def testName(self):
with ops.name_scope("scope"):
queue = data_flow_ops.FIFOQueue(10, dtypes.float32, name="queue")
qr = queue_runner_impl.QueueRunner(queue, [control_flow_ops.no_op()])
self.assertEqual("scope/queue", qr.name)
queue_runner_impl.add_queue_runner(qr)
self.assertEqual(
1, len(ops.get_collection(ops.GraphKeys.QUEUE_RUNNERS, "scope")))
def boston_input_fn_with_queue(num_epochs=None):
features, labels = boston_input_fn(num_epochs=num_epochs)
# Create a minimal queue runner.
fake_queue = data_flow_ops.FIFOQueue(30, dtypes.int32)
queue_runner = queue_runner_impl.QueueRunner(fake_queue,
[constant_op.constant(0)])
queue_runner_impl.add_queue_runner(queue_runner)
return features, labels
def boston_input_fn_with_queue(num_epochs=None):
features, labels = boston_input_fn(num_epochs=num_epochs)
# Create a minimal queue runner.
fake_queue = data_flow_ops.FIFOQueue(30, dtypes.int32)
queue_runner = queue_runner_impl.QueueRunner(fake_queue,
[constant_op.constant(0)])
queue_runner_impl.add_queue_runner(queue_runner)
return features, labels
def testStartQueueRunnersRaisesIfNotASession(self):
zero64 = constant_op.constant(0, dtype=dtypes.int64)
var = variables.VariableV1(zero64)
count_up_to = var.count_up_to(3)
queue = data_flow_ops.FIFOQueue(10, dtypes.float32)
init_op = variables.global_variables_initializer()
qr = queue_runner_impl.QueueRunner(queue, [count_up_to])
queue_runner_impl.add_queue_runner(qr)
with self.cached_session():
init_op.run()
with self.assertRaisesRegexp(TypeError, "tf.Session"):
queue_runner_impl.start_queue_runners("NotASession")
def testStartQueueRunnersIgnoresMonitoredSession(self):
zero64 = constant_op.constant(0, dtype=dtypes.int64)
var = variables.VariableV1(zero64)
count_up_to = var.count_up_to(3)
queue = data_flow_ops.FIFOQueue(10, dtypes.float32)
init_op = variables.global_variables_initializer()
qr = queue_runner_impl.QueueRunner(queue, [count_up_to])
queue_runner_impl.add_queue_runner(qr)
with self.cached_session():
init_op.run()
threads = queue_runner_impl.start_queue_runners(
monitored_session.MonitoredSession())
self.assertFalse(threads)
def _testScopedExportWithQueue(self, test_dir, exported_filename):
graph = ops.Graph()
with graph.as_default():
with ops.name_scope("queue1"):
input_queue = data_flow_ops.FIFOQueue(10, dtypes.float32)
enqueue = input_queue.enqueue((9876), name="enqueue")
close = input_queue.close(name="close")
qr = queue_runner_impl.QueueRunner(input_queue, [enqueue], close)
queue_runner_impl.add_queue_runner(qr)
input_queue.dequeue(name="dequeue")
orig_meta_graph, _ = meta_graph.export_scoped_meta_graph(
filename=os.path.join(test_dir, exported_filename),
graph=ops.get_default_graph(),
export_scope="queue1")
return orig_meta_graph
def testStartQueueRunners(self):
# CountUpTo will raise OUT_OF_RANGE when it reaches the count.
zero64 = constant_op.constant(0, dtype=dtypes.int64)
var = variables.VariableV1(zero64)
count_up_to = var.count_up_to(3)
queue = data_flow_ops.FIFOQueue(10, dtypes.float32)
init_op = variables.global_variables_initializer()
qr = queue_runner_impl.QueueRunner(queue, [count_up_to])
queue_runner_impl.add_queue_runner(qr)
with self.cached_session() as sess:
init_op.run()
threads = queue_runner_impl.start_queue_runners(sess)
for t in threads:
t.join()
self.assertEqual(0, len(qr.exceptions_raised))
# The variable should be 3.
self.assertEqual(3, var.eval())
def _testScopedExportWithQueue(self, test_dir, exported_filename):
graph = ops.Graph()
with graph.as_default():
with ops.name_scope("queue1"):
input_queue = data_flow_ops.FIFOQueue(10, dtypes.float32)
enqueue = input_queue.enqueue((9876), name="enqueue")
close = input_queue.close(name="close")
qr = queue_runner_impl.QueueRunner(input_queue, [enqueue], close)
queue_runner_impl.add_queue_runner(qr)
input_queue.dequeue(name="dequeue")
orig_meta_graph, _ = meta_graph.export_scoped_meta_graph(
filename=os.path.join(test_dir, exported_filename),
graph=ops.get_default_graph(),
export_scope="queue1")
return orig_meta_graph
def testMultiThreadedEstimateDataDistribution(self):
num_classes = 10
# Set up graph.
random_seed.set_random_seed(1234)
label = math_ops.cast(
math_ops.round(random_ops.random_uniform([1]) * num_classes),
dtypes_lib.int32)
prob_estimate = sampling_ops._estimate_data_distribution( # pylint: disable=protected-access
label, num_classes)
# Check that prob_estimate is well-behaved in a multithreaded context.
_, _, [prob_estimate] = sampling_ops._verify_input( # pylint: disable=protected-access
[], label, [prob_estimate])
# Use queues to run multiple threads over the graph, each of which
# fetches `prob_estimate`.
queue = data_flow_ops.FIFOQueue(
capacity=25,
dtypes=[prob_estimate.dtype],
shapes=[prob_estimate.get_shape()])
enqueue_op = queue.enqueue([prob_estimate])
queue_runner_impl.add_queue_runner(
queue_runner_impl.QueueRunner(queue, [enqueue_op] * 25))
out_tensor = queue.dequeue()
# Run the multi-threaded session.
with self.cached_session() as sess:
# Need to initialize variables that keep running total of classes seen.
variables.global_variables_initializer().run()
coord = coordinator.Coordinator()
threads = queue_runner_impl.start_queue_runners(coord=coord)
for _ in range(25):
sess.run([out_tensor])
coord.request_stop()
coord.join(threads)
def testMultiThreadedEstimateDataDistribution(self):
num_classes = 10
# Set up graph.
random_seed.set_random_seed(1234)
label = math_ops.cast(
math_ops.round(random_ops.random_uniform([1]) * num_classes),
dtypes_lib.int32)
prob_estimate = sampling_ops._estimate_data_distribution( # pylint: disable=protected-access
label, num_classes)
# Check that prob_estimate is well-behaved in a multithreaded context.
_, _, [prob_estimate] = sampling_ops._verify_input( # pylint: disable=protected-access
[], label, [prob_estimate])
# Use queues to run multiple threads over the graph, each of which
# fetches `prob_estimate`.
queue = data_flow_ops.FIFOQueue(capacity=25,
dtypes=[prob_estimate.dtype],
shapes=[prob_estimate.get_shape()])
enqueue_op = queue.enqueue([prob_estimate])
queue_runner_impl.add_queue_runner(
queue_runner_impl.QueueRunner(queue, [enqueue_op] * 25))
out_tensor = queue.dequeue()
# Run the multi-threaded session.
with self.cached_session() as sess:
# Need to initialize variables that keep running total of classes seen.
variables.global_variables_initializer().run()
coord = coordinator.Coordinator()
threads = queue_runner_impl.start_queue_runners(coord=coord)
for _ in range(25):
sess.run([out_tensor])
coord.request_stop()
coord.join(threads)
def model_fn(self, features, labels, mode, params):
# forward
wide_fea_cols = params['feature_columns']['wide_feas']
fm_and_dnn_fea_cols = params['feature_columns']['fm_and_dnn_feas']
wide_net = tf.feature_column.input_layer(features, wide_fea_cols)
fm_and_dnn_net = tf.feature_column.input_layer(features,
fm_and_dnn_fea_cols)
y_pred = tf.Variable(0.0, name='y_pred')
y_pred += self.linear(wide_net)
y_pred += self.dnn(fm_and_dnn_net)
if self.model == 'deepfm':
y_pred += self.fm(fm_and_dnn_net)
y_pred = tf.sigmoid(y_pred)
# predict
if mode == tf.estimator.ModeKeys.PREDICT:
predictions = {'probabilities': y_pred, 'logits': y_pred}
return tf.estimator.EstimatorSpec(mode, predictions=predictions)
loss = tf.losses.log_loss(labels=labels, predictions=y_pred)
# eval
auc_ = tf.metrics.auc(labels=labels, predictions=y_pred, name='auc_op')
metrics = {'auc': auc_}
tf.summary.scalar('auc', auc_[1])
if mode == tf.estimator.ModeKeys.EVAL:
return tf.estimator.EstimatorSpec(mode,
loss=loss,
eval_metric_ops=metrics)
# train
if params['optimizer'] == 'Adam':
optimizer = tf.train.AdamOptimizer(learning_rate=params['lr'],
beta1=0.9,
beta2=0.999,
epsilon=1e-8)
elif params['optimizer'] == 'Adagrad':
optimizer = tf.train.AdagradOptimizer(
learning_rate=params['lr'], initial_accumulator_value=1e-8)
elif params['optimizer'] == 'Momentum':
optimizer = tf.train.MomentumOptimizer(learning_rate=params['lr'],
momentum=0.95)
elif params['optimizer'] == 'ftrl':
optimizer = tf.train.FtrlOptimizer(params['lr'])
else:
optimizer = tf.train.GradientDescentOptimizer(params['lr'])
# SyncReplicasOptimizer for distribution
optimizer = tf.train.SyncReplicasOptimizer(
optimizer,
replicas_to_aggregate=self.worker_num,
total_num_replicas=self.worker_num,
use_locking=False)
print("hook_sync_replicas is set")
self.hook_sync_replicas = optimizer.make_session_run_hook(
is_chief=self.is_chief, num_tokens=0)
train_op = optimizer.minimize(loss,
global_step=tf.train.get_global_step())
chief_queue_runner = optimizer.get_chief_queue_runner()
queue_runner_impl.add_queue_runner(chief_queue_runner)
self.sync_init_op = optimizer.get_init_tokens_op()
return tf.estimator.EstimatorSpec(
mode,
loss=loss,
train_op=train_op,
training_hooks=[self.hook_sync_replicas])