Charles Weill

Try upgrading to `adanet=0.5.0` and training from scratch with a new `model_dir`. Let me know if that works.

Thanks for the heads up. We'll update.

@wangzhishi: You can get the **weights** w from TensorBoard, under the `mixture_weight_norms` section. As far as extracting the individual subnetworks, that is currently not easily doable possible, since we only...

> Is there a way to use fixed weights like (0.1, 0.2, 0.7) without learning them? Thanks! Not directly, however one option would be to create the `ComplexityRegularizedEnsembler(optimizer=...)` with a...

Have you tried [`tf.estimator.BinaryClassHead()`](https://www.tensorflow.org/api_docs/python/tf/estimator/BinaryClassHead)? I believe it uses the same loss under the hood, specifically `sigmoid_cross_entropy`.

I agree that it would be a better UX to add some examples of how this works. I'll need to think about this a bit. We need to at least...

It was introduced in https://github.com/cweill/gotests/pull/90, for users who want to call gotests from a separate binary.

You're right, it looks like a bug where I forgot to pass it to the Options in #90. Would you like to send a PR to fix it?

You should contribute to the discussion at https://github.com/golang/go/issues/33688. We should get consensus from the Go community about how to best generate tests like these.

Hi @mindlesslcc, this is an impressive PR. You'll have to give me until this weekend to review this, as I'm swamped this week. In the meantime, perhaps you can help...