mismatch of code and pseudocode

[explorer1212]

Hello,
in SDE-BFN-Solver2(on discrete data),
in the pseudocode, the coefficient is (-1/3), but in the code, the coefficient is (1/3)


So, which one is correct?

[qky18]

Hi, I also found some difference between the formula and the code implementation, for example in Eq. (29)

and this line of code flips the sign of the first drifting term.

x_t = x_s + g**2 * (data_pred - 1/self.K) * self.delta_t + g * self.delta_t**0.5 * noise

I was wondering if there's anything I've overlooked or misunderstood, or is it a bug that needs fixing?

[explorer1212]

@qky18 I've posted some questions under your paper and received great answers, and surprisingly, I meet you here. Let me also take a look at this issue.

[qky18]

@explorer1212 So nice seeing you here and exploring BFN together! Maybe we can talk some day.

[explorer1212]

@qky18 I am looking forward to this talk!

[qky18]

@explorer1212 Hi, I cannot find your email address. Email me at qky22[AT]mails.tsinghua.edu.cn if you'd like to!