pic: Sum of team numbers versus match number in Curie

[Drwurm]

294 had a brutal match schedule this year. first match against 118, then 67, then 33, then 100. I get a bit hazy after that. We were only ever paired with a high caliber team(16) once. We also only played one match against a team with a number higher than 1000.

[meaubry]

We were in Newton , (Team 47) and we also never played with or against any team higher than 1000.

All our matches involved teams like 233, 175, 39, 111 and
in 5 of the 7 we played against a team, then with them in the next match.

On a brighter note - the points piled up like crazy, and we ended up with the 2nd highest RS in Newton (over 70) - we seeded 43 of 85 with a 3-4 (W-L) record.

We had a minimum of an hour between matches - that usually grew as the matches fell behind schedule.

I guess with 85 teams, I have difficulty understanding why the chosen algorithem they chose was used. Was it to somehow level the playing field and force higher #'s up to the top seeding positions?

Mike

[Travis Hoffman]

Originally Posted by meaubry

We were in Newton , (Team 47) and we also never played with or against any team higher than 1000.

All our matches involved teams like 233, 175, 39, 111 and
in 5 of the 7 we played against a team, then with them in the next match.

On a brighter note - the points piled up like crazy, and we ended up with the 2nd highest RS in Newton (over 70) - we seeded 43 of 85 with a 3-4 (W-L) record.

We had a minimum of an hour between matches - that usually grew as the matches fell behind schedule.

Well gee, since my fellow Delphi brethren on 45 and 47 have chimed in, let me pile on.....

48 also played the majority of our matches on Galileo against teams numbered less than 1000, but we did manage to see a few +1000 teams. Of course, one of those teams above 1000 just happened to be 1717, a divisional finalist. They just happened to be paired with the ThunderChickens in that match - event champion. Fun times on D, tell ya what.

Partner Teams:

8, 40, 70, 84, 168, 121, 180, 226, 291, 425, 509, 812, 1540, 2599

Opposing Teams:

8, 25, 65, 84, 88, 115, 121, 134, 176, 195, 217, 343, 364, 384, 469, 494, 509, 1138, 1366, 1717, 2564

It would have been great if the talent were evenly balanced between partners and opponents, but that wasn't to be. By far, our opponent strength far exceeded the cumulative partner strength. It's fun checking out the match list on Thursday and having everyone simultaneously roll their eyes - here let me simulate this -->

We were also queued up in the 2nd match of the day both Friday and Saturday - while we broke the string of being queued in the first dang match at 3 supposedly independent regionals with 3 supposedly independent match lists, this was still an inconvenience - more eyes were rolled -->

Yet despite the 2-5 record (the OPR predictions suggested 3 wins), we still had a good time. But yeah, please fix the algorithm for the offseason events. Especially IRI. Pleeeeeeeeeeeze.

I guess with 85 teams, I have difficulty understanding why the chosen algorithem they chose was used. Was it to somehow level the playing field and force higher #'s up to the top seeding positions?

Mike

Interesting thought.....

[Joe Ross]

Originally Posted by meaubry

I guess with 85 teams, I have difficulty understanding why the chosen algorithem they chose was used. Was it to somehow level the playing field and force higher #'s up to the top seeding positions?

I was told that the field control system did not allow all the options to be passed to the MatchMaker software. The choices it did allow were decent for regionals, but not for 85 teams.

Unfortunately, this was discovered too late in the process to fix for this year, but should be easy to fix for next year.

[waialua359]

I really dont understand what's so hard to fix these kinds of problems. It was brought up last year, they fix one part, but not the rest.
With so many smart math people in the FIRST community, you would think these kinds of things could be avoided.
When people discuss a schedule being random, we dont have to get technical about it and everyone knows what we mean.
Excuses that because we want all the matches as equally timed apart is not good enough to have the scheduling the way it did and to have to wait year after year for slow gradual improvements.
I can honestly say this with conviction because it happened with some regionals and it didnt with others.
With only 7 matches to play after spending $$$$$ to get there, its in the best interest of everyone to maximize the opportunities for everyone to play with/against everyone as equally as possible while meeting other parameters.

The Hawaii regional match pairings alone is proof that this can be done for ALL regionals and championships.

[Ross340]

Originally Posted by Kris Verdeyen

IIRC, in one of the Technocat's alliances, there wasn't a team number over 67.

We were on the other side of that one.

bad news bears. haha

[Rick TYler]

How many of you have read the paper describing the mathematics and reasoning algorithm of this year's new match scheduling software? You can read it here: http://www.idleloop.com/matchmaker/. For the record, this software was written (for free) by two mentors from team 1318, and was meant to address the criticisms from past years. In fact Tom and Cathy wrote it because they despised some of the "we seem to have the same partners in every match" problems. It seems to have fixed that. What I'm guessing might be missing in the algorithm is the assumption that "team numbers determine robot quality." The algorithm is trying to minimize the number of times a given team sees another specific team as an opponent or alliance partner, to make sure the red/blue alliance experience is close to 50-50, to smooth out match separation (to assure no back-to-back matches), and what Cathy and Tom call "round uniformity." Making sure the team numbers are smoothly distributed was not a factor in the algorithm.

If you have suggestions why not email Tom and Cathy at the address on their Website? Adding a factor for "balanced mix of rookies and older teams" would probably not be impossible, and going to someone who can fix the problem would probably be more productive than complaining on CD. For what it's worth, the current match scheduler looks to me like it produces much better results than older solutions. I also think someone ought to thank Cathy and Tom for their contribution of this software.

[sgreco]

It appears to me that a lot of teams were probably in a matches with the same team(s) multiple times. Undoubtedly that would happen sometimes, but it seems like (based on the graph) that it happened a lot.

[Rick TYler]

Originally Posted by Sgreco27

It appears to me that a lot of teams were probably in a matches with the same team(s) multiple times. Undoubtedly that would happen sometimes, but it seems like (based on the graph) that it happened a lot.

Respectfully, I have not heard that complaint. What the current problem seems to be is "old teams are paired with old teams and rookies are paired with rookies."

[The Lucas]

Originally Posted by Rick TYler

How many of you have read the paper describing the mathematics and reasoning algorithm of this year's new match scheduling software? You can read it here: http://www.idleloop.com/matchmaker/. For the record, this software was written (for free) by two mentors from team 1318, and was meant to address the criticisms from past years. In fact Tom and Cathy wrote it because they despised some of the "we seem to have the same partners in every match" problems. It seems to have fixed that. What I'm guessing might be missing in the algorithm is the assumption that "team numbers determine robot quality." The algorithm is trying to minimize the number of times a given team sees another specific team as an opponent or alliance partner, to make sure the red/blue alliance experience is close to 50-50, to smooth out match separation (to assure no back-to-back matches), and what Cathy and Tom call "round uniformity." Making sure the team numbers are smoothly distributed was not a factor in the algorithm.

Again, can't stress this enough, everyone read the excellent white paper that describes the MatchMaker, especially if you are going to criticize. I also recommend downloading and playing with the free version generously provided. This version of the software is also very useful for find potential problems (probably one of the reasons it is available) as Joe did in this thread. The problems that make the scheduler less than optimal:

1. The Team List is fed into the algorithm in numerical order. This leads to the problem displayed above where the numerical team difference correlates with the probability of being in the same match. The problem is not the scheduling, but the ordering of the team list. It is easy to fix, after reading in the list, randomize it (equal probability of any team being in any slot) then fill it into the schedule.
2. The Algorithm allows you to choose a Minimum Match Separation (MMS) but most scorekeepers just use Default (I also heard the GUI selections for MMS are too limited for the scorekeepers). The default MMS is not optimal and limits the the algorithms ability to achieve its other goals (leads to more repeats than necessary). A more optimal default MMS need to be determined through testing. However, I believe current default MMS minus 2 is pretty close.

These are minor bugs in a first year program. Neither of them are fundamental flaws in the core of the algorithm, they are just poor inputs (similar to the old adage, Garbage in, Garbage out). I wouldn't be surprised if they are already fixed.

Originally Posted by Rick TYler

Adding a factor for "balanced mix of rookies and older teams" would probably not be impossible,

NO! Don't add this factor. This factor spawned the Algorithm of Doom in 07. This would cause the problem it the other direction (inverse correlation) making it more likely to pair the lower numbers and high. There should be no correlation between team number and paring probability. Fix the 2 minor problems, Default Match separation and Team list order. That's all, don't change anything else, compile it, use it for years.

Originally Posted by Rick TYler

I also think someone ought to thank Cathy and Tom for their contribution of this software.

Thank you Tom and Cathy for writing this software, documenting it and making it available to teams. The software was great this year and I'm sure it will be virtually perfect next year and on.

[waialua359]

Originally Posted by Rick TYler

Respectfully, I have not heard that complaint. What the current problem seems to be is "old teams are paired with old teams and rookies are paired with rookies."

Actually, I've seen matches many times where you were with teams in one match, and then against them the next. That should and could be fixed considering you had 85-86 teams and only 7 matches to play at CMP.

[The Lucas]

Originally Posted by waialua359

Actually, I've seen matches many times where you were with teams in one match, and then against them the next. That should and could be fixed considering you had 85-86 teams and only 7 matches to play at CMP.

I call those "home and home" matches (kinda like college football). The algorithm tries its best to avoid this but with the default minimum match separation constraint being set too high it is not possible. The default match separation for a division is 12 when there are a little over 14 matches in a round (86 teams / 6 bots per match = 14.33). That doesn't provide many options of separate matches to reschedule those 6 teams as you can see by the peaks and valleys in the graph. It is actually amazing that there where no repeat opponents (that I know of), which is a product of the 5 million schedules that "Best quality" generates (lower qualities do produce repeats).

Obviously, the next step in quality would be to generate all possible schedules, not just 5 million, and choose the best among them. That would take a very long time and would have to be done before hand for the sets of inputs. Then at the event, just randomize the input team list and plug into the perfect schedule.

[Mr B]

The graph is great in that it clearly shows what we all suspected as we looked at the match lists. Thanks for posting it.

I hope that the algorithm wasn't built upon the assertion that lower numbered teams means veteran and therefore better robots. Although that might be generally the case, all teams have up or down years. I'd hate to be a low numbered team that lost their corporate support (I know several) and had to face nothing but low numbered teams. By the same token, I wouldn't want to be a high numbered team that was hot but skated through qualifying because of easy matches. The method of match selection is WAY better than in '07, but it looks like it still needs a little tweaking.

It'd be great if FIRST provided something similar to the awesome Sundial program (which counts down the time till each team's next match). The improved efficiency might give the match algorithm a little more flexibility in the time between matches.

[Vikesrock]

Originally Posted by Mr B

I hope that the algorithm wasn't built upon the assertion that lower numbered teams means veteran and therefore better robots. Although that might be generally the case, all teams have up or down years. I'd hate to be a low numbered team that lost their corporate support (I know several) and had to face nothing but low numbered teams. By the same token, I wouldn't want to be a high numbered team that was hot but skated through qualifying because of easy matches. The method of match selection is WAY better than in '07, but it looks like it still needs a little tweaking.

The algorithm is definitely not built on such an assumption. As Rick mentioned above you can read all about the algorithm here:
http://www.idleloop.com/matchmaker/

This is a pretty good description of the software in mostly low-tech understandable terminology. It's really a must read for anyone commenting on the pairings.