def test_contrastive_distance_setter():
new_base = mlp.Base()
new_siamese = contrastive.ContrastiveSiamese(new_base)
try:
new_siamese.distance = "a"
except:
new_siamese.distance == "L2"
assert new_siamese.distance == "L2"
def test_contrastive_margin_setter():
new_base = mlp.Base()
new_siamese = contrastive.ContrastiveSiamese(new_base)
try:
new_siamese.margin = -1
except:
new_siamese.margin = 1.0
assert new_siamese.margin == 1.0
def test_contrastive_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 = contrastive.ContrastiveSiamese(new_base)
new_siamese.compile(optimizer=tf.optimizers.Adam(learning_rate=0.001))
new_siamese.fit(train.batches, epochs=1)
def test_contrastive_step():
(x, y), (_, _) = tf.keras.datasets.mnist.load_data()
new_base = mlp.MLP()
new_siamese = contrastive.ContrastiveSiamese(new_base)
new_siamese.compile(optimizer=tf.optimizers.Adam(learning_rate=0.001))
x1 = tf.ones((1, 784))
x2 = tf.ones((1, 784))
y = tf.zeros(1)
new_siamese.step(x1, x2, y)
def test_contrastive_margin():
new_base = mlp.Base()
new_siamese = contrastive.ContrastiveSiamese(new_base)
assert new_siamese.margin == 1.0
def test_contrastive_distance():
new_base = mlp.Base()
new_siamese = contrastive.ContrastiveSiamese(new_base)
assert new_siamese.distance == "L2"