def testMinimalRun(self):
x = basic.TrainableVariable(shape=(),
initializers={'w':
tf.ones_initializer()})()
x2 = x**2.0
min_value = 0.5
constr = optimization_constraints.OptimizationConstraints().add(
x > min_value)
self.assertFalse(constr._is_connected)
loss = moving_average.MovingAverage()(x2 + tf.random.normal(
(), stddev=1.0)) + constr()
self.assertTrue(constr._is_connected)
with self.assertRaisesRegexp(ValueError,
'Cannot add further constraints'):
constr.add(x > min_value)
with self.assertRaisesRegexp(ValueError,
'Cannot add further constraints'):
constr.add_geq(x, min_value)
with self.assertRaisesRegexp(ValueError,
'Cannot add further constraints'):
constr.add_leq(min_value < x)
opt = tf.train.AdamOptimizer(1e-2, beta1=0.0)
update = opt.minimize(loss)
with tf.control_dependencies([update]):
x2 = tf.identity(x2)
with tf.train.MonitoredSession() as sess:
for _ in range(500):
v, _ = sess.run([x2, update])
self.assertAllClose(v, min_value**2)
def testFirst(self):
var = tf.Variable(10.0)
avg = moving_average.MovingAverage(decay=0.9)(var)
with tf.train.MonitoredSession() as sess:
avg_value = sess.run(avg)
# The avg should be equal to the var after only one iteration
self.assertEqual(avg_value, 10.0)
def testIsDifferentiable(self):
x = tf.get_variable(name='x', shape=())
mva = moving_average.MovingAverage(decay=0.99, local=False)
y = mva(x)
dydx = tf.gradients(y, x)
z = mva(2 * x)
dzdx = tf.gradients(z, x)
with tf.train.MonitoredSession() as sess:
df = sess.run([dydx, dzdx])
self.assertEqual(df[0], [1.0])
self.assertEqual(df[1], [2.0])
def testReset(self):
val = tf.placeholder(shape=(), dtype=tf.float32)
module = moving_average.MovingAverage(decay=0.9)
avg = module(val)
reset = module.reset()
with tf.train.MonitoredSession() as sess:
avg_value = sess.run(avg, feed_dict={val: 10.0})
# The avg should be equal to the var after only one iteration
self.assertEqual(avg_value, 10.0)
sess.run(reset)
avg_value = sess.run(avg, feed_dict={val: 100.0})
# The avg should be equal to the var after only one iteration, again
self.assertEqual(avg_value, 100.0)
def testAverage(self, use_resource_vars):
decay = 0.9
num_steps = 10
init_value = 3.14
with tf.variable_scope('', use_resource=use_resource_vars):
var = tf.get_variable(
'var', (), initializer=tf.constant_initializer(init_value))
avg = moving_average.MovingAverage(decay=decay)(tf.identity(var))
with tf.control_dependencies([avg]):
increment = tf.assign_add(var, 1.0)
with tf.train.MonitoredSession() as sess:
expected_value = init_value
x = init_value
for _ in range(num_steps):
avg_value, _ = sess.run([avg, increment])
self.assertNear(avg_value, expected_value, 1e-4)
x += 1
expected_value = expected_value * decay + x * (1 - decay)
def testAssertDecayIsValid(self):
with self.assertRaisesRegexp(ValueError, 'Decay must be'):
moving_average.MovingAverage(decay=2.0)
