def test_session_neg():
a = ad.constant(1, 'a')
assert a.value == 1
c_true = ad.constant(-1, 'c')
c = ad.neg(a) # build graph
with ad.Session() as sess:
# eval graph
c_out = sess.run(c)
assert c_out.value == c_true.value
def test_session_reciprocal():
a = ad.constant(3, 'a')
assert a.value == 3
c_true = ad.constant(1 / 3, 'c')
c = ad.reciprocal(a) # build graph
with ad.Session() as sess:
# eval graph
c_out = sess.run(c)
assert c_out.value == c_true.value
def test_session_log():
a = ad.constant(3, 'a')
assert a.value == 3
c_true = ad.constant(math.log(3), 'c')
c = ad.log(a) # build graph
with ad.Session() as sess:
# eval graph
c_out = sess.run(c)
assert c_out.value == c_true.value
def test_session_pow():
a = ad.constant(2, 'a')
b = ad.constant(3, 'b')
assert a.value == 2
assert b.value == 3
c_true = ad.constant(2**3, 'c')
c = ad.pow(a, b) # build graph
with ad.Session() as sess:
# eval graph
c_out = sess.run(c)
assert c_out.value == c_true.value
def test_session_div():
a = ad.constant(1, 'a')
b = ad.constant(2, 'b')
assert a.value == 1
assert b.value == 2
c_true = ad.constant(1 / 2, 'c')
c = ad.div(a, b) # build graph
with ad.Session() as sess:
# eval graph
c_out = sess.run(c)
assert c_out.value == c_true.value
def test_constant():
# int value
a = ad.constant(1, 'a')
assert a.value == 1
assert a.name == 'a'
# float value
a = ad.constant(0.5, 'a')
assert a.value == 0.5
assert a.name == 'a'
# numpy array
np_a = np.random.rand(5)
a = ad.constant(np_a, 'a')
assert np.array_equal(a.value, np_a)
assert a.name == 'a'
def test_trainable_parameters():
node1 = ad.constant(1)
node2 = ad.variable(2)
node3 = ad.variable(3)
node4 = ad.variable(4)
node5 = ad.constant(5)
node1_node2 = ad.mul(node1, node2)
node1_node3 = ad.mul(node1, node3)
node1_node4 = ad.mul(node1, node4)
node_sum = ad.add(node1_node2,
node1_node3,
node1_node4)
final_node = ad.mul(node_sum, node5)
trainable_nodes = final_node.trainable_parameters()
assert len(trainable_nodes) == 3
for node in trainable_nodes:
assert node.value in [2, 3, 4]
def test_trainable_parameters_grads():
node1 = ad.constant(1)
node2 = ad.variable(2)
node3 = ad.variable(3)
node4 = ad.variable(4)
node5 = ad.constant(5)
node1_node2 = ad.mul(node1, node2)
node1_node3 = ad.mul(node1, node3)
node1_node4 = ad.mul(node1, node4)
node_sum = ad.add(node1_node2, node1_node3, node1_node4)
final_node = ad.mul(node_sum, node5)
trainable_nodes = final_node.trainable_parameters()
with Session() as sess:
out = sess.run(final_node)
assert out.value == 45
# test grad
trainable_node_grads = final_node.grad(trainable_nodes)
assert len(trainable_node_grads) == len(trainable_nodes)
with Session() as sess:
trainable_node_grads = sess.run(trainable_node_grads)
for node, grad in zip(trainable_nodes, trainable_node_grads):
print(grad.value, node.value)
continue
if node.name == '2':
assert grad.value == 5
elif node.name == '3':
assert grad.value == 5
elif node.name == '4':
assert grad.value == 5
else:
raise ValueError('Not a trainable node')
assert False