Dean Thompson

Yes, I have been a happy user of this workflow for about 6 months, and am sad to see it break now.

An excessive-memory-use-in-`Model.fit` problem that I am struggling with fits well with this one. It appears that the total size of the `Dataset` I process in `Model.fit` drives total CPU memory...

Thank you, @karllessard, for inviting me to review. I have lately been exploring different aspects of deep learning that don't leave me the time to participate in tensorflow/java for now.

and here: https://github.com/tensorflow/java/blob/5532bf0af4a908ee36b6017a8745e515bb3bb544/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/impl/LossesHelper.java#L123

I generally think of subclassing as more brittle than composition. Plus, I think of the prompt classes as having exactly the responsibilities that I want to customize, versus the Coder...

Reflecting further on this, in the longer run, I think there are two broad buckets of responsibiity: * The hands-on coding skill, represented by today's `Coder` classes. * The expert...

I liked this experiment, but I'm going to try a different approach.

I added a $100 bounty to this on [BountySource](https://www.bountysource.com/issues/48433936-backwards-compatibility-verification-tool). After doing that, I realized this functionality is nearly present in SchemaLoader -- see [this discussion group thread](https://groups.google.com/forum/#!searchin/capnproto/compatibility%7Csort:date/capnproto/dmnD1zWo7Ug/F-6dduXyAwAJ). I also see...

This sounds great, Kenton! My use case is ideally library API rather than command-line, but I imagine the normal pattern would be to implement both? And by the way, I...

I can use SchemaLoader for my purposes — when I originally joined this issue, I hadn’t found the group discussion thread explaining how to do that. But “throws an unspecified...