Thank you so much for sharing your excellent work! I have carefully studied your code and gained valuable insights. I have a few questions that I would like to ask for your advice:
In the _make_spec(self, generator: HCVRPGenerator) method in env.py, regarding the dimensions of the observations—for example, the dimension of locs is shape=(generator.num_loc + 1, 2). Why is it not shape=(generator.num_loc + generator.num_agents, 2)? Similarly, the dimension of action_mask is shape=(generator.num_loc + 1, 1) instead of shape=(generator.num_loc + generator.num_agents, 1). Although each vehicle shares the same depot (merely duplicated), the dimension design does not seem to account for the number of agents. If I were to customize my own problem—for instance, each city requires three tasks to be completed sequentially—should the first dimension of locs and action_mask be generator.num_loc + 1 or generator.num_loc * 3 + generator.num_agents?
Regarding the baseline algorithms mentioned in your paper, do you plan to open-source them as well? I would greatly appreciate it if you could share the implementations!
Thank you so much for sharing your excellent work! I have carefully studied your code and gained valuable insights. I have a few questions that I would like to ask for your advice:
In the _make_spec(self, generator: HCVRPGenerator) method in env.py, regarding the dimensions of the observations—for example, the dimension of locs is shape=(generator.num_loc + 1, 2). Why is it not shape=(generator.num_loc + generator.num_agents, 2)? Similarly, the dimension of action_mask is shape=(generator.num_loc + 1, 1) instead of shape=(generator.num_loc + generator.num_agents, 1). Although each vehicle shares the same depot (merely duplicated), the dimension design does not seem to account for the number of agents. If I were to customize my own problem—for instance, each city requires three tasks to be completed sequentially—should the first dimension of locs and action_mask be generator.num_loc + 1 or generator.num_loc * 3 + generator.num_agents?
Regarding the baseline algorithms mentioned in your paper, do you plan to open-source them as well? I would greatly appreciate it if you could share the implementations!