evaluation : structure of function input parameters

[Tinysimpler]

Hello, thanks for your codes, I learn from it much. there's a problem confused me, hope your help. I want to evaluate the performance of AM and AM-ASP for CVRP problem, but i don't know the structures of input parameters in get_mix_solver(model_args, mix_prob, param_list) from https://github.com/LOGO-CUHKSZ/ASP/evaluation.py, i have tried some ways, but it doesn't work, hope your reply ～

[LOGOLAB-CUHKSZ]

Hi there! The evaluation.py file was only created for testing purposes at the time. For the fair evaluations, you can test the obtained neural solver from ASP using the corresponding neural solver's open-source codes by loading the model parameters solver_param in saved asp_info.pt.
For your specific case, we recommend checking out the AM repository: https://github.com/wouterkool/attention-learn-to-route , and using their approach to load models and data for testing. If you encounter any difficulties during this process, feel free to let us know.

[Tinysimpler]

your reply helps me a lot, appreciate you, based on your advice, these days i try to evaluate models in https://github.com/wouterkool/attention-learn-to-route and in your project ASP，i'm confused by a new problem, that is the performance of AM is better than the 【AM + ASP】in CVRP 20/50/100 AM cases. In that evaluation, AM model i used is the pretrained cvrp_20/50/100, and 【AM + ASP】model is in the saved_asp/CVRP/AM/cvrp_am/all_best_asp_info.pt, Is there something wrong with my evaluation? Or whether i need update the ASP model(such as update the objective functions or model net) and train it again？hope your reply, and Happy New Year for YOU……