chendrag/cep-energy-guided-diffusion

Official codebase for Exact Energy-Guided Diffusion Sampling via Contrastive Energy Prediction （ICML 2023）

cep-energy-guided-diffusion's Issues

list.astype(np.float32) cause err

When running train_critic.py after training behavior model, line 106 raises an error because list doesn't have astype method. How to fix this?

Mismatch of the checkpoint

Thank you very much for sharing the pre-trained models. But I found some unexpected keys with the following error:

RuntimeError: Error(s) in loading state_dict for QGPO_Critic:
Unexpected key(s) in state_dict: "vf.v.0.weight", "vf.v.0.bias", "vf.v.2.weight", "vf.v.2.bias", "vf.v.4.weight", "vf.v.4.bias", "gaussian_policy.net.0.weight", "gaussian_policy.net.0.bias", "gaussian_policy.net.2.weight", "gaussian_policy.net.2.bias", "gaussian_policy.net.4.weight", "gaussian_policy.net.4.bias".

I also didn't find the nets, vf and gaussian_policy in the file model.py. Could you help me what's wrong?

usage of fake_actions and fake_next_actions

I am confused about fake_actions and fake_next_actions. Why do you use fake_actions and fake_next_actions in train_critic? Why their sizes are different from dataset.actions?

Weights for the Behavior Models?

Hi, Could you provide the pre-trained behavior models weights for the locomotion and antmaze environments? This would save me a lot of time for an experiment I need to run!

Thanks