System Requirement

windows : tensorflow=2.13.0
ubuntu 18.04 : tensorflow=2.13.1

Install

First git clone

git clone https://github.com/gordanLiang/highway-environment-NN-model-training.git

cd to file

cd highway-environment-NN-model-training

Using conda environment python==3.8

conda create -n <env name> python=3.8

activate environment

conda activate <env name>

intstall requirements with

pip install -r requirements.txt

Training DQN model with stable baseline 3

Training dqn model using

python highway_dqn_train.py

Testing dqn model using

python highway_dqn_test.py

It will give 10 score each test 10 rounds and collect reward.

Generating dataset with pre-trained dqn model

Load model and generate dataset in highway environment by using

python dataset_making_each1w.py

It will generate total 50000 data for each action 10000 data

Training NN model using dataset

Before generating dataset, we can use it to train the NN model and test it in highway environment.
Training the NN model using

python NN_model_train.py

Testing in highway environment

python NN_model_test.py

This is the same way tesing dqn model so can compare.

Result

DQN 10 reward

[212, 160, 229, 227, 207, 235, 227, 219, 171, 230]
average:211.7

NN model 10 reward

[229, 227, 222, 214, 189, 241, 212, 194, 222, 198]
average:214.8

References

highway envirenment:https://github.com/Farama-Foundation/HighwayEnv
stable baseline 3:https://github.com/DLR-RM/stable-baselines3