Best dataset configuration for reproduction and evaluation protocol

[lukangjie]

Describe the question
I am a graduate student researching multi-goal path planning, and I am interested in reproducing the results of iAstar. After reviewing the codebase, I have a few questions regarding the dataset setup and the testing phase:

1. Dataset Configuration: To reproduce the results in the paper, what is the recommended way to set dataset and val_dataset? Specifically, should I use MP for the training dataset and maze for the val_dataset?
2. Optimal Model: What are the key metrics or checkpoints I should look for to obtain the best .pkl file? Are there specific training tips (e.g., convergence signs) to ensure the model reaches the reported performance?
3. Evaluation Protocol: For the final testing, should the model be evaluated across the three maps mentioned in the paper? Is there a specific script or parameter set for this evaluation phase?
   Thank you for your great work and for providing the source code!