def testConstructNAdamWithLR(self):
opt = nadam.Nadam(lr=1.0)
opt_2 = nadam.Nadam(learning_rate=0.1, lr=1.0)
opt_3 = nadam.Nadam(learning_rate=0.1)
self.assertIsInstance(opt.lr, tf.Variable)
self.assertIsInstance(opt_2.lr, tf.Variable)
self.assertIsInstance(opt_3.lr, tf.Variable)
self.evaluate(tf.compat.v1.global_variables_initializer())
self.assertAllClose(self.evaluate(opt.lr), (1.0))
self.assertAllClose(self.evaluate(opt_2.lr), (1.0))
self.assertAllClose(self.evaluate(opt_3.lr), (0.1))
def testSparse(self):
# TODO(tanzheny, omalleyt): Fix test in eager mode.
sparse_epsilon = 1e-7
for dtype in [tf.half, tf.float32, tf.float64]:
with tf.Graph().as_default(), self.cached_session():
# Initialize variables for numpy implementation.
m0, v0, m1, v1, mcache = 0.0, 0.0, 0.0, 0.0, 1.0
var0_np = np.array([1.0, 1.0, 2.0], dtype=dtype.as_numpy_dtype)
grads0_np = np.array([0.1, 0, 0.1], dtype=dtype.as_numpy_dtype)
var1_np = np.array([3.0, 3.0, 4.0], dtype=dtype.as_numpy_dtype)
grads1_np = np.array([0.01, 0, 0.01], dtype=dtype.as_numpy_dtype)
var0 = tf.Variable(var0_np)
var1 = tf.Variable(var1_np)
grads0_np_indices = np.array([0, 2], dtype=np.int32)
grads0 = tf.IndexedSlices(
tf.constant(grads0_np[grads0_np_indices]),
tf.constant(grads0_np_indices), tf.constant([3]))
grads1_np_indices = np.array([0, 2], dtype=np.int32)
grads1 = tf.IndexedSlices(
tf.constant(grads1_np[grads1_np_indices]),
tf.constant(grads1_np_indices), tf.constant([3]))
opt = nadam.Nadam(epsilon=sparse_epsilon)
update = opt.apply_gradients(zip([grads0, grads1], [var0, var1]))
self.evaluate(tf.compat.v1.global_variables_initializer())
# Fetch params to validate initial values
self.assertAllClose([1.0, 1.0, 2.0], var0)
self.assertAllClose([3.0, 3.0, 4.0], var1)
beta1_power, beta2_power = get_beta_accumulators(opt, dtype)
# Run 3 steps of Nadam
for t in range(3):
self.assertAllCloseAccordingToType(0.9**(t + 1), beta1_power)
self.assertAllCloseAccordingToType(0.999**(t + 1), beta2_power)
update.run()
mcache = update_m_cache(mcache, t)
var0_np, m0, v0 = nadam_update_numpy(
var0_np, grads0_np, t, m0, v0, mcache, epsilon=sparse_epsilon)
var1_np, m1, v1 = nadam_update_numpy(
var1_np, grads1_np, t, m1, v1, mcache, epsilon=sparse_epsilon)
# Validate updated params
self.assertAllCloseAccordingToType(var0_np, var0)
self.assertAllCloseAccordingToType(var1_np, var1)
def testBasic(self):
# TODO(tanzheny, omalleyt): Fix test in eager mode.
for dtype in [tf.half, tf.float32, tf.float64]:
with tf.Graph().as_default(), self.cached_session():
# Initialize variables for numpy implementation.
m0, v0, m1, v1, mcache = 0.0, 0.0, 0.0, 0.0, 1.0
var0_np = np.array([1.0, 2.0], dtype=dtype.as_numpy_dtype)
grads0_np = np.array([0.1, 0.1], dtype=dtype.as_numpy_dtype)
var1_np = np.array([3.0, 4.0], dtype=dtype.as_numpy_dtype)
grads1_np = np.array([0.01, 0.01], dtype=dtype.as_numpy_dtype)
var0 = tf.Variable(var0_np)
var1 = tf.Variable(var1_np)
grads0 = tf.constant(grads0_np)
grads1 = tf.constant(grads1_np)
opt = nadam.Nadam()
update = opt.apply_gradients(
zip([grads0, grads1], [var0, var1]))
self.evaluate(tf.compat.v1.global_variables_initializer())
# Fetch params to validate initial values
self.assertAllClose([1.0, 2.0], var0)
self.assertAllClose([3.0, 4.0], var1)
# Run 3 steps of Nadam
for t in range(3):
update.run()
mcache = update_m_cache(mcache, t)
var0_np, m0, v0 = nadam_update_numpy(
var0_np, grads0_np, t, m0, v0, mcache)
var1_np, m1, v1 = nadam_update_numpy(
var1_np, grads1_np, t, m1, v1, mcache)
# Validate updated params
self.assertAllCloseAccordingToType(var0_np, var0)
self.assertAllCloseAccordingToType(var1_np, var1)
gradient_descent_optimizer_v1_fn, adagrad_optimizer_v1_fn,
ftrl_optimizer_v1_fn, rmsprop_optimizer_v1_fn
]
adadelta_optimizer_keras_v2_fn = tf.__internal__.test.combinations.NamedObject(
"AdadeltaKerasV2", lambda: adadelta_keras_v2.Adadelta(0.001))
adagrad_optimizer_keras_v2_fn = tf.__internal__.test.combinations.NamedObject(
"AdagradKerasV2", lambda: adagrad_keras_v2.Adagrad(0.001))
adam_optimizer_keras_v2_fn = tf.__internal__.test.combinations.NamedObject(
"AdamKerasV2", lambda: adam_keras_v2.Adam(0.001, epsilon=1.0))
adam_experimental_fn = tf.__internal__.test.combinations.NamedObject(
"AdamExperimental", lambda: adam_experimental.Adam(0.001))
adamax_optimizer_keras_v2_fn = tf.__internal__.test.combinations.NamedObject(
"AdamaxKerasV2", lambda: adamax_keras_v2.Adamax(0.001, epsilon=1.0))
nadam_optimizer_keras_v2_fn = tf.__internal__.test.combinations.NamedObject(
"NadamKerasV2", lambda: nadam_keras_v2.Nadam(0.001, epsilon=1.0))
ftrl_optimizer_keras_v2_fn = tf.__internal__.test.combinations.NamedObject(
"FtrlKerasV2", lambda: ftrl_keras_v2.Ftrl(0.001))
gradient_descent_optimizer_keras_v2_fn = tf.__internal__.test.combinations.NamedObject(
"GradientDescentKerasV2", lambda: gradient_descent_keras_v2.SGD(0.001))
rmsprop_optimizer_keras_v2_fn = tf.__internal__.test.combinations.NamedObject(
"RmsPropKerasV2", lambda: rmsprop_keras_v2.RMSprop(0.001))
# TODO(shiningsun): consider adding the other v2 optimizers
optimizers_v2 = [
gradient_descent_optimizer_keras_v2_fn, adagrad_optimizer_keras_v2_fn
]
optimizers_v1_and_v2 = optimizers_v1 + optimizers_v2
def testConstructNAdamWithScheduleDecay(self):
opt = nadam.Nadam(schedule_decay=0.2)
self.assertIsInstance(opt.decay, tf.Variable)
self.evaluate(tf.compat.v1.global_variables_initializer())
self.assertAllClose(self.evaluate(opt.decay), (0.2))
def testNadamCompatibility(self): opt_v1 = optimizer_v1.Nadam(lr=0.001) opt_v2 = nadam.Nadam(learning_rate=0.001) self._testOptimizersCompatibility(opt_v1, opt_v2)