We Keep Coding

tf.global_variable_initializer() with regard to session?

My understandings on Sessions in Tensorflow still seem to be flawed even after reading the official documentation and this tutorial.
In particular, does tf.global_variable_initializer() initialize global variables with regard to a particular session, or for all the sessions in the program? Are there ways to "uninitialize" a variable in / during a session?
Can a tf.variable be used in multiple sessions? The answer seems to be yes (e.g. the following code), but then are there good cases where we want multiple sessions in a program, instead of a single one?
#!/usr/bin/env python
import tensorflow as tf
def main():
x = tf.constant(0.)
with tf.Session() as sess:
print(sess.run(x))
with tf.Session() as sess:
print(sess.run(x))
if __name__ == '__main__':
main()

In particular, does tf.global_variable_initializer() initialize global variables with regard to a particular session, or for all the sessions in the program?
With regards to a particular session. Check this out.
tf.reset_default_graph()
x = tf.Variable(tf.random.normal([1,5]))
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
first_sess_out = sess.run(x)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
second_sess_out = sess.run(x)
np.testing.assert_array_equal(first_sess_out, second_sess_out)
The assertion fails so it is per session.
Are there ways to "uninitialize" a variable in / during a session?
tf.reset_default_graph()
x = tf.Variable(tf.random.normal([1,5]))
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
first_init_out = sess.run(x)
sess.run(tf.global_variables_initializer())
second_init_out = sess.run(x)
np.testing.assert_array_equal(first_init_out, second_init_out)
Apparently there is, after running tf.global_variables_initializer() the variables got reinitialized. Thus, the assertion fails.
Can a tf.Variable be used in multiple sessions? The answer seems to be yes (e.g. the following code), but then are there good cases where we want multiple sessions in a program, instead of a single one?
Yes, it can be used as you can see on the examples above. Good cases are when you want to execute the graph multiple times in a single run.

Tensorflow With Context Manager vs. With Session

I'm trying to understand how a with block behaves in the above scenarios. I assume to always have one graph and one session only. I understand that I have (at least?) 2 ways to work using a session in a with block:
Example 1 : using as_default() that creates a context manager:
Same documentation says:
Use with the with keyword to specify that calls to
tf.Operation.run or tf.Tensor.eval should be executed in this
session.
# Create session
sess = tf.Session()
# Enter with block on a new context manager
with sess.as_default():
# Train: Following line should result calling tf.Operation.run
sess.run([optimizer, loss], feed_dict={x: x_train, y: y_train)
# Eval: Following line should result calling tf.Tensor.eval
sess.run([loss], feed_dict={x: x_eval, y: y_eval)
Example 2 : with block on session as stated in same documentation in a lower section:
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.
# Enter with block on session instead of Context Manager
with tf.Session() as sess:
# Train: Following line seems calling tf.Operation.run as per my test
sess.run([optimizer, loss], feed_dict={x: x_train, y: y_train)
# Eval: Following is unclear
sess.run([loss], feed_dict={x: x_eval, y: y_eval)
I would like to understand what is the correct usage as I see GitHub examples of both cases but of course without the results. In my tests both Example 1 and 2 works for training. For evaluation it seems there is a difference that I can't understand. It exceeds my knowledge to browse the Tensorflow source. Can someone please explain?

They do slightly different things, so each may or may not be correct depending on the usage. tf.Session.as_default() will just ensure that the session is set as the default ones, so calls to eval and run will use that session by default:
import tensorflow as tf
sess = tf.Session()
with sess.as_default():
print(sess is tf.get_default_session()) # True
However, as stated in the documentation, tf.Session.as_default() will not automatically close the session after the with block. If you want that, you can use the session itself as the context manager:
import tensorflow as tf
with tf.Session() as sess:
# Do something with the session
sess.run([]) # RuntimeError: Attempted to use a closed Session.
However, although (from my point of view) not clearly documented, using a session as context manager also makes it the default session.
import tensorflow as tf
with tf.Session() as sess:
print(sess is tf.get_default_session()) # True
What is the point of tf.Session.as_default(), then? Well, simply when you want to temporarily make a session the default one but you do not one to close it after that (in which case you should manually close it later, or use it as an outer context manager). This is probably most relevant when you have more than one open session. Consider the following case:
import tensorflow as tf
with tf.Session() as sess1, tf.Session() as sess2:
print(sess2 is tf.get_default_session()) # True
Here sess2 is the default because its context was added later (it can be considered to be "inner" to the context created by sess1). But now maybe you want to make sess1 the default for a while. However you can not use sess1 itself as context manager again:
import tensorflow as tf
with tf.Session() as sess1, tf.Session() as sess2:
# Do something with sess2
with sess1:
# RuntimeError: Session context managers are not re-entrant.
# Use `Session.as_default()` if you want to enter
# a session multiple times.
So you can switch between one and other default sessions with tf.Session.as_default():
import tensorflow as tf
with tf.Session() as sess1, tf.Session() as sess2:
with sess1.as_default():
# Do something with sess1
with sess2.as_default():
# Do something with sess2
# This is not really needed because sess2 was the default
# in the outer context but you can add it to be explicit
# Both sessions are closed at the end of the outer context
Of course, you can be extra explicit even with one session, if you want:
import tensorflow as tf
with tf.Session() as sess, sess.as_default():
# ...
Personally, I have never used tf.Session.as_default() in my actual code, but then again I have rarely needed to use multiple sessions, and I prefer to use tf.Session.run() instead of relying on the default session anyway, but that is mostly a matter of personal taste I suppose.

Error in tensor evaluation

The below code runs fine in a tutorial but there is an error when I run it locally. Are there any installation errors or something else? Please help. This is link to that tutorial:
https://colab.research.google.com/notebooks/mlcc/tensorflow_programming_concepts.ipynb?utm_source=mlcc&utm_campaign=colab-external&utm_medium=referral&utm_content=tfprogconcepts-colab&hl=en#scrollTo=Md8ze8e9geMi
And the code:
import tensorflow as tf
#create a graph
g = tf.Graph()
#establish the graph as the default graph
with g.as_default():
x = tf.constant(8, name = "x_const")
y = tf.constant(5, name = "y_const")
my_sum = tf.add(x,y, name = "x_y_sum")
#create the session
#this runs the default graph
with tf.Session() as sess:
print(my_sum.eval())
Below is the error that occurs:
gunjan#gunjan-Inspiron-3558:~/Desktop$ python tf1.py
/home/gunjan/anaconda3/lib/python3.5/site-
packages/h5py/__init__.py:34: FutureWarning: Conversion of the second
argument of issubdtype from `float` to `np.floating` is deprecated. In
future, it will be treated as `np.float64 == np.dtype(float).type`.
from ._conv import register_converters as _register_converters
2018-08-20 22:10:41.619062: I
tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports
instructions that this TensorFlow binary was not compiled to use: AVX2
FMA
Traceback (most recent call last):
File "tf1.py", line 15, in <module>
print(my_sum.eval())
File "/home/gunjan/anaconda3/lib/python3.5/site-
packages/tensorflow/python/framework/ops.py", line 680, in eval
return _eval_using_default_session(self, feed_dict, self.graph,
session)
File "/home/gunjan/anaconda3/lib/python3.5/site-
packages/tensorflow/python/framework/ops.py", line 4942, in
_eval_using_default_session
raise ValueError("Cannot use the default session to evaluate tensor: "
ValueError: Cannot use the default session to evaluate tensor: the
tensor's graph is different from the session's graph. Pass an explicit
session to `eval(session=sess)`.

The problem is that you've created one graph (g) and you're executing code in a separate graph (sess). If you don't need two graphs, you can just use sess:
x = tf.constant(8, name = "x_const")
y = tf.constant(5, name = "y_const")
my_sum = tf.add(x,y, name = "x_y_sum")
#create the session
#this runs the default graph
with tf.Session() as sess:
print(my_sum.eval())

To simply get it working you can pass the session explicitly, as suggested by the error message:
print(my_sum.eval(session=sess))
To understand why it doesn't work exactly as the tutorial specifies it, you could start by comparing the versions of Python and TensorFlow to those used in the tutorial.
import tensorflow as tf
import platform
print("Python version: ", platform.python_version())
print("TensorFlow version", tf.__version__)
For the colab environment you linked, this prints:
Python version: 2.7.14
TensorFlow version 1.10.0
EDIT
Taking another look at your code sample, it's not an issue of version compatibility. The issue is that your copy of the tutorial did not properly preserve the indentation. The second with block needs to be enclosed in the first.
# Establish the graph as the "default" graph.
with g.as_default():
# ...
# Now create a session.
# The session will run the default graph.
with tf.Session() as sess:
print(my_sum.eval())
This ensures that g is used as the default graph for the session, instead of creating a new one like MatthewScarpino points out your incorrectly-indented version does.

If you create/use a Graph object explicitly rather than using the default graph, you need to either (a) pass the graph object to your Session constructor, or (b) create the session in the graph context.
graph = tf.Graph()
with graph.as_default():
build_graph()
with tf.Session(graph=graph) as sess:
do_stuff_with(sess)
or
graph = tf.Graph()
with graph.as_default():
build_graph()
with tf.Session() as sess:
do_stuff_with(sess)

Tensorflow leaks 1280 bytes with each session opened and closed?

It seems that each Tensorflow session I open and close consumes 1280 bytes from the GPU memory, which are not released until the python kernel is terminated.
To reproduce, save the following python script as memory_test.py:
import tensorflow as tf
import sys
n_Iterations=int(sys.argv[1])
def open_and_close_session():
with tf.Session() as sess:
pass
for _ in range(n_Iterations):
open_and_close_session()
with tf.Session() as sess:
print("bytes used=",sess.run(tf.contrib.memory_stats.BytesInUse()))
Then run it from command line with different number of iterations:
python memory_test.py 0 yields bytes used= 1280
python memory_test.py 1 yields bytes used= 2560.
python memory_test.py 10 yields bytes used= 14080.
python memory_test.py 100 yields bytes used= 129280.
python memory_test.py 1000 yields bytes used= 1281280.
The math is easy - each session opened and closed leaks 1280 bytes. I tested this script on two different ubuntu 17.10 workstations with tensorflow-gpu 1.6 and 1.7 and different NVIDIA GPUs.
Did I miss some explicit garbage collection or is it a Tensorflow bug?
Edit: Note that unlike the case described in this question, I add nothing to the default global graph within the loop, unless the tf.Session() objects themselves 'count'. If this is the case, how can one delete them? tf.reset_default_graph() or using with tf.Graph().as_default(), tf.Session() as sess: doesn't help.

Turning my comment into an answer:
I can reproduce this behavior. I guess you should create an Issue on the GitHub-Issue-Tracker. TF uses it own Allocator-mechanism and the documentation of the session object clearly states that close()
Calling this method frees all resources associated with the session.
Which is apparently not the case here. However, even the 1281280 bytes could be potentially reused from the memory pool in a consecutive session.
So the answer is: It seems to be a bug (even in a recent '1.8.0-rc0' Version of TensorFlow.) -- either in close() or in the memory_stats Implementation.

In TensorFlow, what is the difference between Session.run() and Tensor.eval()?

TensorFlow has two ways to evaluate part of graph: Session.run on a list of variables and Tensor.eval. Is there a difference between these two?

If you have a Tensor t, calling t.eval() is equivalent to calling tf.get_default_session().run(t).
You can make a session the default as follows:
t = tf.constant(42.0)
sess = tf.Session()
with sess.as_default(): # or `with sess:` to close on exit
assert sess is tf.get_default_session()
assert t.eval() == sess.run(t)
The most important difference is that you can use sess.run() to fetch the values of many tensors in the same step:
t = tf.constant(42.0)
u = tf.constant(37.0)
tu = tf.mul(t, u)
ut = tf.mul(u, t)
with sess.as_default():
tu.eval() # runs one step
ut.eval() # runs one step
sess.run([tu, ut]) # evaluates both tensors in a single step
Note that each call to eval and run will execute the whole graph from scratch. To cache the result of a computation, assign it to a tf.Variable.

The FAQ session on tensor flow has an answer to exactly the same question. I will just go ahead and leave it here:
If t is a Tensor object, t.eval() is shorthand for sess.run(t) (where sess is the current default session. The two following snippets of code are equivalent:
sess = tf.Session()
c = tf.constant(5.0)
print sess.run(c)
c = tf.constant(5.0)
with tf.Session():
print c.eval()
In the second example, the session acts as a context manager, which has the effect of installing it as the default session for the lifetime of the with block. The context manager approach can lead to more concise code for simple use cases (like unit tests); if your code deals with multiple graphs and sessions, it may be more straightforward to explicit calls to Session.run().
I'd recommend that you at least skim throughout the whole FAQ, as it might clarify a lot of things.

eval() can not handle the list object
tf.reset_default_graph()
a = tf.Variable(0.2, name="a")
b = tf.Variable(0.3, name="b")
z = tf.constant(0.0, name="z0")
for i in range(100):
z = a * tf.cos(z + i) + z * tf.sin(b - i)
grad = tf.gradients(z, [a, b])
init = tf.global_variables_initializer()
with tf.Session() as sess:
init.run()
print("z:", z.eval())
print("grad", grad.eval())
but Session.run() can
print("grad", sess.run(grad))
correct me if I am wrong

The most important thing to remember:
The only way to get a constant, variable (any result) from TenorFlow is the session.
Knowing this everything else is easy:
Both tf.Session.run() and tf.Tensor.eval() get results from the session where tf.Tensor.eval() is a shortcut for calling tf.get_default_session().run(t)
I would also outline the method tf.Operation.run() as in here:
After the graph has been launched in a session, an Operation can be executed by passing it to tf.Session.run(). op.run() is a shortcut for calling tf.get_default_session().run(op).

Tensorflow 2.x Compatible Answer: Converting mrry's code to Tensorflow 2.x (>= 2.0) for the benefit of the community.
!pip install tensorflow==2.1
import tensorflow as tf
tf.compat.v1.disable_eager_execution()
t = tf.constant(42.0)
sess = tf.compat.v1.Session()
with sess.as_default(): # or `with sess:` to close on exit
assert sess is tf.compat.v1.get_default_session()
assert t.eval() == sess.run(t)
#The most important difference is that you can use sess.run() to fetch the values of many tensors in the same step:
t = tf.constant(42.0)
u = tf.constant(37.0)
tu = tf.multiply(t, u)
ut = tf.multiply(u, t)
with sess.as_default():
tu.eval() # runs one step
ut.eval() # runs one step
sess.run([tu, ut]) # evaluates both tensors in a single step

In tensorflow you create graphs and pass values to that graph. Graph does all the hardwork and generate the output based on the configuration that you have made in the graph.
Now When you pass values to the graph then first you need to create a tensorflow session.
tf.Session()
Once session is initialized then you are supposed to use that session because all the variables and settings are now part of the session. So, there are two ways to pass external values to the graph so that graph accepts them. One is to call the .run() while you are using the session being executed.
Other way which is basically a shortcut to this is to use .eval(). I said shortcut because the full form of .eval() is
tf.get_default_session().run(values)
You can check that yourself.
At the place of values.eval() run tf.get_default_session().run(values). You must get the same behavior.
what eval is doing is using the default session and then executing run().