Re: "Random" match Schedules
 

I think that in defining the "random" algorithm somebody (the GDC, 4FX?) was trying to get qualification matches where the opposing sides are relatively equally matched. This should, in theory, lead to more exciting qualification matches.

AYSO (American Youth Soccer Organization) years ago started ranking players and then assigning them to teams at random. There ceased to be a "draft". Coaches are just given a list of names. At the end of the season coaches are asked to rank their players as A-E. The next season these rankings are used to create ballanced teams. Each team will have one A player, a couple Bs, etc. The new coach is not told of the rankings, but a good one will figure it out pretty quick. This system works pretty well for producing balanced teams. In fact, the winningest coaches are the ones that improve their players skills the most. The ones that turn their C players into C+ or B players.

But robot teams are not soccer players. Soccer players tend to stay at pretty much the same level from year to year. They might move a grade or so in one direction or the other but not huge changes in performance. On the other hand many robot teams fluctuate wildly in their performance from year to year.

While there is a slight correlation between team age and performance on the field, it is not close enough to use to achieve the presumed goal. So I hope this turns out to be another experiment in Utopia (a la 2001) that is quickly abandoned.

Now to relate things from the match generator's side of things.

The match generator has about the same idea of what is happening that you do. In fact if you are reading these threads and he isn't you have far more idea. There are two "buttons" used on the screen to generate matches. One is the Week 1 algorithm and the other is the "new and improved" version. If the operator is not careful or is ignorant of the difference it would be easy to pick the wrong button.

Now let's look at the SoCal Regional. We had 52 teams which break up into two groups of 17 teams and one of 18. Each team had 8 matches. So each team would have played with at most 16 teams in "other" groups and 8 teams in the same group. If you put 17 marked beans into a sack, pull one out, record which one it was, throw it back in and repeat fifteen times, chances are pretty darn good that you will get a repeat or two.

The software does have means to check the number of "repeat" teams. At SoCal the most repeats was two teams playing each other 3 times. At least that's what the auditing software told us and we believed it. Most teams had two repeats, but we figured that was acceptable considering the situation. We did not think to check for time between matches before publishing the schedule. I'll try to do better in Davis, but the worst was one team that only had a one match break. I'm not sure we would have regenerated the match list even if we knewbecause it was so good in other ways. You can't save a list to se if you get a better one or not.

The auditing tab is not obvious and is a step that some events might have missed. But honestly, even having the same team from another group in four matches would not have been unreasonable given the pool size. Having the same team four times from the same group probably means the wrong algorithm was used.

So somebody made a mistake doing something on which they might or might not have recieved proper instruction. This was a last minute change and was not in the documentation on the system. Oh by the way they were probably using vacation time to be there and probably spent numerous hours before the event getting ready, learning the scoring system etc. So be gracious and thank them for their efforts.

Re: "Random" match Schedules
 

But if the person generating match lists has a bias or conflict of interest, then they could keep going till they got one they liked. A process that is to be repeated until "fair" can be perverted at will. The more likely event is that it will be perverted by someone who’s not even aware they are doing it. From what you say, there will always be something about one list or another that they do not like. But, from what you say, they will know a good list when they see it. :(

Re: "Random" match Schedules
 

By "trying again until you get a better list" I meant getting a list that doesn't have more than two or three repeats for any team and that finishes rounds about the time you want.

Due to the randomness you get more or fewer matches sometimes due to the "extra" teams when the number of teams is not divisible by six. So your ending schedule bounces around a bit with different runs. It took us a while to figure out why that was.

Would it be possible for somebody to mess around with the schedule so they get an "easy" schedule while somebody else gets a "hard" one? I suppose.

We got the final team list around 3:30pm and needed to make the schedule by 5pm to allow for duplication. It took several tries to meet our operating parameters for schedule repeats. Continual regeneration to find a "soft" schedule for one team? It would be easier to get one close and edit it by hand, which can be done as well, though it isn't easy.

Another point, the people selected for this job are people who have been around FIRST for a while. People who have a track record for being fair and even handed. If they had noticable bias, then they wouldn't be there

In SoCal there were three of us, two from different teams and another volunteer who just does the regional who were present during the match generation. The only checks we made on the schedule were on the number of "repeats", and the start times for the final matches each day, which the program gives without revealing what teams are in what matches. A master was then printed and given to the Regional Director for duplication. If we'd messed with the schedule after that it would have been pretty obvious.

So while messing with the schedule is a possibility it is fairly improbable. But now that you mention it, it is probably best for multiple people with different loyalties to be involved in the process. One man can be bought, it is much riskier to buy two or three.

Re: "Random" match Schedules
 

I added an analysis of the LA regional to my site.

Re: "Random" match Schedules
 

Tom,

Thank you for your posting. It looks like the auditing software may need some auditing. That 4+ column was the one we were watching and it was empty on our screen. Either that or we have an IO (Idiot Operator:rolleyes: ) error.

At least only four teams were involved and one of the "repeat" matches was an "extra" match and didn't count for 968. By the way we had 52 teams and only 16 teams wih "extra" matches.

We tried.

Re: "Random" match Schedules
 

Actually, of the week two regionals I've looked at, I think LA had the best schedule. It's too bad it takes so much effort on the part of the regional organizers to weed out bad "random" schedules. The folks who graciously donate their time to run regionals shouldn't have to do work that could be done automatically.

On a related topic, I do have a question about surrogate teams that a regional organizer might be able to answer. From reading the "Tournament" part of the manual (section 9.3.2), it sounds to me like there should only be surrogate teams in the last qualifier match to round out the final match, so I would expect at most 5 surrogate teams. However, all of the schedules that I've looked at for regionals when the teams aren't a multiple of 6 show many more than this, sometimes more than half of the teams. It looks like the scheduling algorithm pads out every round, rather than having matches that mix the last teams getting their first match with the first teams getting their second match, etc.

This seems odd to me, to have so many extra matches, which also make keeping a minimum gap between successive matches more difficult. It doesn't seem bad to me to put the last robots getting their first round against the the first robots getting their second round. All those teams get to scout and make robot improvements during the previous matches.

Is there some reason why it's bad to mix rounds slightly at the edges, assuming the schedule does a good job of giving teams time between matches?

Is there official FIRST policy about this beyond what's in the manual? If every round is padded, how do teams know when they are playing a match as a surrogate? Is it just their last match that doesn't count? If that that's the case, then padding all of the rounds doesn't avoid round-mixing and just adds extra matches that don't count.

I'd like to add this to my scheduling algorithm and want to get it right.

Re: "Random" match Schedules
 

The last round was the only one with surrogates. For the surrogate teams their "throw out" round is always the last match. It is indicated on the schedule if you know what to look for but it is not obvious. Even so the teams I mentioned it to seemed to know about it. Because our "extra" teams were all in the "rookie" group we only had one team on a side that was playing "for real" in the last few matches.

Re: "Random" match Schedules
 

I was just checking up to see if the scheduling issue with the extra surrogates was resolved*. I didn't find any more instances of that problem, which is good.

What I did find, however, was that the team standings from Florida indicate that all teams except 1875 and 1694 played 8 matches. According to the team standings, 1694 played 7, and 1875 played 6. Now, of course, I checked the match results to verify this, and in fact, those teams are listed as playing 8 times each. So what happened? Is the software is interpreting a DQ/no-show as not having played the match (which could affect the rankings, depending on how it's handled internally)? Since I don't know the unpenalized scores, I can't deduce if the ranking score is the result of 8 matches, or of fewer.

Similarly, at GLR, 226 and 519 are both listed as having played 7, when 8 was the norm. 519 actually played 8, according to the match list, but 226 played 9. Does anyone have statistics from GLR to cross-reference against the published standings? Also, with 59 teams, we should be seeing 2 surrogates; who were they?

And in Brazil, 1621 is listed for 9, but played 10. Also, though it's due to an oversight in the rules, technically there should have been an all-surrogate match played, because the number of teams (15) is indivisible by 6, but the product of the number of teams and the number of matches per team (15 × 10 = 150) is divisible by 6. (A better solution would be to correct 9.3.2 to eliminate the useless match that is called for in this situation.)

In L.A., 8 matches was the norm; 702 is listed for 7, but played 9. Also, who were the 4 surrogates?

In Wisconsin, 8 matches was the norm; 1525 and 1103 are listed for 7, but played 9. And there would have been 4 surrogates.

In Louisiana, 11 matches was the norm; 462 was listed for 10 but played 12, and 2190 was listed for 10 but played 11. I presume that 462 was the lone necessary surrogate.

Maybe all of this is due to errors creeping into the match data after the fact, or maybe it's due to a non-intuitive way of displaying disqualifications. It's probably worth a look, just to make sure that the teams with strange numbers of matches aren't having their rankings miscalculated.

*The published match results at St. Louis indicated that several extra matches were played with all 6 teams being surrogates, in excess of the number required by 9.3.2. And 9.3.2 is itself excessive in cases where the number of teams is indivisible by 6, but the product of the number of teams and the number of matches is divisible by 6—like the 45 teams playing 8 times each at St. Louis; it calls for surrogates, even when they aren't required to construct a self-consistent schedule.

Re: "Random" match Schedules
 

With 52 teams playing 8 rounds, you only need one match at the end with four surrogate teams completing the last round for two teams. But the LA schedule had the last four matches, each with two scored teams plus four surrogate teams. I can't understand how that makes sense!

I'm afraid to see what was happening at the regionals with half of the teams playing surrogate rounds...

Re: "Random" match Schedules
 

Well said Alan!
And another point I would like to add, besides advantages and disadvantages to match play.......our team didn't spend all those weeks with blood sweat and tear, just so that we could play the same teams over and over again. Why have a regional with 60 teams if we only see less than 10% of the field. Shouldnt we play with as many different teams as possible?

Re: "Random" match Schedules
 

Having been to GLR last week, I am really anxious to see what FIRST has in store for the Boilermaker and all other 3rd week regionals. All I can say is good luck to everyone and may the best robot/strategy/driver/team/alliance come out on top

Court

Re: "Random" match Schedules
 

The extra surrogate teams seem to be a function of the A/B/C pooling scheme. At GLR the pools were 19 teams in A, and 20 each in B and C.

(59*8)/6 = 78 2/3, indicating that all teams could get their 8 matches in 79 games, with 2 surrogates. Instead they played 80 matches, with 8 surrogates. These were teams 57, 65, 67, 201, 226, 308, 313 and 406, all from Pool A.

I believe "played" in the standings is less than the number of times the team appears in the match schedule because of no-shows (not even a human player) to the field, or because of DQ's.

Re: "Random" match Schedules
 

I've added a Beta release of my scheduling program to our web site. It's a console app so you have to run it from the command line. There are versions for Windows and Mac OS X. To get the beta, scroll to the bottom of the white paper.

Re: "Random" match Schedules
 

Some of this "random-ness" didn't seem to be occuring at Midwest. 1525 never played 71, and a few other teams, and was matched against Motorola 111 3 times, and played with 858 3 times, and against 858 once. There wasn't any real problems with the time between matches, but we just didn't play quite a few teams on the field.

Re: "Random" match Schedules
 

At SVR it appeared that the A/B/C segregation was still in effect, as 190 never played with 8, 100, 114, 115 or 254. 971 also played 3 rounds against 254 (none with) and 3 rounds with 1970 (none against). Overall, however, the pairings seemed a lot better than it had been