Description
Thank you for creating a C++ version of Pytorch. However, I wonder if you could create an example that looks like C++ and not like Python?
The DCGAN sample project makes extensive use of auto so that it can show how it can be made to look and feel like Python by avoiding standard C++ things like unique_ptr<>, shared_ptr<> etc.
However, I am a C++ programmer, not a Python programmer. I am very happy working with standard C++ things like classes with methods and smart pointers. The noble attempt to make "feel like Python" with auto variables isn't helpful for me. For example, it assumes that I will be able to put my entire program into a single method. That's an unfortunate restriction, as I want to build, store and pass objects between a number of different methods.
I have tried unwrapping the auto using some decltype() statements, but the Pytorch C++ templating makes this quite laborious. Perhaps that is an unavoidable result of the way that the underlying library is built? If so, could you create an C++ example that shows how to unwrap the various templates in one case, splitting the operations across several methods of a class for me?
Would that be straightforward to do? It would be a great help for me to get an idea of how your templating structure works and I can then build up from that.
I've only just started working with the library (that's why I'm looking at the example), so maybe I've missed something in the tutorial? I apologize if that's the case and ask if you would point me at the example that I should be looking at?
Many thanks,
Dan