CODEWITHSUNDEEP
tensorflow
Konstantin Solomatov
Konstantin Solomatov

Reputation: 10332

Shapes in tf.test.compute_gradient_error

In addition to 2 tensors, tf.test.compute_gradient_error takes two shapes. Why are they required? Why can't we just evaluate the tensors and calculate their shapes?

Upvotes: 0

Views: 67

Answers (1)

Allen Lavoie
Allen Lavoie

Reputation: 5808

It was written very early in TensorFlow's development, at a point where shape information was not associated with Tensors. Since it's not useful in model code, updating it for usability was never a priority.

Upvotes: 1

Related Questions