I believe that after we discovered and rectified the bug in the application of the ML model, we never appropriately changed the subject set location of those images most effected. i.e. if scratchy were classified less strong when labelled with the bug ---> they were put into a higher workflow (maybe even level 5) and if they wrongly ended in level five they have probably not received nearly as many verification labels and therefore have not experienced expedited retirement that comes from being put in level 1-3/4.
I must go through and make sure with the new ML scores all the images are in the appropraite workflow as described in the Gravity Spy workflow. if not then some images that should be in lower workflows receiving more labels are not getting it.
I believe that after we discovered and rectified the bug in the application of the ML model, we never appropriately changed the subject set location of those images most effected. i.e. if scratchy were classified less strong when labelled with the bug ---> they were put into a higher workflow (maybe even level 5) and if they wrongly ended in level five they have probably not received nearly as many verification labels and therefore have not experienced expedited retirement that comes from being put in level 1-3/4.
I must go through and make sure with the new ML scores all the images are in the appropraite workflow as described in the Gravity Spy workflow. if not then some images that should be in lower workflows receiving more labels are not getting it.
Scotty