使用张量计算错误值的梯度下降

Question

I am implementing simple gradient descent algorithm using tensors. It learns two parameters m and c .
The normal python code for it is :

for i in range(epochs): 
    Y_pred = m*X + c  # The current predicted value of Y
    D_m = (-2/n) * sum(X * (Y - Y_pred))  # Derivative wrt m
    D_c = (-2/n) * sum(Y - Y_pred)  # Derivative wrt c
    m = m - L * D_m  # Update m
    c = c - L * D_c  # Update c
    print (m, c) 

output for python :

0.7424335285442664 0.014629895049575754
1.1126970531591416 0.021962519495058154
1.2973530613155333 0.025655870599552183
1.3894434413955663 0.027534253868790198
1.4353697670010162 0.028507481513901086

Tensorflow equivalent code :

#Graph of gradient descent
y_pred = m*x + c
d_m = (-2/n) * tf.reduce_sum(x*(y-y_pred)) 
d_c = (-2/n) * tf.reduce_sum(y-y_pred)  
upm = tf.assign(m, m - learning_rate * d_m)
upc = tf.assign(c, c - learning_rate * d_c)

#starting session
sess = tf.Session()

#Training for epochs
for i in range(epochs):
    sess.run(y_pred)
    sess.run(d_m)
    sess.run(d_c)
    sess.run(upm)
    sess.run(upc)
    w = sess.run(m)
    b = sess.run(c)
    print(w,b)

Output for tensorflow :

0.7424335285442664 0.007335550424492317
1.1127687194584988 0.011031122807663662
1.2974962163433057 0.012911024540805463
1.3896400798226038 0.013885244876397126
1.4356019721347115 0.014407698787092268

The parameter m has the same value for both but parameter c has different value for both although the implementation is same for both.
The output contains first 5 values of parameter m and c. The output of parameter c using tensors is approximately half of the normal python.
I don't know where my mistake is.

For recreating the entire output: Repo containing data along with both implementations

The repo also contains image of graph obtained through tensorboard in events directory

Answer 1

The problem is that, in the TF implementation, the updates are not being performed atomically. In other words, the implementation of the algorithm is updating m and c in an interleaved manner (eg the new value of m is being used when updating c ). To make the updates atomic, you should simultaneously run upm and upc :

sess.run([upm, upc])