chaitjo / learning-paradigms-for-tsp Goto Github PK

out results reveal - our results reveal
Just in case it's helpful!

Hi!^_^
It's a good job and I'm so interested in your researches!

When I read the paper 'On Learning Paradigms for the Travelling Salesman Problem',I met some problems confusing me.

In the experiment, you train the supervised model on the fixed size like TSP20, then evaluate the generalization ability by testing the pre-trained model on variable size like TSP50 or TSP100.

However, the shape of input feature tensor will vary according to the problem size. For example, the shape of adjacency matrix for TSP20 may be 20 x 20, while it may change into 50 x 50 for TSP50.

So I wonder that how do you deal with this problem so that you can test supervised model on different sizes of TSP?

Maybe I miss some key points when I read your paper, and sorry for disturbing you.
It will be so kind of you if you can solve my problems!

class BatchBeam(NamedTuple):

RuntimeError: class not set defining 'BatchBeam' as <class 'utils.beam_search.BatchBeam'>. Was classcell propagated to type.new?