Super helpful. That’s where I found the formula in enough specificity that I could port it. @stevebronder just helped me figure out what I need for the template guards for types, so I should be able to roll in the custom gradients this week.
The reverse mode is all passing tests, but I’m going to add analytic autodiff. This is another case where I probably should do both var-matrix and matrix-var solutions
Howdy folks! Just stumbled upon this thread as we were looking for a discrete FFT in Stan with the “new” complex numbers support. Do we know if and maybe this is still going to make it into a release? Thank you!
Do you mean if FFT will be in the 2.30 release? If that is the question, then yes, the next release (RC coming out in a few hours) will include the fft() and inv_fft() as well as the 2D implementation fft2() and inv_fft2().
wOOOOOOOO!!! Thank you @rok_cesnovar @WardBrian @Bob_Carpenter and the rest of the team! Stan never fails to impress!