"Random" Match List Generation
 

I've searched the forums for a discussion like this and found none. I'm curious to know if others have noticed this (and I'm sure many have) and what you think of it.

Our team attended the Wisconsin Regional a couple weekends ago and were very intrigued by the "randomness" of the alliance pairings in the quailfying matches. In addition to our own team's awareness of this, many other teams expressed that they noticed the same thing. As an example, our team played four matches in a row with team 525. We played against team 111 three times in a row (won 1 by some small miracle!) and team 111 played against team 494 four times! In our final three matches we either played with our against teams 1714 and 1732, both of whom played on the same alliance for the last three matches.

I'm not sure how the match lists are generated but sometimes it feels like there is a puppet-master working the strings. If the match lists truly are randomly generated then someone needs to work on the sophistication of their generator so there is a bit more mix to things. With 34 teams at the Wisconsin regional we should have played with and against every team there at least once. Instead we ended up playing with our against the same teams over and over again and never played with/against a handful of teams.

I don't want this to sound like sour grapes because it is anything but that. Regardless how many times we played against the WildStang juggernaut we still ended up seeded third and played respectably and we're proud of our performance. It would be nice to have the opportunity to play with some of these teams as well as against them. I can't imagine what it would be like for a rookie team to come in and find the deck "randomly" stacked against them.

Anyone else have thoughts on this or is this regional just a fluke in the pairings?

Thanks!

Sean

Re: "Random" Match List Generation
 

people have raised this issue before. Maybe its time to drive a stake through the random match generator and make something better.

You are right, during the seeding matches every team SHOULD play against and/or with all of the other teams at least once. So why do we assign matches at random?

Why not take the number of teams at the event, arrange a sliding schedule so the teams at the oppositie ends slide towards each other during the seeding rounds, and then assign the teams to those slots at random.

That way you wont have all the high number rookie teams at one end (together) and the low number veteran teams at the other (allied together). Something like a football pool - every team gets a slot, then those team slots are assigned to the sliding schedule at random?

The only way to verify the random-ness of the current system would be to collect all the data from each regional and do a statistical analysis. When you flip coins sometimes heads does come up 5 times in a row - if you are that team then it will not feel random, but the regional as a whole should average out to a random distribution.

maybe that kind of randomness is not what we really want? Maybe we want something more like the playoff tiers, except your team is not eliminated when you lose.

Re: "Random" Match List Generation
 

I don't believe the problem boils down to just the algorithm used. There's a few tunable parameters in the generation algorithm and the setting of these will greatly influence the randomness. For instance, one setting is the minimum number of matches a team must "sit out" between rounds that they play in. If this is set too high then the generator becomes very constrained on which teams it can place in any given match. This is especially noticable at events with a relatively small number of teams (such as Midwest and Wisconsin). If they had reduced the match spacing, it would have been much more random but at the expense of some really short breaks between matches. I know at Midwest for example that we played every 20 minutes. It was barely enough time to make a few quick repairs and get back in the queue. It would have been really tough if that number had been reduced even more.

Unfortunately I think there's a bigger problem here: when you have 6 teams on the field at once, and you only have about ~30 teams in attendance, you don't have a lot to work with. Maybe FIRST needs to consider a game where it can be either 2-on-2 or 3-on-3 depending on the number of teams at the event or something.

Re: "Random" Match List Generation
 

Another option would be to limit the number of matches that each team plays in a regional.

I'm guessing that the smaller regionals play a lot of matches.

Simply extend the time per match from 6 minutes to 8 or 9. Then crank the number of matches that a team sits out to a smaller number to increase the spread.

You'll have more time between rounds, but you'll play fewer matches. Which would you prefer?

Re: "Random" Match List Generation
 

That's a real double-edged sword, isn't it?! The nice thing about six teams on the field at a time is that all teams get more qualifying rounds which equates to more time to tweak an unruly robot and more time to play the game. For those teams that may only attend one or two events a year, more field time is a real bonus. Otherwise these teams are building a robot that sees action maybe a half dozen times.

The downside to more quailfying matches and more teams on the field, like you said, is the short duration between matches for any given team. We've dealt with short turnaround times in the past and it can be a real hassle, especially if things aren't going so well.

How about this for an off-the-wall game concept - not only random pairings but also a random number of robots on the field! Anywhere from 1-on-1 to 3-on-3! That would REALLY mix things up! ANd it would work for this years game! I think you're on to something Dave!

Re: "Random" Match List Generation
 

The problem really is too much randomness, not too little. What we really want is a defined tournament scheme, not a random scheme. Total randomness can create some surprisingly non-random looking results.

There are very well defined schemes for tournaments involving 4 teams/individuals. Bridge, for example. Given "X" players (x divisible by 4), a schedule can be set up so that you are partnered with each other player once, and against every other player twice. (Since at any one time, you have one partner and 2 opponents.) Or for round-robin leagues, where every team plays every other team once.

I don't know if such schemes are defined for a 3v3 competition.

The restrictions on not playing too many matches within a given time period also puts constraints on the selection process.

Re: "Random" Match List Generation
 

How about keeping the number of matches, but extending the day? Allow 8 or more minutes per match (instead of the usual 6 minutes) when the regional is small.

For example - MWR only had 60 total matches (12 per team for 30 teams). At regionals with 60+ teams (such as VCU), they were able to run 59 matches in the first day and 86 total matches. They did this by going from 9:20 AM to 4:12 PM at VCU versus 10:00 AM to 3:56 at MWR. So, if they ran the same schedule at MWR as VCU, they could slow down, still run 60 matches (12 per team) and not rush everyone on and off the field.

Re: "Random" Match List Generation
 

I like this idea, but it would have to be part of the game from day one (kickoff).

Some teams have built simple purely defensive robots that are in-capable of scoring. They did this because they know their bot will be valuable when matched with shooters and dumpers

but if you put them on the field alone they will automatically lose every match (because they cannot score).

Re: "Random" Match List Generation
 

Most chess tournaments of the size of a FIRST regional are paired with what is called a swiss system. In round 1, the players are seeded (by rating), and split into top and bottom halves. The #1 seed in the top half plays the top seed in the bottom half, the #2 seed plays the second highest seed in the bottom half, and so on and so forth. In subsequent rounds the players are divided into groups by score, seeded inside those groups by rating, and each group is split into top and bottom halves and paired as above. The system's goal is to pair players who are playing well against one another. A variation is the accelerated swiss pairing system.

While the system as used in chess tournaments is not clearly not viable for a FIRST competition, the concept behind it (pair strong players against one another) should be.

Re: "Random" Match List Generation
 

I don't ever think this match generator is going to be disseappearing anytime soon. Personally at the NYC regionals the matches were fairly random and so were the alliances. It made the event a little bit more exciting to see different alliances. Personally I like the randomness an event has. It gives teams a chance to change stratedgy to compensate for what they don't have. If anything the amount of robots on the field will be extremely more competitive, the only problem is the day would have to be extended or the time in between matches would have to be shortened so that each team can play all the scenarios.

Any thoughts ?

Re: "Random" Match List Generation
 

I haven't studied algorithms for generating matchlists. Superficially, it seems that repeat alliance partners and/or repeat opponents are more likely to occur

1) as the number of matches is increased -- if the event staff tries to schedule more matches while maintaining a larger minimum interval between appearances of the same team (say four matches off between field appearances) the result is that teams that have been on the field together once will share the field again, probably several times, and

2) as the number of teams is decreased -- for similar reasons.

To allow teams to share the field (as partners or opponents) with a wider variety of other teams, the match list must allow more variability in the interval between appearances of the same team.

It has been proposed here and in previous threads on this topic that, to prevent repeat field partners/opponents, the minimum interval between field appearances should be controlled in minutes rather than matches; i.e., increase the time interval between matches at smaller regionals, and allow more variance in the match interval between field appearances at larger ones. I think a statistical review of repeat field partners/opponents vs. appearance interval might support this proposal, but I don't have the time to do the analysis now. Any volunteers?

Re: "Random" Match List Generation
 

I dont see why a short interval between two matches for one team would be a problem. In the finals you may play three matches back to back, depending on your last match of the semi finals maybe even 4 or 5.

If you have 60 teams at a regional, then there will be an average spacing of 9 matches between each teams time on the field. If a fixed pattern was used (instead of psuedo random) then you might end up with two matches back to back, but then you would also have two matches with 18 or 19 in between - so it would balance out over the 3 days.

Re: "Random" Match List Generation
 

at the Midwest, us, 71, and another team (rookie), were on the same alliance in the same sides on the same positions. On the other alliance, only one changed, so basically the match was the same groups of people, except for one team. It was weird. Match 10, and then Match 15, and all the same bots but one were in line for Match 15.

Re: "Random" Match List Generation
 

I forgot how the NYC Regional Match list was generated. I think it was 12 matches The Start Time was 9:50 End Time was 4 something for the first day lunch was between 11 / 12 & 1, time in between matches i think was like 4 1/2 minutes - 6 1/2 minutes and the start times for saturday was 9:20 and end was somewhere around 11:30ish so that kinda helped in not seeing any alliance repeats. Like I said I was around when the matches were being generated but I was giving one copy one day and a totally different copy the following day. As I said before there should be a SET amount of matches to be played for each regional based on the amount of teams - like 10 matches for 30 teams so on and so forth that way those volunteers who are at the scoring table don't have to use their best judgement in how many matches to run.

Re: "Random" Match List Generation
 

Do you value your sanity? That is a lot of back and forth between the pits and the stands/field. Not to mention that there is very little time to just sit back and enjoy the surroundings.

At both Midwest and Wisconsin, I helped a few teams with their software. The changes were easy, the testing was easy, but there was a huge problem with timing. I'd be working with them and one of our matches would come up, so I'd run out to barely catch our robot beginning to move in auton. I'd return to their pit after the match to find that they were just about to go to a match. By the time they came back it was time for another one of our matches. It became a huge game of tag trying to coordinate time to work together. If we had an extra 2-3 matches worth of spacing, both of our lives would have been a lot less chaotic....they would have been able to play all of their matches with the changes that they needed, and I would have been able to watch the matches with my team.

Don't get me wrong, I really enjoyed helping these teams and watching their students go from down to excited, but I'll tell you what, I sure was exhausted by the end of the day.

I'm all for the extra spacing, even if it means having a 5 minute break every 10 matches. I think it benefits all parties involved.

Re: "Random" Match List Generation
 

I dont think that there should be a SET number, but a rough guideline. I know that at VCU, one of the biggest reasons we try to get as many matches in as possible is to give each team all the play time that they can possibly have. I think that those regionals with lower team attendance shouldn't necessarily take more time inbetween matches. I agree that there should be sufficient time inbetween a teams matches, since i was a driver for my team, but also try to maximize the number of matches that a team can play. Doing this first basically gives the teams what they pay for, but it also helps teams to be able to help out their rank, esp those teams that are able to improve their robot greatly if they came in not up to par shape.

Also just to sum up how the match pairing is generated:
Hatch Technology developed the scoring program that they use at the competitions. The part of the scoring program has the algorithm that generates the matches. The inputs that the competition puts in before it generates the matches are:
-Time inbetween matches for teams
-Ideal number of matches for each team to play
-and, Start and End time for the competition day
I think that is about it, not to sure, can't really remember that well. After these inputs are put in, the program takes them and the list of the teams that is stored in the program for that event and generates the match schedule.

just my .02

Mike O'Brien

Re: "Random" Match List Generation
 

Im not saying we should speed up the timing between all matches.

Right now the 'random match generator' is programmed to ensure all teams have at least 3 or 4 matches between games.

Im proposing we take the randomness out of match selection, and use a fixed pattern instead, that will insure all teams will play both with and against all the other teams in at least one match.

This means you might have two matches back to back maybe once at a regional, and the rest would be spaced out more, with at least one pair spaced out by 15 or 16 matches in between.

So for that one back to back pair you would also end up a with a pair having an hour or more in between.

and you would know all this on thursday morning, so you could plan any major changes when you know you will have enough time.

Re: "Random" Match List Generation
 

At AZ (45 teams), we were against the same team twice in a row. This happened at least twice for two different teams.

At LA (50 teams), that didn't happen.

The quick solution: add something that says "If team x plays against team y, then team y cannot play against team x again on that day, but may play with them." Then add something similar for playing with and you are all set--until you run out of teams or matches to fill.

Re: "Random" Match List Generation
 

From what I hear, this year, FIRST used a minumum of 5 matches before each team could repeat. This (as others have suspected) would severly constrain the "randomness" at small regionals, because if you do the math, if you have 30 teams at an event (like FLR) and there are 6 robots, thats only 5 matches total, so there isnt much room to move the robots around at all.

Personally, I agree with the 2-3 matches, 3 being what I think would be ideal. I can remember in the "old days" where we had potential to have every other match (ie you get off the field and jump right into the loading station). This to me, while hectic and hard if you are damaged, was cool because you did play against so many more teams.

Re: "Random" Match List Generation
 

I like that idea, but here's another one that I thought of while reading:

When TeamA plays in a match, they cannot play in the next match that any three robots from their previous match were in.
Adding that ensures that only two teams from their previous match are in the next match they're in. This may cause some problems at smaller regionals where there aren't as many teams, but I'm sure teams won't mind if the number of matches decrease between a team coming up decreases to 3 or 4 if it ensures that the matches will be a bit more randomly paired.

Those two robots may not also be on the same alliance configuration that they were in last match either.
If TeamA was playing against TeamB and TeamC, one of those two must be on TeamA's alliance in the next match. If TeamA played with both of those teams, one of them must also be playing against TeamA. If TeamA was playing with TeamB and against TeamC, then TeamC must play with TeamA, and TeamB must play against.

Re: "Random" Match List Generation
 

Just for fun, and only for fun:

Wouldn't it be fun to make your own pairing algorythem. We could take all the team lists from this year and put them in - see what you get. It will be harder than you think. And if you don't use the same one over and over, then people will freak out next year when 1975, 1999, and 2001 go up against 33, 71, and 494.

By the way - those are the rookies for next year.
1975 - my birthday year
1999 - Party like it's
2001 - Space Oddesy

If I had more time, I'd work on it for fun.

But I'm in China right now. (The students on my team already told me to start a team over here in the 2 weeks I'm here.) (No, I didn't) (But it would be easy - you can buy a CNC mill out of the back of an OX Cart in the middle of the street.) (It seems almost EVERYONE welds) (Enough ()'s)

Re: "Random" Match List Generation
 

The whole issue with the random generator is that some people were not informed of what needed to be changed. I was at a 30 team regional and we had really good randomization. The only problem was that you sometimes ended up playing 3 games in 17 minutes. I did not hear many complaints about time intervals and none on randomization. We set the one decider on 3 game interval minimum.

Re: "Random" Match List Generation
 

I disagree with you on this one and I know a couple of the other top teams will agree with me on this. We played at Pittsburgh with only 24 teams and saw far more randomness than at NYC. We wound up playing with teams 3+ times either allied or against, and others none. Sometimes allied twice and then agains once, or vice versa. There were some teams that we never played with on a field. This could be great for some teams, but horrible for others depending on the alliance strength. Furthermore, this throws the "Random Generator" Factor into seeding for some of the teams. Top performers should stay on top, but there were some non-scoring bots and sometimes non-driving bots that made the top 8 due to their alliance partners performance. The Master Puppeteer Theory is something we partially believe, our first 2 matches were against 375, NJ #2 Seed, and our Regional Championship Alliance partner 2 years in a row. I'm in agreement with the other engineers/mentors, at smaller Regionals space the matches out further, even if it means playing fewer matches we should play at least all of the teams once.
Best,
Wayne

Re: "Random" Match List Generation
 

random-ness doent mean that your team will play against many other teams - random-ness means it could go both ways, you could have different alliances and opponents on every match, or you could have the same alliance for EVERY match - both are possible when teams are selected at random. The odds against that may be very high, but people win the lottery every day against much higher odds.

Which is why Im saying I dont think we really want a random match selector - I would rather see a match assignment process where every team gets to play both with and against every other team at least once.

Re: "Random" Match List Generation
 

Wayne you can blame joshua who was behind the scenes putting all the figures in for the match schedule =).
I'll stand by my opinion that matches do need to be randominized and will settle on the fact the way FIRST and Hatch should do it should be changed. There have been some pretty good idea(s) on how this can be accomplished and when the end of the season thread opens up this should be one of the items brought up. Hopefully there will be more teams attending each regional so maybe there will be some more randominzation of a different way. And I also agree with Ken so he's not insane ;-)

Re: "Random" Match List Generation
 

Here's a thouht that many of you have already alluded to - how about NO randomness in match pairings? Like some of you have already said, why not have a set match rotation that ensures that each team at any given event plays both with and against every other team there?

Regardless of the time between matches (yes, I do value my sanity and prefer a little more time as well) or the number of matches played or the length of the day, I think giving teams the opportunity to play with and against all other teams will make for a more exciting event and will really allow the cream to rise to the top. A case in point is last year's Boilermaker where we had just plain bad luck and wound up seeded dead last. Our robot worked fine but we ended up playing against Beatty three times and never played with them. Had the alliance been the other way where we played WITH them three times, we would have possibly seeded higher. Fortunately we were selected for elims by team 135 who recognized our abilities despite our ranking.

Bottom line, I think you get a better feel for the capabilities of each team's robot and a more accurate indication of which teams stand out if there is a systematic pairing of teams that allows all to play with and against each other.

Sean

Re: "Random" Match List Generation
 

my team has also noticed that a newer team (past 3-4 years) will often get paired together and face an alliance with three veteran teams (older than 3 years) this happened a lot at Boston, we were often paired up with newer teams (no offense we were all good) but still we were no match for the likes of 233, 121, 40, 126, etc...what i am trying to say is: sometimes it seems that alliances are really not random, and that sometimes they favor the vetran teams, however:

this sort of applied to us this year, where we were ranked 40 out of 45 at boston, but 121 and 233 recognized us for our defensive skills and we helped win the regional

Re: "Random" Match List Generation
 

At Boilermaker, Team 1501 always played on the RED side during quialifying matches. Never in match on BLUE.

http://www2.usfirst.org/2006comp/Events/IN/matches.html

The first three matches we played, we played with team 461.
match 5, 10, 15

We also had several matches with 1018 as well.

We also had two matches in a row with team 829
match 34, 39

We never once got to play with team 135, which was the number 1 seed at Boilermaker. So we never found out if we were a good pair or not because we never got a match with them.

I thought this was very "strange".



-

Re: "Random" Match List Generation
 

I disagree completely, I broke teams up into new and old (I used 1200 as the division.

We were paired with 8 teams over 1200 and 10 teams under 1200

We were against 14 teams over 1200 and 13 teams under 1200

I would say that based on the age of the teams this was pretty even.

We were with 233 and never against them, we were against 121 but never with them (until eliminations) 40 and 126 we were against each once and with each once. I would say that those numbers wok out pretty well. It seems to me that elimination matches are randomn they just include some parameters such as minimum amount of time between matches that make it less likly for some teams to get paired up. The only way to fix this is to run a round then take a break and run another round, this however would take longer and you would get less matches. Other wise It could be done by predictably rotating through teams so it would be more even.

I have never had a problem with elimination pairings, you win some you lose some, in the end there are enough matches so the best teams come out on top.

Re: "Random" Match List Generation
 

There is one on on the volunteer staff to be blame. Please don't make statements that are not true.
The system automatically generated the list. I do not believe anyone on the NYC staff had anything to do with creating the match assignments.

On Friday morning, many of us did notice that certain teams were playing against or with the same group of teams several times in a roll. I brought the issue up with the Regional Director.
Also in a conversation later that day, the senior staff members discussed the possibilities of limiting the number of matches in small venues like NYC with 33 teams only.

Re: "Random" Match List Generation
 

Rich,

I don't believe mtaman02 was attempting to place blame on anyone. I noticed the smile after his statement and it appears to be an "inside joke".

From what I have read this is a very touchy subject to many folks in FIRST. My personal observations are that at the SBPLI regional which had 39 teams the match pairing were very odd. We noticed many teams (ourselves included) which were paired up with the same team 3 or 4 times during the seeding matches. While we did play 12 matches I would have settled for 11 if the parings could be better distributed. One of the interesting things that occurred was that the Friday seeding rounds were ended early because the organizers noticed that there were 2 teams that would have had back-to-back matches. To avoid this they ended the day a few matches early in between what would have been their "no turnaround" pairings and then had their next match as the first one up on Sat morning.

We also attended the Hartford regional 2 weeks earlier to scout some teams and renew old friendships. After looking over the match schedule there it seemed to follow the "normal" layout from the past where teams would be with a certain partner one match and then be against them at some point later in the tournament.

As far as the Hatch Technologies seeding software and the tournament/pairing generation it creates: I have my own opinions on the software but I would rather not post them in a public forum as I don't want to get off topic or create ill feelings.

One other observation; there are many post season competitions which have the same amount of teams in attendance that are present at many regionals. Many of these competitions have no problems whatsoever with pairings, scoring or field glitches. Some of them have generated their own software to insure they have complete control over the end product. A good example of this is BattleCry@WPI.

Re: "Random" Match List Generation
 

Sorry Rich the comment I made was inappropriate even though it was with the greatest intent to be me kidding around. In any case as I said before in my previous post hopefully with more teams joining up next year maybe we will see a different way on how everything will be randomized.

Re: "Random" Match List Generation
 

It was pretty random this week at West Michigan. Although match 70 and 71 was kind of different. We had to play a 2 matches in a row