The question of CRPS metric

ermongroup / CSDI

Codes for "CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation"

MIT License

305 stars 87 forks source link

Closed packer-c closed 10 months ago

packer-c commented 1 year ago

Hi, thank you for your nice work.

I saw the CRPS metric of the eq. (17) in your paper, my question is why the value of "i" is from 1 to 19.

Look forward to your reply. Thanks!

y-tashi commented 1 year ago

Sorry for the late reply. It takes percentiles from 0.05 to 0.95. Some previous studies also used this discretization and we followed it.