CODEWITHSUNDEEP
tensorflow
inc0
inc0

Reputation: 215

Overriding variable in Tensorflow

I have problem with following code:

v_background.assign(
   tf.cond(tf.less(candidate_mse, mse),
   lambda: resp,
   lambda: v_background)
 )

Basically what I want to happen is when candidate_mse is lower than old mse, v_background will be overridden with resp, otherwise it stays the same. Issue is when I call

v_background.eval()

it seems to have initial value regardless of mse.

print(mse.eval())
print(candidate_mse.eval())

Results with

0.0314396114956
0.031410553229

Upvotes: 0

Views: 66

Answers (1)

DomJack
DomJack

Reputation: 4183

This creates the assignment operation, but doesn't actually run it. If you want to do one-time assignment, just run the operation.

assign_op = v_background.assign(
   tf.cond(tf.less(candidate_mse, mse),
   lambda: resp,
   lambda: v_background)
 )
 sess.run(assign_op)

If you want a value that switches between the two depending on the values candidate_mse and mse, use tf.where.

v_background = tf.where(tf.less(candidate_mse, mse), resp, v_background)
sess.run(v_background)

Upvotes: 1

Related Questions