Pfthb

I have a collection of Variables in a TensorFlow graph which I want to update simultaneously all according to a similar rule. An example graph would be

a = tf.placeholder(tf.int32)
x1 = tf.Variable(0, tf.int32)
x2 = tf.Variable(1, tf.int32)

I want to then read in data to a one value at a time, and at each step update x1 and x2 to max(current_value, a). This can be accomplished by adding two assign operations

u1 = x1.assign(tf.maximum(a, x1))
u2 = x2.assign(tf.maximum(a, x2))

and if the input data is in a list data, and the collection of variables vars, this can be accomplished by the loop (apologies for the ugly logic!)

with tf.Session() as sess:
 for d in data:
 if vars = :
 continue
 if vars = ['x1']:
 sess.run(u1, a:d)
 if vars = ['x2']:
 sess.run(u2, a:d)
 if vars = ['x1', 'x2']:
 sess.run([u1, u2], a:d)

However, if I have a large number of xi and want to avoid repetitive code
building by hand each of the updating ui, is there a way to build a function which takes as arguments a list of variables and generates new assignment variables just for them?