包装 Eigen::Vector 类型

Question

I'm working on a Neural Network library and I want to use Eigen to handle the vector and matrix math behind the models but I've been told that I should't expose Eigen types in the final api so, since the network is basically going to get vectors as inputs and return vectors as outputs I was thinking about creating a class(maybe Tensor) which wraps Eigen::VectorXf but I don't really know how to this preserving the Eigen optimizations when operating on such Tensors.

Is there a way to do this or should I change my approach?

For example the OpenNN library also uses Eigen(I think) and its Tensor type is inherited from an std::vector, I haven't found how it is converted to Eigen yet (maybe Map is involved?)

Answer 1

The only reason I see to write a thin wrapper around the Eigen types is if you need the flexibility to change the implementation at a later point.

If you don't have this requirement, you should expose the Eigen types. It is an API your users will be familiar with and you will avoid any unnecessary conversions (copies) between your type and the Eigen Tensors.