Is this repository deprecated?

[erichulburd]

Has this work been subsumed by other projects? It seems that AutoKeras is actively maintained and I'm curious why AdaNet was never supported there: https://github.com/keras-team/autokeras/issues/415. Other alternative AutoDL libraries that seem well maintained are Auto-PyTorch and AutoGluon

Perhaps AdaNet evolved into proprietary algorithms and software that are only available through Google Cloud AutoML?

For reference, I'm interested in AdaNet more from the client side (ie I'm evaluating AutoDL options either through a proprietary vendor or open source software) rather than a researcher looking to push this particular field forward.

[scottyak]

Hi Eric,

Thank you for the question. While the principles of the AdaNet algorithm still stand, the implementation of the AdaNet library is deeply tied to tf.Estimator. With the transition to TF2 and keras.Models, "outer-loop" libraries (such as AutoKeras) have made "outer-loop" algorithms (such as AdaNet) easier to implement, and there is less of a need for a dedicated "one-stop" library for searching, training, and ensembling DNNs. As such, we have decided not to port AdaNet over to Keras.

However, we do have an alternative solution for architecture search + distributed training in the cloud that is more future-proof (i.e., supports Keras): https://github.com/google/model_search

On Thu, Sep 29, 2022 at 2:31 PM Eric Hulburd @.***> wrote:

Has this work been subsumed by other projects? It seems that AutoKeras is actively maintained and I'm curious why AdaNet was never supported there: keras-team/autokeras#415 https://github.com/keras-team/autokeras/issues/415. Another alternative AutoDL library that seems well maintained is Auto-PyTorch https://github.com/automl/Auto-PyTorch.

Perhaps AdaNet evolved into proprietary algorithms and software that are only available through Google Cloud AutoML?

For reference, I'm interested in AdaNet more from the client side (ie I'm evaluating AutoDL options either through a proprietary vendor or open source software) rather than a researcher looking to push the field forward.

— Reply to this email directly, view it on GitHub https://github.com/tensorflow/adanet/issues/163, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABN6VYLSZWTKAESGVGMGANLWAXOADANCNFSM6AAAAAAQZBFXB4 . You are receiving this because you are subscribed to this thread.Message ID: @.***>