Running Graph
This library contains classes for launching graphs and executing operations.
The basic usage guide has
examples of how a graph is launched in a tf.Session .
Session management
class tf.Session
A class for running TensorFlow operations.
A Session object encapsulates the environment in which Operation
objects are executed, and Tensor objects are evaluated. For
example:
# Build a graph.
a = tf.constant(5.0)
b = tf.constant(6.0)
c = a * b
# Launch the graph in a session.
sess = tf.Session()
# Evaluate the tensor `c`.
print(sess.run(c))
A session may own resources, such as
variables, queues,
and readers. It is important to release
these resources when they are no longer required. To do this, either
invoke the close() method on the session, or use
the session as a context manager. The following two examples are
equivalent:
# Using the `close()` method.
sess = tf.Session()
sess.run(...)
sess.close()
# Using the context manager.
with tf.Session() as sess:
sess.run(...)
The ConfigProto
protocol buffer exposes various configuration options for a
session. For example, to create a session that uses soft constraints
for device placement, and log the resulting placement decisions,
create a session as follows:
# Launch the graph in a session that allows soft device placement and
# logs the placement decisions.
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True,
log_device_placement=True))
tf.Session.__init__(target='', graph=None, config=None)
Creates a new TensorFlow session.
If no graph argument is specified when constructing the session,
the default graph will be launched in the session. If you are
using more than one graph (created with tf.Graph() in the same
process, you will have to use different sessions for each graph,
but each graph can be used in multiple sessions. In this case, it
is often clearer to pass the graph to be launched explicitly to
the session constructor.
Args:
target : (Optional.) The execution engine to connect to.
Defaults to using an in-process engine. At present, no value
other than the empty string is supported.
graph : (Optional.) The Graph to be launched (described above).
config : (Optional.) A ConfigProto
protocol buffer with configuration options for the session.
tf.Session.run(fetches, feed_dict=None, options=None, run_metadata=None)
Runs the operations and evaluates the tensors in fetches .
This method runs one "step" of TensorFlow computation, by
running the necessary graph fragment to execute every Operation
and evaluate every Tensor in fetches , substituting the values in
feed_dict for the corresponding input values.
The fetches argument may be a list of graph elements or a single
graph element, and these determine the return value of this
method. A graph element can be one of the following types:
- If the *i*th element of
fetches is an
Operation , the *i*th
return value will be None .
- If the *i*th element of
fetches is a
Tensor , the *i*th return
value will be a numpy ndarray containing the value of that tensor.
- If the *i*th element of
fetches is a
SparseTensor ,
the *i*th return value will be a
SparseTensorValue
containing the value of that sparse tensor.
- If the *i*th element of
fetches is produced by a get_tensor_handle op,
the *i*th return value will be a numpy ndarray containing the handle of
that tensor.
The optional feed_dict argument allows the caller to override
the value of tensors in the graph. Each key in feed_dict can be
one of the following types:
- If the key is a
Tensor , the
value may be a Python scalar, string, list, or numpy ndarray
that can be converted to the same dtype as that
tensor. Additionally, if the key is a
placeholder, the shape of
the value will be checked for compatibility with the placeholder.
- If the key is a
SparseTensor ,
the value should be a
SparseTensorValue .
The optional options argument expects a [RunOptions ] proto. The options
allow controlling the behavior of this particular step (e.g. turning tracing
on).
The optional run_metadata argument expects a [RunMetadata ] proto. When
appropriate, the non-Tensor output of this step will be collected there. For
example, when users turn on tracing in options , the profiled info will be
collected into this argument and passed back.
Args:
fetches : A single graph element, or a list of graph elements
(described above).
feed_dict : A dictionary that maps graph elements to values
(described above).
options : A [RunOptions ] protocol buffer
run_metadata : A [RunMetadata ] protocol buffer
Returns:
Either a single value if fetches is a single graph element, or
a list of values if fetches is a list (described above).
Raises:
RuntimeError : If this Session is in an invalid state (e.g. has been
closed).
TypeError : If fetches or feed_dict keys are of an inappropriate type.
ValueError : If fetches or feed_dict keys are invalid or refer to a
Tensor that doesn't exist.
tf.Session.close()
Closes this session.
Calling this method frees all resources associated with the session.
Raises:
RuntimeError : If an error occurs while closing the session.
tf.Session.graph
The graph that was launched in this session.
tf.Session.as_default()
Returns a context manager that makes this object the default session.
Use with the with keyword to specify that calls to
Operation.run() or
Tensor.run() should be
executed in this session.
c = tf.constant(..)
sess = tf.Session()
with sess.as_default():
assert tf.get_default_session() is sess
print(c.eval())
To get the current default session, use
tf.get_default_session() .
N.B. The as_default context manager does not close the
session when you exit the context, and you must close the session
explicitly.
c = tf.constant(...)
sess = tf.Session()
with sess.as_default():
print(c.eval())
# ...
with sess.as_default():
print(c.eval())
sess.close()
Alternatively, you can use with tf.Session(): to create a
session that is automatically closed on exiting the context,
including when an uncaught exception is raised.
N.B. The default graph is a property of the current thread. If you
create a new thread, and wish to use the default session in that
thread, you must explicitly add a with sess.as_default(): in that
thread's function.
Returns:
A context manager using this session as the default session.
class tf.InteractiveSession
A TensorFlow Session for use in interactive contexts, such as a shell.
The only difference with a regular Session is that an InteractiveSession
installs itself as the default session on construction.
The methods Tensor.eval()
and Operation.run()
will use that session to run ops.
This is convenient in interactive shells and IPython
notebooks, as it avoids having to pass an explicit
Session object to run ops.
For example:
sess = tf.InteractiveSession()
a = tf.constant(5.0)
b = tf.constant(6.0)
c = a * b
# We can just use 'c.eval()' without passing 'sess'
print(c.eval())
sess.close()
Note that a regular session installs itself as the default session when it
is created in a with statement. The common usage in non-interactive
programs is to follow that pattern:
a = tf.constant(5.0)
b = tf.constant(6.0)
c = a * b
with tf.Session():
# We can also use 'c.eval()' here.
print(c.eval())
tf.InteractiveSession.__init__(target='', graph=None, config=None)
Creates a new interactive TensorFlow session.
If no graph argument is specified when constructing the session,
the default graph will be launched in the session. If you are
using more than one graph (created with tf.Graph() in the same
process, you will have to use different sessions for each graph,
but each graph can be used in multiple sessions. In this case, it
is often clearer to pass the graph to be launched explicitly to
the session constructor.
Args:
target : (Optional.) The execution engine to connect to.
Defaults to using an in-process engine. At present, no value
other than the empty string is supported.
graph : (Optional.) The Graph to be launched (described above).
config : (Optional) ConfigProto proto used to configure the session.
tf.InteractiveSession.close()
Closes an InteractiveSession .
tf.get_default_session()
Returns the default session for the current thread.
The returned Session will be the innermost session on which a
Session or Session.as_default() context has been entered.
NOTE: The default session is a property of the current thread. If you
create a new thread, and wish to use the default session in that
thread, you must explicitly add a with sess.as_default(): in that
thread's function.
Returns:
The default Session being used in the current thread.
Error classes
class tf.OpError
A generic error that is raised when TensorFlow execution fails.
Whenever possible, the session will raise a more specific subclass
of OpError from the tf.errors module.
tf.OpError.op
The operation that failed, if known.
N.B. If the failed op was synthesized at runtime, e.g. a Send
or Recv op, there will be no corresponding
Operation
object. In that case, this will return None , and you should
instead use the OpError.node_def to
discover information about the op.
Returns:
The Operation that failed, or None.
tf.OpError.node_def
The NodeDef proto representing the op that failed.
Other Methods
tf.OpError.__init__(node_def, op, message, error_code)
Creates a new OpError indicating that a particular op failed.
Args:
node_def : The graph_pb2.NodeDef proto representing the op that failed,
if known; otherwise None.
op : The ops.Operation that failed, if known; otherwise None.
message : The message string describing the failure.
error_code : The error_codes_pb2.Code describing the error.
tf.OpError.error_code
The integer error code that describes the error.
tf.OpError.message
The error message that describes the error.
class tf.errors.CancelledError
Raised when an operation or step is cancelled.
For example, a long-running operation (e.g.
queue.enqueue() may be
cancelled by running another operation (e.g.
queue.close(cancel_pending_enqueues=True) ,
or by closing the session.
A step that is running such a long-running operation will fail by raising
CancelledError .
tf.errors.CancelledError.__init__(node_def, op, message)
Creates a CancelledError .
class tf.errors.UnknownError
Unknown error.
An example of where this error may be returned is if a Status value
received from another address space belongs to an error-space that
is not known to this address space. Also errors raised by APIs that
do not return enough error information may be converted to this
error.
tf.errors.UnknownError.__init__(node_def, op, message, error_code=2)
Creates an UnknownError .
class tf.errors.InvalidArgumentError
Raised when an operation receives an invalid argument.
This may occur, for example, if an operation is receives an input
tensor that has an invalid value or shape. For example, the
tf.matmul() op will raise this
error if it receives an input that is not a matrix, and the
tf.reshape() op will raise
this error if the new shape does not match the number of elements in the input
tensor.
tf.errors.InvalidArgumentError.__init__(node_def, op, message)
Creates an InvalidArgumentError .
class tf.errors.DeadlineExceededError
Raised when a deadline expires before an operation could complete.
This exception is not currently used.
tf.errors.DeadlineExceededError.__init__(node_def, op, message)
Creates a DeadlineExceededError .
class tf.errors.NotFoundError
Raised when a requested entity (e.g., a file or directory) was not found.
For example, running the
tf.WholeFileReader.read()
operation could raise NotFoundError if it receives the name of a file that
does not exist.
tf.errors.NotFoundError.__init__(node_def, op, message)
Creates a NotFoundError .
class tf.errors.AlreadyExistsError
Raised when an entity that we attempted to create already exists.
For example, running an operation that saves a file
(e.g. tf.train.Saver.save() )
could potentially raise this exception if an explicit filename for an
existing file was passed.
tf.errors.AlreadyExistsError.__init__(node_def, op, message)
Creates an AlreadyExistsError .
class tf.errors.PermissionDeniedError
Raised when the caller does not have permission to run an operation.
For example, running the
tf.WholeFileReader.read()
operation could raise PermissionDeniedError if it receives the name of a
file for which the user does not have the read file permission.
tf.errors.PermissionDeniedError.__init__(node_def, op, message)
Creates a PermissionDeniedError .
class tf.errors.UnauthenticatedError
The request does not have valid authentication credentials.
This exception is not currently used.
tf.errors.UnauthenticatedError.__init__(node_def, op, message)
Creates an UnauthenticatedError .
class tf.errors.ResourceExhaustedError
Some resource has been exhausted.
For example, this error might be raised if a per-user quota is
exhausted, or perhaps the entire file system is out of space.
tf.errors.ResourceExhaustedError.__init__(node_def, op, message)
Creates a ResourceExhaustedError .
class tf.errors.FailedPreconditionError
Operation was rejected because the system is not in a state to execute it.
This exception is most commonly raised when running an operation
that reads a tf.Variable
before it has been initialized.
tf.errors.FailedPreconditionError.__init__(node_def, op, message)
Creates a FailedPreconditionError .
class tf.errors.AbortedError
The operation was aborted, typically due to a concurrent action.
For example, running a
queue.enqueue()
operation may raise AbortedError if a
queue.close() operation
previously ran.
tf.errors.AbortedError.__init__(node_def, op, message)
Creates an AbortedError .
class tf.errors.OutOfRangeError
Raised when an operation iterates past the valid input range.
This exception is raised in "end-of-file" conditions, such as when a
queue.dequeue()
operation is blocked on an empty queue, and a
queue.close()
operation executes.
tf.errors.OutOfRangeError.__init__(node_def, op, message)
Creates an OutOfRangeError .
class tf.errors.UnimplementedError
Raised when an operation has not been implemented.
Some operations may raise this error when passed otherwise-valid
arguments that it does not currently support. For example, running
the tf.nn.max_pool() operation
would raise this error if pooling was requested on the batch dimension,
because this is not yet supported.
tf.errors.UnimplementedError.__init__(node_def, op, message)
Creates an UnimplementedError .
class tf.errors.InternalError
Raised when the system experiences an internal error.
This exception is raised when some invariant expected by the runtime
has been broken. Catching this exception is not recommended.
tf.errors.InternalError.__init__(node_def, op, message)
Creates an InternalError .
class tf.errors.UnavailableError
Raised when the runtime is currently unavailable.
This exception is not currently used.
tf.errors.UnavailableError.__init__(node_def, op, message)
Creates an UnavailableError .
class tf.errors.DataLossError
Raised when unrecoverable data loss or corruption is encountered.
For example, this may be raised by running a
tf.WholeFileReader.read()
operation, if the file is truncated while it is being read.
tf.errors.DataLossError.__init__(node_def, op, message)
Creates a DataLossError .
|