Official code for: Conformal prediction interval for dynamic time-series (conference, ICML 21 Long Presentation) AND Conformal prediction for time-series (journal, IEEE TPAMI)
MIT License
95
stars
21
forks
source link
Question about the Error Update in the Online Case #2
I have read your paper and I am curious about Line 16-18 in your algorithm. Could you take a look at the following question?
Specifically, when you say the set $\epsilon$ gets rid of the first error, and includes the new prediction error (line 18), do you mean the set always removes the first error in terms of time? For example, if $s=1$ and I have $\epsilon = [e_1, e_2, e_3]$ denoting the errors at time 1, 2, 3, and I just got the prediction error for time 4, $e_4$, will it always be the case that $\epsilon$ gets rid of $e_1$ and adds $e_4$ to the list? Like a queue? I am a little confused by your set union notation and what you mean by "reset index of $\epsilon$"
Hi Chen,
I have read your paper and I am curious about Line 16-18 in your algorithm. Could you take a look at the following question?
Specifically, when you say the set $\epsilon$ gets rid of the first error, and includes the new prediction error (line 18), do you mean the set always removes the first error in terms of time? For example, if $s=1$ and I have $\epsilon = [e_1, e_2, e_3]$ denoting the errors at time 1, 2, 3, and I just got the prediction error for time 4, $e_4$, will it always be the case that $\epsilon$ gets rid of $e_1$ and adds $e_4$ to the list? Like a queue? I am a little confused by your set union notation and what you mean by "reset index of $\epsilon$"
Thank you so much.
Sincerely, William