Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

Rich,

You and I agree.

When I wrote "once the event staff announces them" I meant at the start of the event, not predictions announced before then.

I presumed that once the actual team assignments (to the placeholder IDs that were used to pre-compute the schedule), were entered into the scouting utility, it would then execute some code, or would use Excel's capabilities, or would ... to cook up lists of the matches/teams the scout wants to study.

Blake

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

gotcha...I misread the first time

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

In reply to my "long-shot" question about scheduling matches comprised of more than two alliances, Tom wrote: My answer is "Beats me" :)

I don't have the answers to those questions either.

When/if the subject becomes worthy of serious investigation, maybe we can get some insight from other gaming domains or from game-theory research papers (or from economic theories?).

Blake

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

I've just posted an update to the MatchMaker program that Cathy and I wrote which includes an option for scheduling any surrogate matches in a specified round.

http://www.idleloop.com/matchmaker/

As described in the 2008 manual, surrogate matches will be scheduled in the third round instead of in the last round, so this new option implements that strategy.

This is the engine that FIRST will be using this year to schedule matches at the regionals and the championship, so it would be great to get in some testing before the regionals start off. If you do find any problems, please let me know ASAP!

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

So................

How has the new alliance pairing algorithm treated you so far this season?

I still notice some repetition - are regional staffers correctly parameterizing the algorithm to yield optimal results?

http://www.thebluealliance.net/tbatv/team.php?team=48

Personally, I'm rather agitated and puzzled by the fact that 48 has been stuck in Qualification Match #1 at 3 different regionals, regardless of whether we are the lowest numbered team there or not. At Midwest, 33 didn't even play til Match #5, so it's not like all the lowest-numbered teams are stuck in Matches 1-3.

One begins to wonder if this rather annoying fate is intentionally being forced upon us, given the supposedly random nature of this year's algorithm over last year's...... It's a running joke with us to predict, accurately, that we're in Match #1 at every event we attend, and find out later that we are correct.

After last year, where the Algorithm of Death pretty much forced us to be in one of the first three queued matches, it's GETTING OLD. I'd prefer not to be queued up as soon as we get to the venue on Friday, just for ONE TIME. Is that too much to ask? Perhaps in Atlanta we can have a bit of a break? Cuz, ya know, sometimes it's nice to be able to get a breath in and perform functional checks and such without having the event pit announcers breathing down our necks as soon as we walk in. K? thx, bai.

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

On a side note: Interestingly enough, my team has never been in the first match of any event in all of my years of FIRST.

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

We found at Waterloo regional that the pairing was almost always that you would play WITH a team, and then one match later you would play against that team.

ie. Match X 1075/1114/m v p/q/r, Match Y 1075/188/b v 1114/c/d, Match Z 1075/e/f v 188/g/h

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

The algorithim was very harsh for us. We frequently found ourselves paired with very inexperienced teams and went up against many of the titans of the CT regional (Uberbots, Gaelhawks). It wasn't until our matches on Saturday that we were paired with any really strong teams (love ya Cyberknights), and there was one team we saw three times (once with us, twice against), while many of the 62 teams we never saw, either with or against us.

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

I wouldn't swear that this is the case, but I definitely noticed this trend in our match schedules. This was probably not an intentional design, but rather a side effect of the constraints on match spacing and opponent diversity. If a team was your partner in one match, then the way to ensure that you both get the same amount of downtime is to put you both together in the next match, but on opposite sides. This seems like a pretty good behavior... it should guarantee that no team gets to play with a whole bunch of good teams without having to face some of them, or vice versa.

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

Of course, it still hosed us, because 1114's distro block disconnected itself in our match with them, and they demolished us when against us.

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

I think there's still issues with how the individual regionals are setting things up. At Midwest, we felt the scheduling algorithm did a very poor job (especially since we had to play against 1625 three times in a row). However, the schedule at Boilermaker seemed to have excellent diversity (at the expense of occasionally very short cycle times between matches). I also felt that the diversity of the Boilermaker schedule was responsible for a more accurate Top 8 at the end of qualification matches (completely subjective, of course).

I'd say the algorithm is better when used correctly, but honestly I think a lot of the problems in the past were also due to incorrect use (minimum match spacing set too high). I think that option really needs to be taken out of the control of the local regional.

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

I have to say that I am very pleased with how the new scheduling algorithm is working out. Through 18 qualifying matches in Portland and Seattle we saw and played with a wide range of partners and opponents and had matches that were reasonably well spaced.

As for starting first at the three regionals that Team 48 has been to so far, it would seem the chances of that happening randomly are about 0.3% Which sounds pretty unlikely.

On the other hand there are other patterns that would have similar significance to teams... always having the first match on Saturday, or always having the first match after lunch... always being scheduled for the last match of the qualifiers, etc. so that means that the chance of a team seeing a significant pattern to their matches across these three regionals increases to maybe a bit over 1%. Considering the number of teams at regionals, it may not be surprising that at least one of them has found a pattern to their matches.

This doesn't mean that 48 shouldn't be asking "is something weird here?" There are certainly grounds for them to ask. It merely means that if the answer comes back as "no", then no one should be too surprised. Luck... good, bad, or just weird (I would consider three Q1 matches good luck, as I like to have as many people as possible see our robot) does create funny patterns sometimes and we humans are very good at assigning significance to those patterns.

Jason

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

You may note that we have LOST each of those Match 1's in which we have played. Good luck, indeed. :D

You want our spot? ;) I'd rather have the extra prep time. To each, his own.

We're used to weird patterns this year, whether we are in control of their creation or not.

For instance, in addition to being in the first match at each event, we've finished as quarterfinalists at all three regionals this season, losing the first match, winning the 2nd, and losing the 3rd. :rolleyes:

We've also won 2 awards at each event, with the UL Industrial Safety Award being the common component.

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

I might suggest that the weirder co-incidence is Team 48 losing any three specific matches. You guys always put up a solid machine and have a very focussed team.

While I appreciate the desire for more pit time on Friday morning, there are actually 18 teams that have to be queued up before the opening ceremonies, so starting in Q1 doesn't actually reduce the pit time any more than starting in Q2 or Q3. I note that at Pittsburgh there were 36 teams, so fully half of the teams would have been queued prior to the start of the opening ceremonies.

And, to put it bluntly, "yeah, I want your spot". Although we've often had to queue up before opening ceremonies, we've only been in Q1 once as far as I can recall (last year in Portland) and put on a wild show in autonomous and nearly scored... before going on to lose the match. But I can understand... especially if you're working out bugs, how a later start would be appreciated.

Jason

Re: **FIRST EMAIL**/Updated Alliance Pairing Algorithm
 

For all of 330's years, we have felt really wierd on the rare occasions when we were not in Match 1 or 2. This began long before people started messing with the algorithm. In fact, NOT being in the first match was the only thing we liked about last years's pairings. Sometimes random things are just random, they only look like there is a pattern.

Before the match schedule is locked in, it can be "audited" to see how many different teams each team will play with and against. If there are "too many" repeats, then the algorithm can be run again. But it takes 10 or 15 minutes each run and the Field Management System can't be doing anything else during that time. If you have to run it more than twice, you can forget handing out match schedules at lunch time. In fact, the scorekeeper and FTA can forget lunch altogether.

There is a lot af variability in the performance depending on the constraints. If you have 48 teams and tell it you want 7 matches between times a team plays, then you only have 6 options for opponents and partners. A match schedule set up this way will have a lot of repeat teams. On the other hand, if you set it for 3 matches between times a team plays, then you have more sets of three opponents than you have matches. The algorithm is very sensitive to this and it takes some experimentation to get it right.

At the LA regional the fewest number of unique teams a team faced was 24, the maximum was 27, the average was somewhere between those two. ( The number of unique partners for every team was 18, the maximum except for surrogates, they had 30 oppponents and 20 partners). But we purposely set the minimum time between matches very low. I forget if we used 3 or 4. This guaranteed teams a minimum of 18 minutes between matches. Since a round took 8 matches to complete, the maximum time between matches was 90 minutes. This is typical of our experience under the pre-2007 algorithm. In actual practice I believe the average time between matches was more like 45 minutes, with very few, if any having turnaround times that short or long.

The best thing to do is set the minimum time between matches fairly low and let the algorithm sort out which schedule gives teams the most time between matches. It will probably pick a schedule within 5% of optimal.