Gaudenz Steinlin <gaud...@debian.org> writes: > I'm not sure if I interpret your PDF graph correct. Do you suggest that > everyone on the same line forms a cohort and should have the same > priority?
Yes. > If so I don't understand how those that don't have any outgoing edges in > the (reduced) graph get there level. It looks like they are just moved > to the highest possible level. But is that really what we want? People get pushed down by incoming edges. So yes, we fund people as soon as we have funded everyone who is obviously more deserving. I mentioned at the end it might be worth having cutoffs for the two factors (need, contribution) to avoid people being pushed up too far. In the actual test data I think the levels work out more or less sensibly. > > Or to put it another way, why are number 29 and 36 ranked higher than 19 > and 7. According to the ordering specified above there is no clear > justification for this. 29 and 36 are ranked higher because after we have funded 1-5, 28, they are "sources"; nobody else is "clearly better". > It seems to me this problem arises because your graph drops edges over > more than one level. No, that's definitely not the cause. The ranking uses all of the edges. Only the edges to the next level are drawn because it makes the picture less crowded, and because it's enough to provide a "witness" for why someone is _not_ funded. For example it provides an easily checkable way of seeing why 10 is in the fifth cohort. > I'm not sure if this is really the case, but the graph looks like > results in contribution taking precedence over need. Is that correct > and intended? Not exactly. It (implicitely) provides a way of mixing the two factors in making a decision. If we insist on lexicographic ordering (need taking strict precedence over contribution) then we would fund 18 before e.g. 19-21, which doesn't seem like what we'd want. Compared to choosing some weights, the scheme I've proposed has some pros and cons. The main con, is that it's a bit more complicated. For me, it's easiest to understand as a process/algorithm of going through the "queue", making sure we don't fund obviously less deserving (leaving aside the subjectivity of the numbers) people first. On the plus side, it's much more transparent than a set of weights arrived at by the bursaries team in secret. It's also independent of the actual numerical scales used in the ranking. This gives us flexibility to experiment with different scales for need and contribution without reworking a set of weights. It has seemed to me in the past that these weighting schemes have a strong effect on who gets funded. Of course we can "fine tune" the weights until the results "make sense", but then we are getting closer to a completely subjective ranking, since we just adjust the weights until the ranking agrees with our gut feelings. Another important plus is that the core rule (a definitely better than b) is easy to reach concensus on. One concern you bring up is that this scheme maybe doesn't do enough to prevent "undeserving" people being funded (at least, that's my interpretation of your comment about people being moved to the highest possible level). I'm not sure if that's a problem in practice, but a possible way of addressing it would be to choose a cutoff for one or both factors so that in the absence of budget constraints we'd find it acceptable to fund everyone remaining on the list. My gut feeling is that such a cutoff would keep roughly everyone eligible for food and accomodation funding, but note that f+a funding does not currently take into account financial need. d _______________________________________________ Debconf-team mailing list Debconf-team@lists.debconf.org http://lists.debconf.org/mailman/listinfo/debconf-team