search

Chacha-Chen / MPLight

56 stars 19 forks source link

Not found error #1

Closed oroojlooy closed 4 years ago

oroojlooy commented 4 years ago

Hi, It seems that either the README or the repository are not update. When I follow the instruction, I get:

python -O run_batch.py
python: can't open file 'run_batch.py': [Errno 2] No such file or directory

Beside, when I directly call runexp.py, I get:

python runexp.py
Using TensorFlow backend.
Traceback (most recent call last):
  File "runexp.py", line 1, in <module>
    import config
  File "/net/ge.unx.sas.com/vol/vol110/u11/aforoo/traffic/MPLight/config.py", line 12, in <module>
    from dgn_agent import DGNAgent
ModuleNotFoundError: No module named 'dgn_agent'

Would you mind take a look?

Chacha-Chen commented 4 years ago

Hi, I'll update the README when I got time.

If you are interested in applying RL in a multi-agent scenario, I recommend you look into these two repos: 1. https://github.com/wingsweihua/presslight 2. https://github.com/wingsweihua/colight

oroojlooy commented 4 years ago

Thanks for the quick response. I was able to run presslight, but was interested to MPLight since it has smaller state definition. Beside, in the paper there is not any point about SumoEnv, although the code can either use Sumo or CityFlow. Did you use Sumo in the experiments?

Chacha-Chen commented 4 years ago

I used Cityflow (anon_env.py). The state definition is quite similar to PressLight. I just replaced the lane-num-vehicles to delta-lane-num-vehicles (the discrepancy of the vehicle number in the current lane and its downstream lane). You could find the implementation at the anon_env.py.