_size)
X_vec = keras_vectorizer.fit_transform(X)
assert X_vec.max() == vocab_size
def test_sequence_length():
X = ["One", "Two", "Three"]
sequence_length = 5
keras_vectorizer = KerasVectorizer(sequence_length=sequence_length)
X_vec = keras_vectorizer.fit_transform(X)
assert X_vec.shape[1] == sequence_length
def test_vanilla():
X = ["One", "Two", "Three Four"]
keras_vectorizer = KerasVectorizer()
X_vec = keras_vectorizer.fit_transform(X)
assert X_vec.shape[0] == 3
assert X_vec.shape[1] == 2
assert X_vec.max() == 5 # 4 tokens including OOV
def test_XY_dataset_sparse_y():
X = ["One and two", "One only", "Two nothing else", "Two and three"]
Y = np.array([[1, 1, 0, 0], [1, 0, 0, 0], [0, 1, 0, 0], [0, 1, 1, 0]])
Y_sparse = csr_matrix(Y)
vec = KerasVectorizer()
X_vec = vec.fit_transform(X)
data = tf.data.Dataset.from_tensor_slices((X_vec, Y))
data = data.shuffle(100, seed=42)
clf = CNNClassifier(batch_size=2, sparse_y=True, multilabel=True)
clf.fit(data)
assert clf.score(data, Y_sparse) > 0.3
def test_XY_dataset():
X = ["One", "One only", "Two nothing else", "Two and three"]
Y = np.array([0, 0, 1, 1])
vec = KerasVectorizer()
X_vec = vec.fit_transform(X)
data = tf.data.Dataset.from_tensor_slices((X_vec, Y))
data = data.shuffle(100, seed=42)
clf = CNNClassifier(batch_size=2)
clf.fit(data)
assert clf.score(data, Y) > 0.3
def test_save_load():
X = ["One", "One only", "Two nothing else", "Two and three"]
Y = np.array([0, 0, 1, 1])
vec = KerasVectorizer()
X_vec = vec.fit_transform(X)
model = CNNClassifier()
model.fit(X_vec, Y)
with tempfile.TemporaryDirectory() as tmp_dir:
model.save(tmp_dir)
loaded_model = CNNClassifier()
loaded_model.load(tmp_dir)
assert hasattr(loaded_model, 'model')
assert loaded_model.score(X_vec, Y) > 0.6
