'Tensor' object has no attribute 'ndim' about tf-keras-vis

[keisen]

Hi, @YIsonP . Thank you for your reporting.

I tried to reproduce your problem, I didn't get the error.
Could you please the code snippet to reproduce it?
And please let me know the versions of tf-keras-vis, Tensorflow and Python.

Thanks!

[YIsonP]

Hi @keisen. Thanks for reply. here is my env:
tf-keras-vis:0.8.2
tensorflow:2.10.0
python: 3.9.12
and this is the link for the snippet :https://github.com/YIsonP/saliency-map
a simple bg: I use pre-trained model in Keras and try to visualize its decision by saliency map.

[keisen]

Hi, @YIsonP , Thank you for sharing the notebook.

The cause of the error is tf.compat.v1.disable_eager_execution() at 2nd cell in your notebook.
tf-keras-vis only support TF2.0+, but not TF1.x.
So if you want to visualize with the model built on TF1.x, I recommend you to use keras-vis.
If not, please use tf-keras-vis without tf.compat.v1.disable_eager_execution() .

Thanks!

[YIsonP]

Hi, @YIsonP , Thank you for sharing the notebook.

The cause of the error is tf.compat.v1.disable_eager_execution() at 2nd cell in your notebook. tf-keras-vis only support TF2.0+, but not TF1.x. So if you want to visualize with the model built on TF1.x, I recommend you to use keras-vis. If not, please use tf-keras-vis without tf.compat.v1.disable_eager_execution() .

Thanks!

Hi @keisen
It's been super helpful. Thank you so much for the help!!

[keisen]

Anytime!