The two functions fn1
and fn2
can return multiple tensors, but they have to return the exact same number and types of outputs.
x = tf.constant(1.)
bool = tf.constant(True)
def fn1():
return tf.add(x, 1.), x
def fn2():
return tf.add(x, 10.), x
res1, res2 = tf.cond(bool, fn1, fn2)
# tf.cond returns a list of two tensors
# sess.run([res1, res2]) will return [2., 1.]