Ryan Nett
Ryan Nett
Wouldn't that type need to be set at the class level, rather than either of the inputs? I'm thinking of something like the cosine similarity of int vectors: predictions and...
Because presumably they will all be cast to the result type, i.e. `TFloat32` for my example, not any of the input types (i.e. `TInt32`). It's not so much flexibility, it's...
There's a kind of similar issue for Initializer, where if I understand the intent right, you should easily enough be able to do something like: ```java Ones ones = new...
(I'm writing the framework Kotlin wrappers, which is why this is coming up).
Hmm, that's true. We're still silently casting `labels` though, so we may want to standardize that.
For my wrappers, I'm trying to write something like: ```kotlin inline fun Loss(block: KotlinOps.(Operand) -> Operand): ? ``` which would define a loss for a single datatype (`T`). This isn't...
Yeah, the more I think about it the less I like the solution I proposed, you would still be able to do things in the Kotlin wrapper like `Loss{ ......
Yeah, can you? That's the info I found in the [maven docs](https://maven.apache.org/plugins/maven-compiler-plugin/multirelease.html), see the [Single Project](http://in.relation.to/2017/02/13/building-multi-release-jars-with-maven/) link.
> Now for merging this now or later... this change is now quite subtle, I'm fine merging it now. Were you planning to add cleaner code for the 0.4.0 release?...
Since 8 doesn't support `-release`, and since we're using 11 I can test the base jar with 9.