I'm new to learning neural nets and trying to recreate stuff in order to get an understanding of how things work. I'd like to replicate a pointer network to train on the travelling salesman problem.
While I'm able to understand most of what the paper by Vinyals et al 2015. The point of confusion is the use of the special starting token (=>)
I understand that for text you could create such a token. But how does it translate to co-ordinate data?
Would be really useful if someone could expand on this or point to a resource.