Styjun

I'm trying to evaluate aX^2+bX+c, as [a,b,c]*[X*X X 1] in tensorflow.

I've tried the following code:

import tensorflow as tf
X = tf.placeholder(tf.float32, name="X")
W = tf.Variable([1,2,1], dtype=tf.float32, name="weights")
W=tf.reshape(W,[1,3])
F = tf.Variable([X*X,X,1.0], dtype=tf.float32, name="Filter")
F=tf.reshape(F,[3,1])
print(W.shape)
print(F.shape)
Y=tf.matmul(W,F)
with tf.Session() as sess:
 sess.run(tf.global_variables_initializer())
 for i in range(10): 
 sess.run(Y, feed_dict=X: i)
 Y=sess.run(Y)
print("Y:",Y)

However, initializer is not happy:

(1, 3)
(3, 1)
...
tensorflow.python.framework.errors_impl.InvalidArgumentError: You must feed a value for placeholder tensor 'X' with dtype float
 [[node X]]
During handling of the above exception, another exception occurred:
...
Caused by op 'X', defined at:
 File "sample.py", line 2, in <module>
 X = tf.placeholder(tf.float32, name="X") 
...

Any thoughts, as to possible alternatives?

You just need to modify the code a little bit. The value of tf.Variable should not be tf.placeholder, otherwise it will cause your initialization error when running sess.run(tf.global_variables_initializer()). You can use tf.stack instead of it.
In addition, please remember to feed data when you run sess.run(Y).

import tensorflow as tf

X = tf.placeholder(tf.float32, name="X")
W = tf.Variable([1,2,1], dtype=tf.float32, name="weights")
W = tf.reshape(W,[1,3])
F = tf.stack([X*X,X,1.0])
F = tf.reshape(F,[3,1])
Y = tf.matmul(W,F)
with tf.Session() as sess:
 sess.run(tf.global_variables_initializer())
 for i in range(10):
 Y_val = sess.run(Y, feed_dict=X: i)
 print("Y:",Y_val)

Y: [[1.]]
Y: [[4.]]
Y: [[9.]]
Y: [[16.]]
Y: [[25.]]
Y: [[36.]]
Y: [[49.]]
Y: [[64.]]
Y: [[81.]]
Y: [[100.]]

I think even though you could still initialize a variable that depends on a placeholder like this, W will get initialized repeatedly unless you add more code to initialize only uninitialized variables. That is more effort.

Hope I haven't missed other inefficiencies in this approach.

import tensorflow as tf

sess = tf.InteractiveSession()

X = tf.placeholder(tf.float32, name="X")

W = tf.Variable([1, 2, 1], dtype=tf.float32, name="weights")
W = tf.reshape(W, [1, 3])

var = tf.reshape([X*X,X,1],[3,1])
F = tf.get_variable('F', dtype=tf.float32, initializer=var)

init = tf.global_variables_initializer()
Y=tf.matmul(W,F)

for i in range(10):
 sess.run([init], feed_dict=X: i)
 print(sess.run(Y))


[[1.]]
[[4.]]
[[9.]]
[[16.]]
[[25.]]
[[36.]]
[[49.]]
[[64.]]
[[81.]]
[[100.]]