def test_base_test(self):
with self.test_session() as sess:
adj_mt = tf.convert_to_tensor(
np.array([[0, 1, 1], [1, 0, 1], [1, 1, 0]], dtype=np.int32))
points_data = np.array([[1], [2], [3]], dtype=np.float32)
A = DynamicAdjacentMatrix(adj_mt=adj_mt,
points_data=points_data,
edges_data=None,
edges_data_dim=1)
e_data_s = tf.gather(points_data, A.senders_indexs)
e_data_r = tf.gather(points_data, A.receivers_indexs)
e_data0 = e_data_r + e_data_s
e_data1 = e_data_r - e_data_s
A.edges_data = e_data0
edges_to_points_index = tf.stack(
[A.senders_indexs, A.receivers_indexs], axis=1)
self.assertAllClose(e_data0.eval(), [[3], [4], [3], [5], [4], [5]],
atol=1e-4)
self.assertAllClose(e_data1.eval(),
[[1], [2], [-1], [1], [-2], [-1]],
atol=1e-4)
self.assertAllEqual(A.real_edge_nr.eval(), 6)
self.assertAllEqual(
edges_to_points_index.eval(),
[[0, 1], [0, 2], [1, 0], [1, 2], [2, 0], [2, 1]])
wmlu.show_nparray(edges_to_points_index.eval())
target_length = [2, 2, 2]
target_p2e = [[[0, 1, 0], [2, 4, 0]], [[2, 3, 0], [0, 5, 0]],
[[4, 5, 0], [1, 3, 0]]]
'''data = A.points_to_edges_index[0].eval()
def test_update_points(self):
with self.test_session() as sess:
adj_mt = tf.convert_to_tensor(
np.array([[0, 1, 1], [1, 0, 1], [1, 1, 0]], dtype=np.int32))
points_data = np.array([[1], [2], [3]], dtype=np.float32)
edges_data = tf.constant(
np.array([[5], [6], [7], [8], [9], [10]], dtype=np.float32))
A = DynamicAdjacentMatrix(adj_mt=adj_mt,
points_data=points_data,
edges_data=edges_data)
#A.edges_reducer_for_points = tf.unsorted_segment_mean
A.global_attr = tf.convert_to_tensor(
np.array([[21]], dtype=np.float32))
def update_points(x):
def fn(x):
return tf.reshape(tf.reduce_sum(x), [1])
return tf.map_fn(fn, elems=(x))
A.update_points(update_points)
sess.run(tf.global_variables_initializer())
points_data = A.points_data.eval()
print(points_data)
self.assertAllClose(points_data, [[35.5], [38.0], [40.5]],
atol=1e-5)
def test_update_edges_independent(self):
with self.test_session() as sess:
adj_mt = tf.convert_to_tensor(
np.array([[0, 1, 1], [1, 0, 1], [1, 1, 0]], dtype=np.int32))
points_data = np.array([[1], [2], [3]], dtype=np.float32)
edges_data = tf.constant(
np.array([[5], [6], [7], [8], [9], [10]], dtype=np.float32))
A = DynamicAdjacentMatrix(adj_mt=adj_mt,
points_data=points_data,
edges_data=edges_data)
A.global_attr = tf.convert_to_tensor(
np.array([[21]], dtype=np.float32))
def update_edges(x):
def fn(x):
return x * 2
return tf.map_fn(fn, elems=(x))
A.update_edges_independent(update_edges)
self.assertAllClose(A.edges_data.eval(), edges_data * 2, atol=1e-5)
def test_update_globals(self):
with self.test_session() as sess:
adj_mt = tf.convert_to_tensor(
np.array([[0, 1, 1], [1, 0, 1], [1, 1, 0]], dtype=np.int32))
points_data = np.array([[1], [2], [3]], dtype=np.float32)
edges_data = tf.constant(
np.array([[5], [6], [7], [8], [9], [10]], dtype=np.float32))
A = DynamicAdjacentMatrix(adj_mt=adj_mt,
points_data=points_data,
edges_data=edges_data)
A.global_attr = tf.convert_to_tensor(
np.array([[21]], dtype=np.float32))
def update_global(x):
def fn(x):
return tf.reshape(tf.reduce_sum(x), [1])
return tf.map_fn(fn, elems=(x))
A.update_global(update_global)
self.assertAllClose(A.global_attr.eval(), [[30.5]], atol=1e-5)