search

Y-debug-sys / Diffusion-TS

[ICLR 2024] Official Implementation of "Diffusion-TS: Interpretable Diffusion for General Time Series Generation"
MIT License
186 stars 27 forks source link

reproducing the results of Figure 6 #18

Closed cmzsbql closed 7 months ago

cmzsbql commented 7 months ago

Hello. Great research work. I have some questions about reproducing the results of Figure 6.

  1. Do I only need to train the model once to obtain all the results in Figure 6 through inference, or do I need to train different models according to different situations?
  2. In the prediction task in Figure 6, do you always keep the history window + prediction window = 48?
  3. In code, how do I switch between Diffusion-TS-G and Diffusion-TS-R？ Thanks.
Y-debug-sys commented 7 months ago

Hi, for your problems :

  1. Yes, of course you can train it just once.
  2. Yes, your right.
  3. Regarding Diffusion-TS-R, just comment the code in gaussian_diffusion.py like following screeshot

Comment1 Comment2

Thanks !

cmzsbql commented 7 months ago

Thanks for your reply, my problem is solved.