Question about apply_conditioning with self.predict_epsilon during training

[OAHC2022]

Firstly, thank you for your amazing work!

I have a question regarding the code in diffusion.py (https://github.com/jannerm/diffuser/blob/main/diffuser/models/diffusion.py#L215). Specifically, when self.predict_epsilon is set to True, I understand that x_recon represents the predicted noise.

In this scenario, does applying apply_conditioning to x_recon still make sense? Additionally, if the cond is empty, it seems like the line

x_noisy = apply_conditioning(x_noisy, cond, self.action_dim)

at line 212(https://github.com/jannerm/diffuser/blob/main/diffuser/models/diffusion.py#L212) would have no effect. However, I assume this could still be useful to constrain the start condition during training?

Thank you for your time and consideration!