Can Wikipedia Teach Us About the World?

Impact of Language-Pretraining on Model-Based RL

Currently WIP.

Install packages from requirements.txt

Execute the following command to run an experiment:

python -O twm/main.py --game CartPole --seed 0 --device cuda:0 --cpu_p 1.0 --config cartpole --wandb disabled 

Use --wandb online to log the metrics in weights and biases.
To use other hyperparameters, edit the file twm/config.py.

See analysis folder for notebooks used to generate analysis for the report.