Is it possible to visualize a tensorflow graph without a training op?

Question

I know how to visualize a tensorflow graph after training with tensorboard. Now, is it possible to visualize just the forward part of the graph, ie, with no training operator defined?

The reason I'm asking this is that I'm getting this error:

No gradients provided for any variable, check your graph for ops that do not support gradients, between variables [ ... list of model variables here ... ] and loss Tensor("Mean:0", dtype=float32).

I'd like to inspect the graph to find out where the gradient tensor flow (pun intended) is broken.

Answer 1

Yes, you can visualize any graph. Try this simple script:

import tensorflow as tf

a = tf.add(1, 2, name="Add_these_numbers")
b = tf.multiply(a, 3)
c = tf.add(4, 5, name="And_These_ones")
d = tf.multiply(c, 6, name="Multiply_these_numbers")
e = tf.multiply(4, 5, name="B_add")
f = tf.div(c, 6, name="B_mul")
g = tf.add(b, d)
h = tf.multiply(g, f)

with tf.Session() as sess:
    writer = tf.summary.FileWriter("output", sess.graph)
    print(sess.run(h))
    writer.close()

Then run...

tensorboard --logdir=output

... and you'll see:

So you can simply create a session just to write the graph to the FileWriter and not do anything else.