Given a few symbolic variables to fetch, I need to know which placeholders are dependency.
In Theano, we have:
import theano as th
import theano.tensor as T
x, y, z = T.scalars('xyz')
u, v = x*y, y*z
w = u + v
th.gof.graph.inputs([w]) # gives [x, y, z]
th.gof.graph.inputs([u]) # gives [x, y]
th.gof.graph.inputs([v]) # gives [y, z]
th.gof.graph.inputs([u, v]) # gives [x, y, z]
How to do the same thing in TensorFlow?