Multi-agent traffic signal control for DeeCamp2019
Training
DQN
python run_rl_control.py --algo DQN --epoch 200 --num_step 2000 --phase_step 1
Double DQN
python run_rl_control.py --algo DDQN --epoch 200 --num_step 2000 --phase_step 1
Dueling DQN
python run_rl_control.py --algo DuelDQN --epoch 200 --num_step 2000 --phase_step 1
Inference
DQN
python run_rl_control.py --algo DQN --inference --num_step 3000 --ckpt model/DQN_20190803_150924/DQN-200.h5
DDQN
python run_rl_control.py --algo DDQN --inference --num_step 2000 --ckpt model/DDQN_20190801_085209/DDQN-100.h5
Dueling DQN
python run_rl_control.py --algo DuelDQN --inference --num_step 2000 --ckpt model/DuelDQN_20190730_165409/DuelDQN-ckpt-10
Simulation
. simulation.sh
open firefox with the url: http://localhost:8080/?roadnetFile=roadnet.json&logFile=replay.txt
Training
QMIX (based on Ray)
python ray_multi_agent.py
MDQN (manul implementation)
python run_rl_multi_control.py --algo MDQN --epoch 1000 --num_step 500 --phase_step 10
MDQN (based on Ray)
python ray_multi_dqn.py
Inference
MDQN (manul implementation)
python run_rl_multi_control.py --algo MDQN --inference --num_step 1500 --phase_step 15 --ckpt model/XXXXXXX/MDQN-1.h5
MDQN (based on Ray) (in lab linux)
python ray_multi_dqn_rollout.py --run DQN --checkpoint ~/ray_results/DQN_cityflow_multi_2019-08-11_00-44-52khzt8bnq/checkpoint_400/checkpoint-400 --env cityflow_multi --steps 1000
1*6 roadnet
Generate checkpoint
python run_rl_multi_control.py --algo MDQN --epoch 1 --num_step 1 --phase_step 10
Generate replay file
python run_rl_multi_control.py --algo MDQN --inference --num_step 130 --phase_step 30 --ckpt model/MDQN_20190809_134734/MDQN-1.h5
Simulation
. simulation.sh
open firefox with the url: http://localhost:8080/?roadnetFile=roadnet.json&logFile=replay.txt
Ciryflow deecamp branch
git clone -b deecamp https://github.com/zhc134/CityFlow.git
pip install .
