Hello,
I am trying to port a Python project that is using tf.estimator.DNNLinearCombinedClassifier. Unfortunately, I can’t seem to find the modern implementation of this. I understand that there should be or once was tf.compat.v1.keras.experimental.WideDeepModel which I could migrate to, but I can’t find that on the current TF docs page nor by exploring the tensorflow package. Does anyone here know how I can get this DNNLinearCombinedClassifier or equivalent to work with the current version of TF?
Any help would be appreciated.
Hi @Govind_Salinas, The estimator package was available till tensorflow 2.15 after 2.15 release it has been removed. To continue using tf.estimator, you will need to use TF 2.15 or an earlier version. From 2.16 you have built the model using the keras API. Thank You.
Hi, thanks for responding. I understand i need to use the keras package but I can’t find the appropriate model/classifier to use. I found a changelog entry from 2.7.0 that said I should start using WideDeepModel instead, but I can’t find that anywhere in the keras package or anywhere else in TF. Do you know where this model exists or if there is a different model I should be using?
Another quick note, I also require numpy >= 2.0 so I can’t drop back to TF 2.15 which requires numpy < 2.0. So unfortunately that is not an option for me.