Skip to content

Applying representation learning to reinforcement learning

Notifications You must be signed in to change notification settings

cpuheater/representation_rl_lab

Repository files navigation

Applying representation learning to reinforcement learning

Using denoising autoencoder to learn compress representations for the purpose of using it to train RL agent.

Collect data while playing Doom

python play_and_collect.py --image-dir=images

Train an autoencoder using previously collected data

python train_ae.py --images-dir=images --model-dir=trained_models

Train PPO agent using compress representation of the state.

python train_ppo.py --ae-path=trained_models/mymodel.pt

Train DQN agent using compress representation of the state.

python train_dqn.py --ae-path=trained_models/mymodel.pt

Releases

No releases published

Packages

No packages published