Match Scheduling Algorithm Competition

[Joe Ross]

I wonder why FIRST didn't publicize this?

[Billfred]

Originally Posted by Joe Ross

I wonder why FIRST didn't publicize this?

Best I can tell from rooting through their site, this particular challenge is their championship round and is thus invite-only.

The challenge has seven quality metrics they're trying to minimize (with each metric carrying its own definable weight):

-Difference in average team age on each alliance between the two alliances
-Difference in average rank of a team (defined from 1 to 10 in the challenge, where 1 is rookie and 10 is reigning champion) between the two alliances
-Unique partners
-Unique opponents
-Time between matches (more is better)
-Assignment to red or blue
-Position in player station

If FIRST took the current match assignment criteria (as defined in The Tournament--PDF link) and put the age and (reasonably-calculated) rank criteria in beneath the unique opponent/partner criteria, then it could reduce the number of "oh-crap-we-play-against-Beatty-AND-Simbots-this-match?!" rounds--at that point, the algorithm is likely to switch one of them to your alliance (especially at Championship, where they'd have the benefit of regular-season data).

TLDR: This could evolve into a good thing or a bad thing, depending on what weight you put on the factors.

[AdamHeard]

FIRST has no valid way of ranking teams, and thus the only fair way to do selections is randomly. Factoring age or some arbitrary rank will never be fair.

Originally Posted by Billfred

Best I can tell from rooting through their site, this particular challenge is their championship round and is thus invite-only.

The challenge has seven quality metrics they're trying to minimize (with each metric carrying its own definable weight):

-Difference in average team age on each alliance between the two alliances
-Difference in average rank of a team (defined from 1 to 10 in the challenge, where 1 is rookie and 10 is reigning champion) between the two alliances
-Unique partners
-Unique opponents
-Time between matches (more is better)
-Assignment to red or blue
-Position in player station

If FIRST took the current match assignment criteria (as defined in The Tournament--PDF link) and put the age and (reasonably-calculated) rank criteria in beneath the unique opponent/partner criteria, then it could reduce the number of "oh-crap-we-play-against-Beatty-AND-Simbots-this-match?!" rounds--at that point, the algorithm is likely to switch one of them to your alliance (especially at Championship, where they'd have the benefit of regular-season data).

TLDR: This could evolve into a good thing or a bad thing, depending on what weight you put on the factors.

[Billfred]

Originally Posted by AdamHeard

FIRST has no valid way of ranking teams, and thus the only fair way to do selections is randomly. Factoring age or some arbitrary rank will never be fair.

Agreed--there will always be really great rookies and really awful veterans, and there are teams that reach Einstein one year that fall off the radar entirely the next. No formula accounting past years' performances will account for student and mentor turnover perfectly.

However, one could certainly compute rank for within the season, especially at the Region Championship and Championship levels where everyone has played this game before.

Edit: Dave got in between my reading and my posting. I'm not advocating for those two to be anywhere before line D in this year's criteria (read: just ahead of equal red/blue assignments), if anywhere. I'd definitely want to see some sample outputs before I put my blessing on any such method.

[EricH]

Originally Posted by Billfred

The challenge has seven quality metrics they're trying to minimize (with each metric carrying its own definable weight):

-Difference in average team age on each alliance between the two alliances
-Difference in average rank of a team (defined from 1 to 10 in the challenge, where 1 is rookie and 10 is reigning champion) between the two alliances

TLDR: This could evolve into a good thing or a bad thing, depending on what weight you put on the factors.

I'd go with ditching the age entirely. Remember the Algorithm of Doom in 2007, where you might end up against 1114 AND 2056 (a rookie that year) AND 71 at the same time, and all of them multiple times (assuming that the age breakup worked out that way), and very rarely with one or the other?

Yeah. Age has very little to do with anything around here--I'm somewhat scared of what Dr. Joe is starting up in Boston...or any team 1114 mentors.

The rank part, OTOH, I can live with, if it's done right. If you're careful about the implementation, and use the available data about record/seeding/result, then I think it could work very well. If you're arbitrary about it, like the Algorithm of Doom was with age, it'll go down as a lousy idea.

[Dave Scheck]

Originally Posted by Billfred

then it could reduce the number of "oh-crap-we-play-against-Beatty-AND-Simbots-this-match?!" rounds

Take the other point of view and ask yourself how to minimize the "Hey, we're one of the top teams at this event and we get to play with a box on wheels and one that only spun in circles last match". True random is the only way that you can make fair schedules. The last time FIRST tried to introduce "smarts" into the system the worst schedules in history were seen (remember the low/mid/high system)...I hope they never try to do that again.

EDIT: Looks like Eric was thinking along the same lines as me.