The PCTSP related beam search issue

[bywbilly]

Hi, Thanks for open sourcing this excellent code! I am playing with the beam search decoder. However, I found that the beam search does not work correctly for PCTSP (including stochastic variants). The greedy decoder can give me around 3.19 on PCTSP-20, however, beam search with width 1 (this should be the same as greedy) gives me 7.44. I am wondering if there is a bug in the code or I did something wrong?

Thanks!

[wouterkool]

Hi! I am sorry it's not working for you as expected. As I am quite busy, can you maybe try to investigate the issue? Please use a batch size of 1 and put some breakpoints to compare the probabilities generated by the model both with greedy and beam search with size 1.

Also, als a general comment, the stochastic PCTSP is not intended for use with beam search. Beam search (or sampling multiple solutions) considers many solutions at once, which, realistically, is not possible with the stochastic PCTSP as you cannot 'try' many solutions since you'll only get to see the actual prize once you (physically) move to a certain node, which is only possible for a single node.