Re: "Random" match Schedules
 

Oh boy, hard numbers!

I wrote a C program to crunch the numbers of teams that teams played with or against at the first day of various regionals.

NASA/VCU Regional
New Jersey Regional
St Louis Regional
Interpretation
First column is how many games a team played.

Second column (|S|) is how many teams that team played either with or against. Obviously P and A are subsets of this.

Third column is how many teams a given team played AGAINST. So this is the size of the set that ALL their opponents were chosen from.

Fourth column is how many teams a given team played WITH. So this is the size of the set that ALL their alliance partners were chosen from.

You can see how a small size in either P or A might be problematic. If a team has a small pool to draw partners or opponents from, it is easy for that pool to be full of very good or very poor teams, thus throwing off that particular team's rankings.

Edit: See my post a few posts down for some stats from 2006. In general, teams did play with a wider variety back then.

Re: "Random" match Schedules
 

I agree with Travis - fix the problem or consider reducing the # of seeding matches to 3.

How is the strength/true seeding of a team is to be measured without variation of the opponents, the true strengths and weakness will not be validated via competition.

Weaker teams beating even weaker teams do not provide a true ranking. If that is what is intended (for any reason) - than just eliminate the seeding rankings and randomly select the #1 though #8 seeds after the pre-elimination matches. The resultant will just about match what is going to occur using the current method.

Besides, how many times does it take to demoralize the losing team, particularly if the robot capabilities and strategy isn't significantly changed from match to match.

Whats next? Requiring different drivers for every match?? After all, if the robot and strategy doesn't change from match to match - the only way to get any kind of variation is to swap out drivers, with varying capability.

I hope FIRST fixes the problem soon.

Mike

Re: "Random" match Schedules
 

And now, I have analyzed the schedule for the first day at NASA/VCU from 2006.
Let's just get to the point:
Conclusion (tentative, the algorithm might switch it up on day 2):
The set of teams a team could expect to play with or against at NASA/VCU last year was much larger. On average, a 2006 team could expect to play with or against 27 teams in the first day of the regional. In the 2007 regional, a team could expect to play with or against 18 teams.

However, the larger set of teams from 2006 was split up interestingly. A 2006team could expect to see their opponents each match come from a pool of 16.5 teams, while their allies came from a pool of 12.0 teams. In 2007, opponents come from a pool of 13.5 teams, and allies from from a pool of 13.5 teams. So a 2007 team actually has more variety in allies than in 2006, but much less variety in opponents.

And here is the raw stats from the first day of the NASA/VCU regional in 2006.

Re: "Random" match Schedules
 

We (1279) played against 1218 every match in NJ. (100%) Very non-random, and I hope this is corrected.

Re: "Random" match Schedules
 

I could see how this whole thing was a plan by FIRST to make the competition a little less harsh to newer teams. However, as a few people have already mentioned, team number is not a measure of how good a robot will be. Hence, the plan backfired.

I think the problem needs to be fixed for the coming week's regionals, if nothing else go back to last year's system.

Can ANYONE confirm or deny a plan to change the system for the coming weeks, has anyone been told anything?

Re: "Random" match Schedules
 

IMHO, a team's number should not have ANY impact on the seeding of the teams, as there are exceptions to the rule. Relying on team number alone is a very poor method of randomly seeding matches, as although there might be a slight trend towards lower number teams as better teams, "picking and choosing" your data to randomly sample a population creates biased results.

It's like asking everyone who's standing outside a building if they smoke, to try to ascertain the number of people who smoke in a company. Your data is obviously going to be biased.

So to solve this problem, we need to go back to elementary statistics and what they have to say about randomness. What we want is a match schedule that "mixes teams up", while still allowing for a minimum time span between matches. To do that, we will need a normally distributed match schedule.

Here's a quick sketch I made to illustrate the idea:



Of course I picked 1 hour as the mean and 15 minutes as the standard deviation, but that data can be changed as necessary. Increasing the mean would allow for greater separated matches, while increasing the standard deviation would also allow more "mixing up" of the matches, but at the sacrifice of some of the minimum timespan between matches.

And of course the real algorithm would have to take into account more variables than this, but this is just a concept. ;)

Re: "Random" match Schedules
 

we played 25 in at least half of our matches today. IMHO, i was very entertained by the eliminations at the regional, they were some very intense matches so i'm not complaining, but it seemed to me that they put a low number with a mid-ish number, and a rookie team/high number (<2000) I don't see how you can expect a fair and or "random" match from the same groups/sets of teams based on the teams number and not their robots specific ability/skill/efficiency this year

Re: "Random" match Schedules
 

Yeah I noticed that the matches were not quite random at NJ. My old team went up against another team all day Friday and Saturday. As far as Time in between matches, my old team went up every 10 - 12 matches which is probably almost 30 minutes (or more) if you count all of the field resets in between and 1 - 5 minor scoring table or robot technacalities - for the most part we were resetting the matches about 5 minutes every round (which isn't too bad) The Teams were very patient & I did not recieve any complaints about the matches not being random enough. I don't think the Match Generation System can be fixed while the software is out on the road traveling to about 4 - 5 regionals but I guess is FIRST has a private internet line I guess they (the Scoring Table) can d/l any updates to the System. The world may never know. :confused:

Re: "Random" match Schedules
 

Yet another MORT member heard from. (Sorry guys, but we did get hosed.)
I think I understand the intent of this system. The belief (misguided though it is) was that by pairing a rookie with an experienced team, the rookies would be able to advance, making the competitions more "fair" to rookie teams.
Now, this system is based on two assumptions. The first is that a team's age and its robot's abilities are inversely proportional. This isn't always true, as our rookie all-stars prove every year. The second assumption was that rookies deserve a better shot at winning than experienced teams. Coming from an experienced team, I may be biased, but with three "rookie-only" awards, do we also need to skew the qualification matches to "help" rookies?
Sorry if I'm stepping on any toes, but this looks a bit like they're stacking it to me. I'd much prefer true randomized pairings, and let everyone rise or fall on their own merit, not their perpetual teammate/opponent's.

Re: "Random" match Schedules
 

Yes, my comments in no way, shape, or form represent Team 118, in any fashion and are solely the views of me.

What I meant:

To explain what I mean is that I believe it is more fair for the rookies to kind of have a fair chance at the Finals by having them compete more with rookies rather than veterans. I think that the matches should be 60% Vet vs Vet||Rooks vs Rooks, and that the other 40% should be a combination of integration.

But for it to fit their algorithm of the time between matches they would have to be similar to this:

(Odd = Veteran Team; Even = Rookie Team)
Match 1: 1,3,5,7,9,11
Match 2: 2,4,6,8,10,12
Match 3: 13,15,17,19,21,23
Match 5: 14,16,18,20,22,24
(Few matches later)
Match 7: 1,2,13,4,8,10
et cetera.

That is just how I feel it would be more fair because that way rookies would have a more fair chance of getting to the top 8 seeds.

Now some of you say well, the vets will kill the rooks, but if you think about it, you will have usually 4 vets and 4 rookies or 5 vets and 3 rookies or even 3 vets and 5 rookies at the final 8 before they pick their teams and after they pick they will probably be around one or two veterans per team to (naturally) balance things out.

Pavan.

Re: "Random" match Schedules
 

While the idea of trying to give rookies a leg up is a really nice thing to do I'm not sure if it is really fair. If the system is setup so that there is adequate time in between the teams should be playing a random set of teams. If a rookie robot is good then it will seed high just like any other good bot. Messing with the deck and stacking it even for the weaker players doesn't help make people feel better about the results.

Re: "Random" match Schedules
 

Side comment:
Pavan, what I meant by that was that 118's robot(s) can go in any direction beautifully and it seems that your thoughts can as well. 'Diagonal step forward' is a great way to express a thought and sounds like 118 movement. :)
Sorry for the confusion.
Jane

Re: "Random" match Schedules
 

I recommend that someone write a short program which takes a look at the ranking after the "seeding rounds" take place. Look at the top 10 teams, see if they had repeat opponents, and note the final rank of those opponents.

Remember that rankings are very important to how alliances are chosen for the final rounds.

Does the data show a pattern?

I did this check by hand for the PNW Regional. (So my count may be slightly off.) There were 54 competitors.

Rank 1
- faced the same competitor 8 times who finished with a rank of 54

Rank 2
- faced the same competitor 4 times who finished with a rank of 32
- faced the same competitor 4 times who finished with a rank of 43

Rank 3
- faced the same competitor 8 times who finished with a rank of 52

Rank 4
- faced the same competitor 8 times who finished with a rank of 50

Rank 5
- faced the same competitor 4 times who finished with a rank of 18
- faced the same competitor 4 times who finished with a rank of 44

Rank 6
- faced the same competitor 8 times who finished with a rank of 51

Rank 7
- faced the same competitor 8 times who finished with a rank of 45

Rank 8
- faced the same competitor 8 times who finished with a rank of 46

Rank 9
- faced the same competitor 4 times who finished with a rank of 53
- faced the same competitor 4 times who finished with a rank of 33
- faced the same competitor 2 times who finished with a rank of 35
- faced the same competitor 2 times who finished with a rank of 49

Rank 10
- faced the same competitor 8 times who finished with a rank of 48

The PNW Regional data shows a pattern. Not to take away from every team who performed well, but is this really the "intended" outcome? Note that the rules clearly state the schedule cannot be changed after it is published.

Re: "Random" match Schedules
 

2108 played 2186 every single match. I fell sorry for them because we only lost one match. 2106 played 2107 the whole time too. We did well though and we ended up winning. I never really expected to win but we did work hard.

Re: "Random" match Schedules
 

I am going to have to be blunt about this. This must be fixed. Teams that work hard to field a robot and come to a competition will not accept this sort of scheduling, and should not. There are many highly skilled mentors involved in the FIRST program who understand random scheduling, while satisfying constraints, and who would happily turn out validated scheduling software for FIRST to use. Yes, I know that it is a hard problem, but it is a solvable one. The match scheduling should be based only on the index of the teams, (1-N) for the teams at the regional, should be random while satisfying the needed constraints on the schedule and should be blind to the team numbers. The index schedule for a given value of "N teams" at a regional need never change, only the task of randomly matching teams to the indicies needs to be done at the regional.