joonsumisu / durleca

The released code for DUal-objective Reinforcement-Learning Epidemic Control Agent.

DURLECA

Here is the released code for the DUal-objective Reinforcement-Learning Epidemic Control Agent (DURLECA) presented in our KDD paper Reinforced Epidemic Control: Saving Both Lives and Economy.

Use your own OD dataset

Please note that we cannot provide public access to the real-world Beijing dataset due to privacy and ethical concerns. However, the dataset may be accessed upon request and the data provider's authorization. We also suggest and welcome other users test DURLECA on other real-world datasets from their own resources.

Please change the dataset path in the utils.py file.

Train

  python main.py \
    --gpu 2 --task train \
    --steps 400000  --batch_size 16 --lr 1e-4\
    --expert_h 1 --expert_lockdown 168 --prob_imitation_steps 200000 --base_prob_imitation 0.5 \
    --repeat 24 --rd_no_policy_days 25 --fixed_no_policy_days_list 0 10 20 \
    --mobility_decay 0.99 --L0 72 --H0 3\
    --I_threshold 100 --lockdown_threshold 336 \
    --beta_s 0.1 --beta_m 3 --gamma 0.3 --theta 0.3 \

Users could also adjust parameters on their own to simulate different diseases or to have different objectives.

Test

  python main.py \
    --gpu 0 --task test_list --verbose True --save_path [YOUR PATH] \
    --fixed_no_policy_days_list 20\