Usage
Installation
Install SUMO (ideally 1.12, but 1.13 seems to work as well) according to your OS, and make sure that the env var SUMO_HOME is set.
Create a conda env or venv with python 3.7
Install dependencies using following command:
pip install -r requirements.txt
Run Instructions
<dir>
is where all the training artifacts will be stored and/or the checkpoints will be retrieved (to evaluate or restart the training).
Will be created if it doesn’t exist.
From the root of the repo and the correct venv/ conda env, run the following command to train the agents. Check code/train.py
for more options.
python code/train.py --dir <exp_dir>
python code/train.py --dir <exp_dir> --kwargs <python dict with arguments to override the config in main.py>
Example:
python code/train.py --dir wd/test --kwargs "{'wandb_proj':'scenario-env'}"
From the root of the repo and the correct venv/ conda env, run the following command to evaluate the trained agents on a given intersection dataset.
Check code/evaluate.py
for more options. Note that <dir>
should point to the same directory used for training.
python code/evaluate.py --dir <exp_dir>
From the root of the repo and the correct venv/ conda env, run the following command to visulize the trained agents on a given intersection dataset.
Check code/visualize.py
for more options. Note that <dir>
should point to the same directory used for training.
python code/visualize.py --dir <exp_dir>