def test_hungarian_reorder_notypes(self):
X = np.random.normal(size=(10, 100, 5))
ix = np.arange(100)
Xshuffled = []
for i in range(len(X)):
np.random.shuffle(ix)
Xshuffled.append(X[i][ix])
Xshuffled = np.stack(Xshuffled)
D1 = tf.constant(utils.to_distance_matrices(X), dtype=tf.float32)
D2 = tf.constant(utils.to_distance_matrices(Xshuffled), dtype=tf.float32)
D1_reordered, t1_reordered = layers.HungarianReorder()([D1, D2])
np.testing.assert_array_almost_equal(D1_reordered, D2, decimal=5)
def usage_numpy():
X = np.random.normal(size=(10, 100, 5)).astype(np.float32)
types = np.random.randint(0, 10, size=(10, 100), dtype=np.int32)
ix = np.arange(100)
Xshuffled = []
types_shuffled = []
for i in range(len(X)):
np.random.shuffle(ix)
Xshuffled.append(X[i][ix])
types_shuffled.append(types[i][ix])
Xshuffled = np.stack(Xshuffled)
types_shuffled = np.stack(types_shuffled)
D1 = utils.to_distance_matrices(X)
D2 = utils.to_distance_matrices(Xshuffled)
D1reordered, types_reordered = utils.reorder_distance_matrices(
D1, D2, types1=types, types2=types_shuffled)
np.testing.assert_array_almost_equal(D2, D1reordered)
np.testing.assert_array_almost_equal(types_shuffled, types_reordered)
def test_D2M_M2D(self):
X = tf.random.normal(stddev=10, shape=(100, 20, 20))
D = layers.to_distmat(X)
np.testing.assert_array_almost_equal(D, utils.to_distance_matrices(X), decimal=3)
M = layers.D2M()(D)
Mref = []
for d in D.numpy():
Mref.append(utils.to_M_matrix(d))
Mref = np.stack(Mref)
np.testing.assert_array_almost_equal(M, Mref, decimal=3)
D2 = layers.M2D()(M)
np.testing.assert_array_almost_equal(D2, D, decimal=3)
def test_hungarian_reorder(self):
X = np.random.normal(size=(10, 100, 5))
types = np.random.randint(0, 10, size=(10, 100), dtype=np.int32)
ix = np.arange(100)
Xshuffled = []
types_shuffled = []
for i in range(len(X)):
np.random.shuffle(ix)
Xshuffled.append(X[i][ix])
types_shuffled.append(types[i][ix])
Xshuffled = np.stack(Xshuffled)
types_shuffled = np.stack(types_shuffled)
D1 = tf.constant(utils.to_distance_matrices(X), dtype=tf.float32)
D2 = tf.constant(utils.to_distance_matrices(Xshuffled), dtype=tf.float32)
t1 = tf.constant(types, dtype=tf.int32)
t2 = tf.constant(types_shuffled, dtype=tf.int32)
D1_reordered, t1_reordered = layers.HungarianReorder()([D1, D2, t1, t2])
np.testing.assert_array_almost_equal(D1_reordered, D2)
np.testing.assert_array_almost_equal(t1_reordered, t2)
def test_D2T(self):
X = tf.random.normal(stddev=10, shape=(100, 20, 20))
Tref = np.stack([utils.to_T_matrix(d) for d in utils.to_distance_matrices(X)])
D = layers.to_distmat(X)
T = layers.D2T(n_atoms=20)(D)
np.testing.assert_array_almost_equal(T, Tref, decimal=3)