def test_local_variable(self):
with self.cached_session() as sess:
self.assertEquals([], variables_lib.local_variables())
value0 = 42
variables_lib2.local_variable(value0)
value1 = 43
variables_lib2.local_variable(value1)
variables = variables_lib.local_variables()
self.assertEquals(2, len(variables))
self.assertRaises(errors_impl.OpError, sess.run, variables)
variables_lib.variables_initializer(variables).run()
self.assertAllEqual(set([value0, value1]),
set(sess.run(variables)))
def testEvaluateWithEvalFeedDict(self):
# Create a checkpoint.
checkpoint_dir = os.path.join(self.get_temp_dir(),
'evaluate_with_eval_feed_dict')
self._train_model(checkpoint_dir, num_steps=1)
# We need a variable that the saver will try to restore.
variables.get_or_create_global_step()
# Create a variable and an eval op that increments it with a placeholder.
my_var = variables.local_variable(0.0, name='my_var')
increment = array_ops.placeholder(dtype=dtypes.float32)
eval_ops = state_ops.assign_add(my_var, increment)
increment_value = 3
num_evals = 5
expected_value = increment_value * num_evals
final_values = evaluation.evaluate_repeatedly(
checkpoint_dir=checkpoint_dir,
eval_ops=eval_ops,
feed_dict={increment: 3},
final_ops={'my_var': array_ops.identity(my_var)},
hooks=[
evaluation.StopAfterNEvalsHook(num_evals),
],
max_number_of_evaluations=1)
self.assertEqual(final_values['my_var'], expected_value)
def testEvalOpAndFinalOp(self):
checkpoint_dir = os.path.join(self.get_temp_dir(),
'eval_ops_and_final_ops')
# Train a model for a single step to get a checkpoint.
self._train_model(checkpoint_dir, num_steps=1)
checkpoint_path = evaluation.wait_for_new_checkpoint(checkpoint_dir)
# Create the model so we have something to restore.
inputs = constant_op.constant(self._inputs, dtype=dtypes.float32)
logistic_classifier(inputs)
num_evals = 5
final_increment = 9.0
my_var = variables.local_variable(0.0, name='MyVar')
eval_ops = state_ops.assign_add(my_var, 1.0)
final_ops = array_ops.identity(my_var) + final_increment
final_ops_values = evaluation.evaluate_once(
checkpoint_path=checkpoint_path,
eval_ops=eval_ops,
final_ops={'value': final_ops},
hooks=[
evaluation.StopAfterNEvalsHook(num_evals),
])
self.assertEqual(final_ops_values['value'],
num_evals + final_increment)
def testTrainWithLocalVariable(self):
with ops.Graph().as_default():
random_seed.set_random_seed(0)
tf_inputs = constant_op.constant(self._inputs,
dtype=dtypes.float32)
tf_labels = constant_op.constant(self._labels,
dtype=dtypes.float32)
local_multiplier = variables_lib.local_variable(1.0)
tf_predictions = logistic_classifier(tf_inputs) * local_multiplier
losses.log_loss(tf_labels, tf_predictions)
total_loss = losses.get_total_loss()
optimizer = gradient_descent.GradientDescentOptimizer(
learning_rate=1.0)
train_op = training.create_train_op(total_loss, optimizer)
loss = training.train(
train_op,
None,
hooks=[basic_session_run_hooks.StopAtStepHook(num_steps=300)],
save_summaries_steps=None,
save_checkpoint_secs=None)
self.assertIsNotNone(loss)
self.assertLess(loss, .015)
def _build_inference_graph(self):
"""Build simple inference graph.
This includes a regular variable, local variable, and fake table.
Returns:
Tuple of 3 `Tensor` objects, 2 input and 1 output.
"""
variables_lib.create_global_step()
in0 = variables.VariableV1(1.0)
in1 = variables_lib.local_variable(2.0)
fake_table = variables.VariableV1(3.0,
trainable=False,
collections=['fake_tables'],
name='fake_table_var')
in0.graph.add_to_collections([ops.GraphKeys.TABLE_INITIALIZERS],
fake_table.initializer)
out = in0 + in1 + fake_table
return in0, in1, out
def testOnlyFinalOp(self):
checkpoint_dir = os.path.join(self.get_temp_dir(), 'only_final_ops')
# Train a model for a single step to get a checkpoint.
self._train_model(checkpoint_dir, num_steps=1)
checkpoint_path = evaluation.wait_for_new_checkpoint(checkpoint_dir)
# Create the model so we have something to restore.
inputs = constant_op.constant(self._inputs, dtype=dtypes.float32)
logistic_classifier(inputs)
final_increment = 9.0
my_var = variables.local_variable(0.0, name='MyVar')
final_ops = array_ops.identity(my_var) + final_increment
final_ops_values = evaluation.evaluate_once(
checkpoint_path=checkpoint_path, final_ops={'value': final_ops})
self.assertEqual(final_ops_values['value'], final_increment)
def test_train_loss(self):
with ops.Graph().as_default() as g, self.session(g):
variables_lib.create_global_step()
loss_var = variables_lib.local_variable(10.0)
train_op = control_flow_ops.group(
state_ops.assign_add(variables_lib.get_global_step(), 1),
state_ops.assign_add(loss_var, -1.0))
self._assert_summaries(self._output_dir)
self._assert_ckpt(self._output_dir, False)
loss = learn.graph_actions.train(g,
output_dir=self._output_dir,
train_op=train_op,
loss_op=loss_var.value(),
steps=6)
# TODO(ebrevdo,ptucker,ispir): this meta_graph_def lacks the
# SaverDef, so we can't add it to the summary assertion test below.
# meta_graph_def = meta_graph.create_meta_graph_def()
self.assertEqual(4.0, loss)
self._assert_summaries(self._output_dir, expected_graphs=[g])
self._assert_ckpt(self._output_dir, True)
def testTrainWithLocalVariable(self):
logdir = os.path.join(
tempfile.mkdtemp(prefix=self.get_temp_dir()), 'tmp_logs')
with ops.Graph().as_default():
random_seed.set_random_seed(0)
tf_inputs = constant_op.constant(self._inputs, dtype=dtypes.float32)
tf_labels = constant_op.constant(self._labels, dtype=dtypes.float32)
local_multiplier = variables_lib2.local_variable(1.0)
tf_predictions = LogisticClassifier(tf_inputs) * local_multiplier
loss_ops.log_loss(tf_predictions, tf_labels)
total_loss = loss_ops.get_total_loss()
optimizer = gradient_descent.GradientDescentOptimizer(learning_rate=1.0)
train_op = learning.create_train_op(total_loss, optimizer)
loss = learning.train(
train_op, logdir, number_of_steps=300, log_every_n_steps=10)
self.assertIsNotNone(loss)
self.assertLess(loss, .015)
