search

Pi-Star-Lab / RESCO

Reinforcement Learning Benchmarks for Traffic Signal Control (RESCO)
116 stars 36 forks source link

Cumulative rewards and Delay time after training #2

Closed TrinhTuanHung2021 closed 2 years ago

TrinhTuanHung2021 commented 2 years ago

Thank you for sharing your code. Your code is very helpful but there are some things I don't understand.

After training model "python main.py --agent IDQN --map ingolstadt21" , I found that it generates several files in the logs directory including agent.pt, metric.csv and tripinfor.xml. But I don't see the cumulative rewards per episode because I want to know when the model converges. And I also don't see where the model is saved, how can I test the model after training (trained_model.h5). Thank you very much

jault commented 2 years ago

The performance output needs some work right now for easier analysis. metrics_x.csv will contain the reward for each agent for each time step for episode x. Summing these would give cumulative reward. Each agent type will need a load function to support separate testing parameters. They'll be included in the next update.