class SingleLossStepTest(test.TestCase, parameterized.TestCase):
@combinations.generate(
combinations.times(
strategy_combinations.distributions_and_v1_optimizers(),
combinations.combine(
mode=strategy_combinations.graph_and_eager_modes),
combinations.combine(is_tpu=[False])) + combinations.combine(
distribution=[strategy_combinations.tpu_strategy],
optimizer_fn=strategy_combinations.optimizers_v1,
mode=["graph"],
is_tpu=[True]))
def testTrainNetwork(self, distribution, optimizer_fn, is_tpu):
with distribution.scope():
single_loss_step, layer = single_loss_example(
optimizer_fn, distribution, use_bias=True, iterations_per_step=2)
if context.executing_eagerly():
single_loss_step.initialize()
run_step = single_loss_step
else:
with self.cached_session() as sess:
sess.run(single_loss_step.initialize())
run_step = sess.make_callable(single_loss_step())
self.evaluate(variables.global_variables_initializer())
weights, biases = [], []
for _ in range(5):
run_step()
weights.append(self.evaluate(layer.kernel))
biases.append(self.evaluate(layer.bias))
error = abs(numpy.add(numpy.squeeze(weights), numpy.squeeze(biases)) - 1)
is_not_increasing = all(y <= x for x, y in zip(error, error[1:]))
self.assertTrue(is_not_increasing)
class MonitorTest(test.TestCase, parameterized.TestCase):
@combinations.generate(
combinations.times(
strategy_combinations.distributions_and_v1_optimizers(),
combinations.combine(
mode=strategy_combinations.graph_and_eager_modes)))
def testTrainNetwork(self, distribution, optimizer_fn):
with distribution.scope():
single_loss_step, layer = single_loss_example(optimizer_fn, distribution)
if context.executing_eagerly():
monitor = monitor_lib.Monitor(single_loss_step, None)
else:
with self.cached_session() as sess:
monitor = monitor_lib.Monitor(single_loss_step, sess)
monitor.run_steps(1)
self.assertEqual(1, len(layer.trainable_variables))
mirrored_weight_variable = layer.trainable_variables[0]
start_error = self.evaluate(mirrored_weight_variable)
start_error = abs(numpy.array(start_error) - 1)
monitor.run_steps(9)
end_error = self.evaluate(mirrored_weight_variable)
end_error = abs(numpy.array(end_error) - 1)
self.assertGreaterEqual(start_error, end_error)
def testPassingASessionInEager(self):
distribution = one_device_strategy.OneDeviceStrategy(
"/device:CPU:0")
step_function, _ = single_loss_example(
lambda: gradient_descent.GradientDescentOptimizer(0.2), distribution)
with session.Session() as sess, context.eager_mode():
with self.assertRaisesRegexp(ValueError, "Should not provide"):
_ = monitor_lib.Monitor(step_function, sess)
def testNotPassingASessionInGraph(self):
distribution = one_device_strategy.OneDeviceStrategy(
"/device:CPU:0")
step_function, _ = single_loss_example(
lambda: gradient_descent.GradientDescentOptimizer(0.2), distribution)
with context.graph_mode(), ops.Graph().as_default():
with self.assertRaisesRegexp(ValueError, "Should provide"):
_ = monitor_lib.Monitor(step_function, session=None)
class MinimizeLossStepTest(test.TestCase, parameterized.TestCase):
def _get_iterator(self, strategy, input_fn):
iterator = strategy.make_input_fn_iterator(lambda _: input_fn())
self.evaluate(iterator.initializer)
return iterator
@combinations.generate(
combinations.times(
strategy_combinations.distributions_and_v1_optimizers(),
combinations.combine(mode=["graph"],
use_callable_loss=[True, False]) +
combinations.combine(mode=["eager"], use_callable_loss=[True])) +
combinations.times(
strategy_combinations.distributions_and_v2_optimizers(),
combinations.combine(mode=["graph", "eager"],
use_callable_loss=[True])) +
combinations.combine(distribution=[strategy_combinations.tpu_strategy],
optimizer_fn=strategy_combinations.optimizers_v2,
mode=["graph"],
use_callable_loss=[True]) +
combinations.combine(distribution=[strategy_combinations.tpu_strategy],
optimizer_fn=strategy_combinations.optimizers_v1,
mode=["graph"],
use_callable_loss=[True, False]))
def testTrainNetwork(self, distribution, optimizer_fn, use_callable_loss):
with distribution.scope():
optimizer = optimizer_fn()
model_fn, dataset_fn, layer = minimize_loss_example(
optimizer, use_bias=True, use_callable_loss=use_callable_loss)
def step_fn(ctx, inputs):
del ctx # Unused
return distribution.group(
distribution.extended.call_for_each_replica(
model_fn, args=(inputs, )))
iterator = self._get_iterator(distribution, dataset_fn)
def run_step():
return distribution.extended.experimental_run_steps_on_iterator(
step_fn, iterator, iterations=2).run_op
if not context.executing_eagerly():
with self.cached_session() as sess:
run_step = sess.make_callable(run_step())
self.evaluate(variables_lib.global_variables_initializer())
weights, biases = [], []
for _ in range(5):
run_step()
weights.append(self.evaluate(layer.kernel))
biases.append(self.evaluate(layer.bias))
error = abs(
numpy.add(numpy.squeeze(weights), numpy.squeeze(biases)) - 1)
is_not_increasing = all(y <= x for x, y in zip(error, error[1:]))
self.assertTrue(is_not_increasing)
@combinations.generate(
combinations.times(
strategy_combinations.distributions_and_v1_optimizers(),
combinations.combine(mode=["graph"],
use_callable_loss=[True, False]) +
combinations.combine(mode=["eager"], use_callable_loss=[True])) +
combinations.times(
strategy_combinations.distributions_and_v2_optimizers(),
combinations.combine(mode=["graph", "eager"],
use_callable_loss=[True])))
def testTrainNetworkByCallForEachReplica(self, distribution, optimizer_fn,
use_callable_loss):
with distribution.scope():
optimizer = optimizer_fn()
model_fn, dataset_fn, layer = minimize_loss_example(
optimizer, use_bias=True, use_callable_loss=use_callable_loss)
iterator = self._get_iterator(distribution, dataset_fn)
def run_step():
return distribution.group(
distribution.extended.call_for_each_replica(
model_fn, args=(iterator.get_next(), )))
if not context.executing_eagerly():
with self.cached_session() as sess:
run_step = sess.make_callable(run_step())
self.evaluate(variables_lib.global_variables_initializer())
weights, biases = [], []
for _ in range(10):
run_step()
weights.append(self.evaluate(layer.kernel))
biases.append(self.evaluate(layer.bias))
error = abs(
numpy.add(numpy.squeeze(weights), numpy.squeeze(biases)) - 1)
is_not_increasing = all(y <= x for x, y in zip(error, error[1:]))
self.assertTrue(is_not_increasing)
@combinations.generate(
combinations.times(
strategy_combinations.distributions_and_v1_and_v2_optimizers(),
combinations.combine(mode=["graph", "eager"])) +
combinations.combine(
distribution=[strategy_combinations.tpu_strategy],
optimizer_fn=strategy_combinations.optimizers_v1_and_v2,
mode=["graph"]))
def testOptimizerInsideModelFn(self, distribution, optimizer_fn):
if (not context.executing_eagerly()
and control_flow_v2_toggles.control_flow_v2_enabled()):
self.skipTest("b/138751864")
created_variables = []
trainable_variables = []
def appending_creator(next_creator, **kwargs):
v = next_creator(**kwargs)
created_variables.append(v.name)
if "trainable" in kwargs and kwargs["trainable"]:
trainable_variables.append(v.name)
return v
# Creator scope needs to be set before it's used inside
# `distribution.scope`.
with variable_scope.variable_creator_scope(
appending_creator), distribution.scope():
optimizer = optimizer_fn()
model_fn, dataset_fn, _ = minimize_loss_example(
optimizer, use_bias=True, use_callable_loss=True)
def step_fn(ctx, inputs):
del ctx # Unused
return distribution.group(
distribution.extended.call_for_each_replica(
model_fn, args=(inputs, )))
iterator = self._get_iterator(distribution, dataset_fn)
def run_step():
return distribution.extended.experimental_run_steps_on_iterator(
step_fn, iterator, iterations=1).run_op
if not context.executing_eagerly():
with self.cached_session() as sess:
run_step = sess.make_callable(run_step())
self.evaluate(variables_lib.global_variables_initializer())
run_step()
def get_expected_variables(num_parameter_devices):
name = optimizer._name
if isinstance(optimizer, optimizer_v2.OptimizerV2):
variables = VAR_MAP_V2[name]
else:
variables = VAR_MAP_V1[name]
extended_variables = [
v + "/replica_{}".format(replica) for v in variables
for replica in range(1, num_parameter_devices)
]
variables = list(variables) + extended_variables
return set(v + ":0" for v in variables)
self.assertEqual(
get_expected_variables(
len(distribution.extended.parameter_devices)),
set(created_variables))
@combinations.generate(
combinations.times(
combinations.combine(momentum=[0.8, 0.9, 0.99],
renorm=[False, True]),
combinations.times(
strategy_combinations.distributions_and_v1_and_v2_optimizers(),
combinations.combine(
mode=["graph", "eager"],
# TODO(isaprykin): Allow False here. Currently subsequent
# replicas will re-execute UPDATE_OPS of previous replicas.
update_ops_in_cross_replica_mode=[True])) +
combinations.combine(
distribution=[strategy_combinations.tpu_strategy],
optimizer_fn=strategy_combinations.optimizers_v1_and_v2,
mode=["graph"],
update_ops_in_cross_replica_mode=[False])))
def testTrainNetworkWithBatchNorm(self, distribution, optimizer_fn,
momentum, renorm,
update_ops_in_cross_replica_mode):
"""Verifies that moving mean updates are reduced across replicas."""
with distribution.scope():
num_replicas = distribution.num_replicas_in_sync
model_fn, dataset_fn, batchnorm = batchnorm_example(
optimizer_fn,
batch_per_epoch=num_replicas,
momentum=momentum,
renorm=renorm,
update_ops_in_replica_mode=not update_ops_in_cross_replica_mode
)
def step_fn(ctx, inputs):
del ctx # Unused
fetches = distribution.experimental_local_results(
distribution.extended.call_for_each_replica(
model_fn, args=(inputs, )))
if update_ops_in_cross_replica_mode:
fetches += tuple(
ops.get_collection(ops.GraphKeys.UPDATE_OPS))
return control_flow_ops.group(fetches)
iterator = self._get_iterator(distribution, dataset_fn)
def run_step():
return distribution.extended.experimental_run_steps_on_iterator(
step_fn, iterator, iterations=1).run_op
if not context.executing_eagerly():
with self.cached_session() as sess:
run_step = sess.make_callable(run_step())
self.evaluate(variables_lib.global_variables_initializer())
expected_moving_means = [0.] * 8
def averaged_batch_mean(i):
# Each batch has shape [16, 8] where the ith element in jth list is
# (8 * j + i + replica_id * 100). So the batch mean in each replica is
# (60 + i + replica_id * 100). So here comes its batch mean over all
# replicas:
return 60. + i + (num_replicas - 1.) / 2. * 100.
for _ in range(10):
run_step()
moving_means = self.evaluate(batchnorm.moving_mean)
# We make sure that the moving_mean is updated as if the sample mean is
# calculated over all replicas.
for i, expected_moving_mean in enumerate(
expected_moving_means):
expected_moving_means[i] -= (
(expected_moving_mean - averaged_batch_mean(i)) *
(1.0 - momentum))
self.assertNear(expected_moving_means[i], moving_means[i],
0.0001)
@combinations.generate(
combinations.times(
combinations.combine(loss_reduction=[
losses_impl.Reduction.SUM, losses_impl.Reduction.MEAN,
losses_impl.Reduction.SUM_OVER_BATCH_SIZE,
losses_impl.Reduction.SUM_OVER_NONZERO_WEIGHTS
]),
combinations.times(
combinations.combine(distribution=[
strategy_combinations.one_device_strategy,
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
strategy_combinations.mirrored_strategy_with_two_gpus
]),
combinations.times(
combinations.combine(optimizer_fn=strategy_combinations.
gradient_descent_optimizer_v1_fn),
combinations.combine(mode=["graph"],
use_callable_loss=[True, False]) +
combinations.combine(mode=["eager"],
use_callable_loss=[True])) +
combinations.times(
combinations.combine(
optimizer_fn=strategy_combinations.
gradient_descent_optimizer_keras_v2_fn),
combinations.combine(mode=["graph", "eager"],
use_callable_loss=[True]))) +
combinations.combine(
distribution=[strategy_combinations.tpu_strategy],
optimizer_fn=strategy_combinations.
gradient_descent_optimizer_v1_fn,
mode=["graph"],
use_callable_loss=[True, False]) + combinations.combine(
distribution=[strategy_combinations.tpu_strategy],
optimizer_fn=strategy_combinations.
gradient_descent_optimizer_keras_v2_fn,
mode=["graph"],
use_callable_loss=[True])))
def testMeanVsSum(self, distribution, optimizer_fn, loss_reduction,
use_callable_loss):
with distribution.scope():
all_vars = []
def model_fn(inputs):
x, y = inputs
w = variable_scope.get_variable("w", initializer=[[2.]])
all_vars.append(w)
def loss_fn():
# Use fixed initialization to make the steps deterministic.
predict = math_ops.matmul(x, w)
loss = losses_impl.mean_squared_error(
y, predict, reduction=loss_reduction)
if loss_reduction == losses_impl.Reduction.SUM:
return loss
return loss / distribution.num_replicas_in_sync
optimizer = optimizer_fn(
) # GradientDescent with 0.2 learning rate
if isinstance(optimizer, optimizer_v2.OptimizerV2):
return optimizer.minimize(loss_fn, [w])
else:
if use_callable_loss:
return optimizer.minimize(loss_fn)
else:
return optimizer.minimize(loss_fn())
def dataset_fn():
features = dataset_ops.Dataset.from_tensors([[2.], [7.]])
labels = dataset_ops.Dataset.from_tensors([[6.], [21.]])
return dataset_ops.Dataset.zip((features, labels)).repeat()
def step_fn(ctx, inputs):
del ctx # Unused
return distribution.group(
distribution.extended.call_for_each_replica(
model_fn, args=(inputs, )))
iterator = self._get_iterator(distribution, dataset_fn)
def run_step():
return distribution.extended.experimental_run_steps_on_iterator(
step_fn, iterator, iterations=1).run_op
if not context.executing_eagerly():
with self.cached_session() as sess:
run_step = sess.make_callable(run_step())
self.evaluate(variables_lib.global_variables_initializer())
run_step()
v = all_vars[0]
self.assertTrue(all(v is vi for vi in all_vars[1:]))
weight = numpy.squeeze(self.evaluate(v))
# Our model is:
# predict = x * w
# loss = (predict - y)^2
# dloss/dpredict = 2*(predict - y)
# dloss/dw = 2 * x^T @ (predict - y)
# For our batch size of 2, assuming sum loss reduction:
# x = [2, 7]
# y = [6, 21]
# w_initial = 2
# predict = [4, 14]
# predict - y = [-2, -7]
# dloss/dw = 2 <[2, 7], [-2, -7]> = - 2(4 + 49) = -106
# So unreplicated the update to w with lr=0.001 is -0.2 * -106 = 0.106
# with sum loss reduction, or 0.053 with mean.
if loss_reduction == losses_impl.Reduction.SUM:
# Note that the "distribution.num_replicas_in_sync" factor will go away
# once we split the input across replicas, instead of pulling a complete
# batch of input per replica.
self.assertNear(weight,
2 + 0.106 * distribution.num_replicas_in_sync,
0.0001)
else:
# One of the mean loss reductions.
self.assertNear(weight, 2 + 0.053, 0.0001)
@combinations.generate(
combinations.times(
strategy_combinations.distributions_and_v1_and_v2_optimizers(),
combinations.combine(mode=["graph", "eager"]),
combinations.combine(is_tpu=[False])) + combinations.combine(
distribution=[strategy_combinations.tpu_strategy],
optimizer_fn=strategy_combinations.optimizers_v1_and_v2,
mode=["graph"],
is_tpu=[True]))
def testRunStepsWithOutputContext(self, distribution, optimizer_fn,
is_tpu):
with distribution.scope():
def dataset_fn():
dataset = dataset_ops.Dataset.from_tensors([[1.]]).repeat()
# TODO(priyag): batch with drop_remainder=True causes shapes to be
# fully defined for TPU. Remove this when XLA supports dynamic shapes.
return dataset.batch(batch_size=1, drop_remainder=True)
optimizer = optimizer_fn()
layer = core.Dense(1, use_bias=True)
key1 = "foo"
value1 = "bar"
def model_fn(output_context, x):
"""A very simple model written by the user."""
def loss_fn():
y = array_ops.reshape(layer(x),
[]) - constant_op.constant(1.)
return y * y
if isinstance(optimizer, optimizer_v2.OptimizerV2):
train_op = optimizer.minimize(
loss_fn, lambda: layer.trainable_variables)
else:
train_op = optimizer.minimize(loss_fn)
loss = loss_fn()
output_context.set_last_step_output(
name="replica_loss_reduced",
output=loss,
reduce_op=reduce_util.ReduceOp.MEAN)
output_context.set_non_tensor_output(key1, value1)
return (train_op, loss)
def step_fn(output_context, inputs):
(train_op, loss) = distribution.extended.call_for_each_replica(
model_fn, args=(output_context, inputs))
output_context.set_last_step_output(
name="cross_replica_loss_reduced",
output=loss,
reduce_op=reduce_util.ReduceOp.MEAN)
output_context.set_last_step_output(
name="cross_replica_loss_not_reduced", output=loss)
return distribution.group(train_op)
iterator = self._get_iterator(distribution, dataset_fn)
def run_step():
initial_loss = lambda: constant_op.constant(1e7)
# Initial values corresponding to reduced losses are just single
# tensors. But for non reduced losses, we need to have initial
# values that are of the same structure as non reduced losses. In
# MirroredStrategy, this will be a list of losses, in TPUStrategy
# it will be single tensor. Using `call_for_each_replica` followed
# by `experimental_local_results` gives us the desired initial
# value structure.
not_reduced = distribution.experimental_local_results(
distribution.extended.call_for_each_replica(initial_loss))
initial_loop_values = {
"replica_loss_reduced": initial_loss(),
"cross_replica_loss_reduced": initial_loss(),
"cross_replica_loss_not_reduced": not_reduced,
}
ctx = distribution.extended.experimental_run_steps_on_iterator(
step_fn,
iterator,
iterations=2,
initial_loop_values=initial_loop_values)
self.assertEqual({key1: (value1, )}, ctx.non_tensor_outputs)
self._verify_loss_output(
initial_loss(),
loss_output=ctx.last_step_outputs["replica_loss_reduced"],
reduced=True,
distribution=distribution)
self._verify_loss_output(
initial_loss(),
loss_output=ctx.
last_step_outputs["cross_replica_loss_reduced"],
reduced=True,
distribution=distribution)
self._verify_loss_output(
initial_loss(),
loss_output=ctx.
last_step_outputs["cross_replica_loss_not_reduced"],
reduced=False,
distribution=distribution)
return (ctx.run_op,
ctx.last_step_outputs["replica_loss_reduced"])
if not context.executing_eagerly():
with self.cached_session() as sess:
run_step = sess.make_callable(run_step())
self.evaluate(variables_lib.global_variables_initializer())
weights, biases, losses = [], [], []
for _ in range(5):
_, loss = run_step()
losses.append(loss)
weights.append(self.evaluate(layer.kernel))
biases.append(self.evaluate(layer.bias))
loss_is_not_increasing = all(y <= x
for x, y in zip(losses, losses[1:]))
self.assertTrue(loss_is_not_increasing)
error = abs(
numpy.add(numpy.squeeze(weights), numpy.squeeze(biases)) - 1)
error_is_not_increasing = all(y <= x
for x, y in zip(error, error[1:]))
self.assertTrue(error_is_not_increasing)
def _verify_loss_output(self, initial_loss, loss_output, reduced,
distribution):
if not reduced:
self.assertLen(
distribution.experimental_local_results(loss_output),
distribution.num_replicas_in_sync)
loss_tensor = distribution.reduce(reduce_util.ReduceOp.MEAN,
loss_output,
axis=None)
else:
unwrapped_output = distribution.experimental_local_results(
loss_output)
self.assertLen(unwrapped_output, 1)
loss_tensor = unwrapped_output[0]
self.assertEqual(initial_loss.dtype, loss_tensor.dtype)
self.assertEqual(initial_loss.shape, loss_tensor.shape)
@combinations.generate(
strategy_combinations.distributions_and_v2_optimizers())
def test_empty_var_list(self, distribution, optimizer_fn):
opt = optimizer_fn()
with distribution.scope():
def run_fn():
opt.minimize(lambda: constant_op.constant(1.), [])
opt.apply_gradients([])
distribution.run(run_fn)
