tensorflow - Difference between tf.assign and assignment operator (=) -

i'm trying understand difference between tf.assign , assignment operator(=). have 3 sets of code

first, using simple tf.assign

import tensorflow tf  tf.graph().as_default():   = tf.variable(1, name="a")   assign_op = tf.assign(a, tf.add(a,1))   tf.session() sess:     sess.run(tf.global_variables_initializer())     print sess.run(assign_op)     print a.eval()     print a.eval() 

the output expected as

2 2 2 

second, using assignment operator

import tensorflow tf  tf.graph().as_default():   = tf.variable(1, name="a")   = + 1   tf.session() sess:    sess.run(tf.global_variables_initializer())    print sess.run(a)    print a.eval()    print a.eval() 

the results still 2, 2, 2.

third, use both tf.assign , assignment operator

import tensorflow tf  tf.graph().as_default():   = tf.variable(1, name="a")   = tf.assign(a, tf.add(a,1))   tf.session() sess:     sess.run(tf.global_variables_initializer())     print sess.run(a)     print a.eval()     print a.eval() 

now, output becomes 2, 3, 4.

my questions are

1. in 2nd snippet using (=), when have sess.run(a), seems i'm running assign op. "a = a+1" internally create assignment op assign_op = tf.assign(a, a+1)? op run session assign_op? when run a.eval(), doesn't continue increment a, hence appears eval evaluating "static" variable.

2. i'm not sure how explain 3rd snippet. why 2 evals increment a, 2 evals in 2nd snippet doesn't?

thanks.

the main confusion here doing a = + 1 reassign python variable a resulting tensor of addition operation a + 1. tf.assign, on other hand, operation setting value of tensorflow variable.

a = tf.variable(1, name="a") = + 1 

this equivalent to:

a = tf.add(tf.variable(1, name="a"), 1) 

with in mind:

in 2nd snippet using (=), when have sess.run(a), seems i'm running assign op. "a = a+1" internally create assignment op assign_op = tf.assign(a, a+1)? [...]

it might so, not true. explained above, reassign python variable. , without tf.assign or other operation changes variable, stays value 1. each time a evaluated, program calculate a + 1 => 1 + 1.

i'm not sure how explain 3rd snippet. why 2 evals increment a, 2 evals in 2nd snippet doesn't?

that's because calling eval() on assignment tensor in third snippet triggers variable assignment (note isn't different doing session.run(a) current session).