Open KrisThielemans opened 1 year ago
@KrisThielemans Which notebook are you referring to?
SIRF-Exercises
notebook ML_reconstruction
compares OSEM with gradient ascent, and so far as I remember, other PET notebooks in SIRF-Exercises
do not deal with gradient descent too.
Synergistic/gradient_descent_mr_pet_ct.ipynb
. No need to fix this now though. Next release!
b6b0d6be1bf7ce5a6d18090aa6a1879d286c1d98 edited the notebook a bit to warn people about the confusion on signs.
@ckolbPTB should we move this notebook to Introductory
in the next release? There isn't anything "synergistic" here really, and it nicely follows up from the other ones in that folder.
@ckolbPTB what do you think about moving this to Introductory
?
@evgueni-ovtchinnikov would you have time to change the sign here and adjust?
currently the "ascent" is addressed by updating text in the notebook. This is good enough for 3.5. We'll multiply the PET objective function with -1 and descend in the next version
Move now completed.
The only thing remaining here is the strange line in the PET reconstructions.
the PET recon has a funny line![image](https://user-images.githubusercontent.com/6362141/214108091-da4de06c-e419-4f96-b3b1-600325dfd9d9.png)
Also, it is called gradient descent, but it is actually ascending due to some minus signs. Better to switch the sign in the PET, but keep it for MR and CT.