Clarification regarding conditional generation

Y-debug-sys / Diffusion-TS

[ICLR 2024] Official Implementation of "Diffusion-TS: Interpretable Diffusion for General Time Series Generation"

MIT License

173 stars 24 forks source link

Clarification regarding conditional generation #70

Open saishankarn opened 4 hours ago

saishankarn commented 4 hours ago

Hi, from my understanding of the code, in line 303 of Models/interpretable_diffusion/gaussiandiffusion.py, the sample refers to x{t-1}. However, within the langevin_fn, the input to the denoiser in line 390 is xt and t. While execution, I believe the input is x{t-1} (i.e., sample) and t. Is my understanding correct?

Y-debug-sys commented 4 hours ago

Yes, you are correct.

saishankarn commented 1 hour ago

The implementation refers to Eq. 11 in the paper. So, should it be t-1 as input in line 390?