Related to #25, currently we take an average of the FI obtained for different theta_i in order to obtain a final FI metric.
But this is actually an average FI, which is not always what you want to report (mostly heuristic in its construction).
Should try to verify if there is actually some variation for different thetas (shouldn't in the nice smooth case, everything being translation invariant). This will have a clear effect when the coverage breaks down (and could be good to report or show).
Maybe try to fix theta for the FI comparisons.
Related to #25, currently we take an average of the FI obtained for different theta_i in order to obtain a final FI metric.
But this is actually an average FI, which is not always what you want to report (mostly heuristic in its construction).
Should try to verify if there is actually some variation for different thetas (shouldn't in the nice smooth case, everything being translation invariant). This will have a clear effect when the coverage breaks down (and could be good to report or show).
Maybe try to fix theta for the FI comparisons.