Charles Weill

Interesting. Are you using `tf.keras.layers` or `tf.layers`? I think I ran into this issue with `tf.keras.layers` because they handle variable scoping differently.

Thank you, for confirming @Npccc. We'll look into this.

@nicholasbreckwoldt: We just released `adanet=0.9.0` which includes better TPU, and TF 2 support. Please try installing it, and let us know if it resolves your issue.

Thank you for your kind words. We would like to make AdaNet as flexible as possible to support different real-world tasks. Could you please provide some examples of object detection...

@GRIGORR That is correct: all the candidate subnetworks (and their associated ensemble) are trained in parallel in the same TensorFlow graph. At the end of each iteration, the best subnetwork...

@shendiaomo: You are correct on both counts. For this reason, we request that the user configures the `max_iteration_steps` to be the number of repetitions desired, which unfortunately requires the user...

@le-dawg I think you make a good point: we need to update our tutorials to use the `tf.ensemble.Ensemblers`.

> According to the doc and original paper this would imply that the tutorials actually do not form ensembles. Is that the case? They still create ensembles, it just that...

Could you please share the log outputs with the failure?

@JodyZXD: Good observation. The GIF just demos a simple example. The AdaNet framework supports a superset of the subnetworks and connections that the paper defines. You can recreate the neural...