def test_balanced_pair_dataset_n_pairs():
x = np.zeros((2, 784))
try:
y = np.zeros(2)
new_balanced_pair_dataset = pair.BalancedPairDataset(x, y)
except:
y = np.asarray([0, 1])
new_balanced_pair_dataset = pair.BalancedPairDataset(x, y)
assert new_balanced_pair_dataset.n_pairs == 2
def test_balanced_pair_dataset_batches():
x = np.zeros((2, 784))
y = np.asarray([0, 1])
new_balanced_pair_dataset = pair.BalancedPairDataset(x, y)
assert new_balanced_pair_dataset.batches is not None
def test_cross_entropy_fit():
(x, y), (_, _) = tf.keras.datasets.mnist.load_data()
train = pair.BalancedPairDataset(x, y, n_pairs=10, input_shape=(x.shape[0], 784))
new_base = mlp.MLP()
new_siamese = cross_entropy.CrossEntropySiamese(new_base)
new_siamese.compile(optimizer=tf.optimizers.Adam(learning_rate=0.001))
new_siamese.fit(train.batches, epochs=1)
def test_balanced_pair_dataset_create_pairs():
x = np.zeros((2, 784))
y = np.asarray([0, 1])
new_balanced_pair_dataset = pair.BalancedPairDataset(x, y)
pairs = new_balanced_pair_dataset.create_pairs(x, y)
assert len(pairs) == 3
def test_balanced_pair_dataset_build():
x = np.zeros((2, 784))
y = np.asarray([0, 1])
new_balanced_pair_dataset = pair.BalancedPairDataset(x, y, shuffle=True)
pairs = new_balanced_pair_dataset.create_pairs(x, y)
new_balanced_pair_dataset._build(pairs)
assert new_balanced_pair_dataset.batches is not None
new_balanced_pair_dataset = pair.BalancedPairDataset(x, y, shuffle=False)
pairs = new_balanced_pair_dataset.create_pairs(x, y)
new_balanced_pair_dataset._build(pairs)
assert new_balanced_pair_dataset.batches is not None
def test_balanced_pair_dataset_batches_setter():
x = np.zeros((2, 784))
y = np.asarray([0, 1])
new_balanced_pair_dataset = pair.BalancedPairDataset(x, y)
try:
new_balanced_pair_dataset.batches = 1
except:
pass
assert new_balanced_pair_dataset.batches != 1
def test_contrastive_predict():
(x, y), (_, _) = tf.keras.datasets.mnist.load_data()
train = pair.BalancedPairDataset(x,
y,
n_pairs=10,
input_shape=(x.shape[0], 784))
new_base = mlp.MLP()
new_siamese = contrastive.ContrastiveSiamese(new_base)
new_siamese.compile(optimizer=tf.optimizers.Adam(learning_rate=0.001))
new_siamese.fit(train.batches, epochs=1)
x1 = tf.ones((1, 784))
x2 = tf.ones((1, 784))
new_siamese.distance = "L1"
new_siamese.predict(x1, x2)
new_siamese.distance = "L2"
new_siamese.predict(x1, x2)
new_siamese.distance = "angular"
new_siamese.predict(x1, x2)
