Why not using `.fit()` method to train TF2 model?

Questions
#1

Hey all!
I am following your example here in my own codebase to train a GCN. My question is, since GCN is a child of tf.keras.Model, why can’t we use .compile() and .fit() to do training? Does it relate to the performance drop mentioned here? Thanks

#2

I tried calling .fit() instead of using tf.GradientTape() and I got this error related to DGLDataset object. Any help? Thanks

Screenshot 2021-10-27 at 14.03.43
Screenshot 2021-10-27 at 14.03.431896×498 200 KB

#3

DGL has some customized operator which cannot be scripted by Tensorflow. Thus only eager mode is supported for now

#5

I might be missing something, but isn’t .fit() eager mode?

#7

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.