A Consciousness-Inspired Planning Agent for Model-Based Reinforcement Learning

By Mingde "Harry" Zhao, Zhen Liu, Sitao Luan, Shuyuan Zhang, Doina Precup and Yoshua Bengio

(BLOGPOST)

Install Dependencies

pip install -r requirements.txt

Reproducing Results (for main manuscript)

CP

python run_distshift_randomized_mp.py --method DQN_CP --num_explorers 8 --ignore_model 0 --layers_model 1 --signal_predict_action 1 --disable_bottleneck 0 --size_bottleneck 8

UP

python run_distshift_randomized_mp.py --method DQN_CP --num_explorers 8 --ignore_model 0 --layers_model 1 --signal_predict_action 1 --disable_bottleneck 1

WM

python run_distshift_randomized_mp.py --method DQN_WM --num_explorers 8 --ignore_model 0 --layers_model 1 --signal_predict_action 1 --disable_bottleneck 0 --size_bottleneck 8 --period_warmup 1000000

Dyna

python run_distshift_randomized_mp.py --method DQN_WM --num_explorers 8 --ignore_model 0 --layers_model 1 --disable_bottleneck 0 --size_bottleneck 8 --learn_dyna_model 1

Dyna*

python run_distshift_randomized_mp.py --method DQN_WM --num_explorers 8 --ignore_model 0 --layers_model 1 --disable_bottleneck 0 --size_bottleneck 8 --learn_dyna_model 0

NOSET

python run_distshift_randomized_mp.py --method DQN_WM --num_explorers 8 --ignore_model 0 --layers_model 2 --len_hidden 256

Reproducing Results (for additional dynamics)

Changes are to be synced from our private repository.