Model generates only invalid solutions when trained without complete solution

[yingycl]

Hi,

Thank you for sharing the codebase. I am trying to reproduce the results by following the parameter settings described in the article. However, the trained model on the combinatorial auction dataset without complete solution only generates invalid solutions--either the outputs violate the constraints, or the solutions are all zeros (which are meaningless).

Could you please provide a more detailed training procedure for reproducing the results in this setting? Are there any additional tricks, specific parameter choices, or implementation details necessary to obtain valied results?

Thanks!

[Dddsheggy]

Hi,

Thank you for sharing the codebase. I am trying to reproduce the results by following the parameter settings described in the article. However, the trained model on the combinatorial auction dataset without complete solution only generates invalid solutions--either the outputs violate the constraints, or the solutions are all zeros (which are meaningless).

Could you please provide a more detailed training procedure for reproducing the results in this setting? Are there any additional tricks, specific parameter choices, or implementation details necessary to obtain valied results?

Thanks!

Hi, I've noticed the same problem, have u found any possible approaches to tackle this? Really appreciate it if u've got any insights to share!