CODEWITHSUNDEEP
tensorflowkeras
Sugh
Sugh

Reputation: 11

'tensorflow.python.keras.api._v1.keras.losses' has no attribute 'Reduction'

I am using Huber loss implementation in tf.keras in tensorflow 1.14.0 as follows:

huber_keras_loss = tf.keras.losses.Huber(
        delta=delta,
        reduction=tf.keras.losses.Reduction.SUM,
        name='huber_loss'
    )

I am getting the error AttributeError: module 'tensorflow.python.keras.api._v1.keras.losses' has no attribute 'Reduction'

I have tried using tf.losses.Reduction, tf.compat.v2.losses.Reduction nothing seems to work.

Did tensorflow remove Reduction from tf.keras.losses, it is strange if they did so because their documentation still shows: https://www.tensorflow.org/versions/r1.14/api_docs/python/tf/keras/losses/Huber#args

Upvotes: 1

Views: 6466

Answers (3)

OliverM98
OliverM98

Reputation: 21

When I get an error similar to that, applying tf.compat.v1. usually works.

This is the specific function where I applied it:

tf.compat.v1.losses.sparse_softmax_cross_entropy(labels=labels, logits=output)

I also applied it to similar functions and it worked.

Upvotes: 0

Markus
Markus

Reputation: 36

Replacing

tf.keras.losses.Reduction

with

tf.compat.v1.losses.Reduction

solved the issue for me in TensorFlow 1.14.0

Upvotes: 1

Colton Neary
Colton Neary

Reputation: 90

I ran into a similar problem when I installed the tensorflow 1.14.0. What I did to get this to work was to go to upgrade the tensorflow which seemed to revert me to 1.12.0.

    pip install --upgrade tensorflow

Upvotes: 0

Related Questions