This branch contains the DDPM code we used for the experiments in Section 5.1 of the paper.

git checkout ddpm

Assuming you have followed the installation steps on the main branch, you'll need to run in addition:

pip install diffusers==0.17.0

Note that the experiments in Section 5.1 trained the world models on a static dataset collected with an expert policy (use training.model_free=True to train a model-free expert, then collect a dataset and put its path into collection.path_to_static_dataset).

Otherwise, to train from scratch a DDPM-based diamond agent, run:

python src/main.py