Some minor note

[Zoran Sevarac]

Hi,

I was going through codebase and I've noticed few things so I have some questions and suggestions.

So for start in https://github.com/tensorflow/java/blob/f85623ed366d903cfddb75177725dc276f359b15/tensorflow-core/tensorflow-core-api/src/gen/java/org/tensorflow/op/nn/DepthwiseConv2dNative.java

What is the difference between methods output() and asOutput()? It seems there is none so can we have only one, or it's something that's not finished?

public Output<T> output() {

return output;

}

@Override

public Output<T> asOutput() {

return output;

}

Best,

Zoran

--

Zoran Sevarac, PhD, Associate Professor
University of Belgrade, Faculty of Organisational Sciences, Department for Software Engineering
Java Champion | Oracle Groundbreaker Ambassador | Deep Netts Co-founder & CEO

[Ryan Nett]

output() is the auto generated named accessor for one of the outputs (of course, there's only one), like Add.z.  It's based on the tensorflow api_def proto.  The main use is there's a few ops that have multiple outputs and thus don't extend Operand, like RequantizationRangePerChannel, and they need some way to access the outputs.

asOutput() converts the Operand to an Output (which is the op name and output index) so it can be used as input to another op.  It's generated (w/ the Operand superinterface) for ops with one output.

The Operand ops probably don't need the output accessors too, but since it's generated it was probably easier to just leave them in there.  Plus that way it matches the proto.

[Zoran Sevarac]

Thanks for explaining that. I'll keep investigating the points you outlined above for a possible solution.

Intuitively I would expect to have just one getOutput() method

Best