custom metrics

[nuaazj]

https://flow-project.github.io/contact.html

[nuaazj]

Hello, everyone!
I adopt multi-agent reinforcement learning to research on autonomous vehicles driving decision making problems at intersection without traffic lights, where all the vehicle's acceleration is decided by the trained policy in order to rapidly and safely through the intersection. However, in FLOW project, the end of the training through policy replay to assess performance, how do I calculate the custom metrics in the process of training process just like reward, such as average speed, collision rate and travel time delay, I would appreciate your sincere help.