Unable to install "benchmark" in the requirement.txt

[Joan-YAO]

I have tried several versions of python and upgraded the pip, setuptools but cannot install the benchmark module. Can you provide the python version you used or are there any external dependencies needed for this benchmark module? Or should it be installed manually?

[Joan-YAO]

It seems the benchmark in the requirement.txt is not a the one available online but the module in the TransSimHub. I just remove the benchmark in the requirement.txt.

[Joan-YAO]

Hi I am still struggling to debug the code as it reports the following error even I configured all the needed environment and modules. The gym version I installed is 0.26.2 and the stable-baseline3 is 2.1.0. Did you encounter the same error of "Get Edge Variable"? It seems to be an error caused by the TransSimHub instead of the RL function. If yes, we can discuss on how to debug. Thanks!

…

-- Jiarong Yao Email: ***@***.*** At 2024-10-22 21:19:37, "jianshangmao" ***@***.***> wrote: May I ask if you have successfully run it? Can you provide all the software package versions required for the operation, especially the Python gym stable_maselines3 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[jianshangmao]

I have successfully run rule_based. py and am currently running sb3_ppo. py. I did not encounter the problem you encountered. I use the following code to verify that my TransSimHub version is 1.1
import tshub
print(tshub.__ version__)

[Joan-YAO]

Hi, I encountered the error of "Get Edge Variable" when I am running sb3_ppo. py. Did you succeed in running the sb3_ppo. py without error? Regards

…

-- Jiarong Yao Email: ***@***.*** At 2024-10-23 12:24:29, "jianshangmao" ***@***.***> wrote: I have successfully run rule_based. py and am currently running sb3_ppo. py. I did not encounter the problem you encountered. I use the following code to verify that my TransSimHub version is 1.1 import tshub print(tshub.__ version__) — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[jianshangmao]

I have been running sb3_ppo.py for 16 hours, but it's still not finished.So I don't know if I will encounter any problems in the future。

[Joan-YAO]

Hi, I suspect that the error of "Get Edge Variable" may be due to a low version of SUMO as the environment variable "SUMO_HOME" is called in the training of sumo simulation. I am using SUMO 1.13. What is the version of your SUMO installed on your device? Regards,

…

-- Jiarong Yao Email: ***@***.*** At 2024-10-23 13:22:37, "jianshangmao" ***@***.***> wrote: I have been running sb3_ppo.py for 16 hours, but it's still not finished.So I don't know if I will encounter any problems in the future。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[jianshangmao]

I use SUMO 1.21.0

[Joan-YAO]

Hi Thanks for your reply! I have checked the error of "Get Edge Variable", and found that it is related to a command of getting the angle of edges, which matters little to the function of the codes, so I just commented it and successfully got the codes running. As it keeps training through multi-process and multi-timestep, I have checked the total timestep of the training (1e7) and the number of threads (5), also count the duration of one simulation (3600s), which is about 1min on my desktop. Thus it may take 1min * 5 *1e7/3600 = 1389h to finish the training, which may account for the long time you spent on running the code. I decided to cut down the total timesteps to have the trained model first to test the sequent LLM part. Hope to give you some reference!

…

-- Jiarong Yao Email: ***@***.*** 在 2024-10-23 14:01:21，"jianshangmao" ***@***.***> 写道： I use SUMO 1.21.0 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[jianshangmao]

I ran Python eval_rl_agent.py. Encountered this issue during runtime, do you have a good solution?

[Joan-YAO]

I ran Python eval_rl_agent.py. Encountered this issue during runtime, do you have a good solution?

It seems the class "make_env" lacks one attribute "trip_info", so I just added it to make it run.

[jianshangmao]

Thanks for your reply! I set trip_info = path_convert(f'./Result/RL.tripinfo.xml') in eval_rl_agent.py, and then the RL.tripinfo.xml file was generated. The Evaluation Metrics in the paper, such as Mean waiting time. Do I need to calculate the mean waiting time myself in this XML file.

[Joan-YAO]

Thanks for your reply! I set trip_info = path_convert(f'./Result/RL.tripinfo.xml') in eval_rl_agent.py, and then the RL.tripinfo.xml file was generated. The Evaluation Metrics in the paper, such as Mean waiting time. Do I need to calculate the mean waiting time myself in this XML file.

Hi, the data in tripinfo.xml are performance indicators of each vehicle/trip. The waitingtime is the total time when the vehicle speed was below or equal 0.1 m/s during its trip. If you want to have a mean waiting time of all the vehicles input in the simulation, I would recommend to use the statis output file.

[jianshangmao]

thank you，i get it

[pangay]

Sorry, I just saw this issue. Regarding the training step size, generally, 3e^5 should be sufficient for convergence. Could you let me know if there are still any unresolved issues? The 'trip_info' has been updated by TransSimhub, but this code has not been updated accordingly. Apologies for the inconvenience, I will sync and update a new version soon. Regarding the calculation of vehicle waiting times, TransSimhub does support it. You can refer to the documentation here: https://transsimhub.readthedocs.io/en/latest/locales/zh_CN/index.html. You can also use this script to analyze the output tripinfo file: https://github.com/Traffic-Alpha/TransSimHub/blob/main/examples/sumo_tools/analysis_output/analysis_tripinfo.py. If you still have issues, feel free to leave a comment or contact me via email. I apologize again for the inconvenience caused by the usage.

[Joan-YAO]

Thanks for your update! I have one more question regarding the LLM prompt. I designed my own prompts and added them into the tool list of the agent attribute but the newly added prompts did not work. Also, I cannot find the exact codes of this function "self.agent.run()" thus I fail to trace where the tools of the LLM agent take effect. All usages of these tools like "GetIntersectionLayout(), GetSignalPhaseStructure()" are only found in the function definition and agent initialization in the code package.