search

ermongroup / CSDI

Codes for "CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation"
MIT License
261 stars 79 forks source link

testing time #8

Closed xchuwenbo closed 6 months ago

xchuwenbo commented 10 months ago

I ran the code you posted on github and found that the time of training for a epoch takes about half a minute but the testing takes about half a hour(The command is “python exe_physio.py --testmissingratio 0.1 --nsample 100”). May I ask if this is in line with the actual situation.

y-tashi commented 7 months ago

Hi, sorry for the late reply. Yes, it's working correctly. Our code is not fast at test time. To accelerate the speed, You can combine CSDI with methods which speed up diffusion models.

xchuwenbo commented 6 months ago

Hi, Tashiro,

Thanks for your reply. We have already finished the comparison experiment and report as it was described in the paper.

Wenbo

tldr: our recent work is deep ensembles method combined with quantile regression. it is faster than the generative models, such as the diffusion models. Deep Ensembles Meets Quantile Regression: Uncertainty-aware Imputation for Time Series. Ying Liu+, Peng Cui+, Wenbo Hu, Richang Hong https://arxiv.org/abs/2312.01294