Hello,
We are attempting to using variational bayesian inference to estimate an event-related fMRI model, and have been comparing the models and estimates created by our scripts and those made from the GUI. Editing the scripts to make the model specification SPM.mat identical to the GUI SPM.mat, the only difference in the estimated model is within SPM.PPM.Sess.block.elapsed_seconds. There doesn’t seem to be a clear pattern as to how the handmade and scripted models differ here, but each individual value can differ by as much as a whole second — again with all the other values of the entire estimated model struct consistent between the two.
Could you elucidate the function of this elapsed_seconds vector? I have scoured the spm_spm_vb.m and spm_vb_taylor_R.m scripts, which seem to be calculating that section, and see where the elapsed_seconds vector is moved into the final struct (spm_spm_vb line 926) but not how it is calculated. Since the rest of the structs are the same, could we assume we’re getting identical results outside of this vector? Any help would be appreciated!
Best,
John Thorp
|