Implementation of World Models paper

Initially we collect rollouts of the environment using a random policy.

python datasets/carracing.py

Generated rollouts will be placed in random directory in data/carracing folder

Later we train all the three modules i.e., VAE module, Memory module and Controller module independently

python src/train_convVAE.py

python src/train_lstm.py

python src/evolution_pooling.py

To test average rewards in gym environment

python src/test_gym.py

To visualise vae original and reconstructed images for analysis

python utils/visualize_vae.py

To plot the graphs from the controller training for analysis

Run utils/plot_utility.ipynb file

Logging from model training are available in logs folder

Saved model checkpoints are available in checkpoints folder

We have referred ctallec repo