Should new sample and training data have same time stamp if it is the input

[JunHanStudy]

Thank you for nice repo on Diffusion model for time series data.

When running your codebase, I observe that the the time stamp is an input to both encoder and latent diffusion model. When drawing new samples, the same encoding of time stamps as the training data are used as inputs. Idealy, could the new samples have different time stamps from the training data as inputs?

[namjoonsuh]

Hi Han. Thanks for the inquiry and interests in our work.
In VAE, the timestamp is not needed by design of our VAE. Just refreshed the updated code.
For sampling, as you pointed out, different timestamps within the timeframe in the training dataset can be used!