You're right, Ogre, if 157 or any of the other Regional champion teams which were not signed up for Nats as of Monday DO get signed up, it will alter my predictions somewhat. All of the non-champion teams, however will still be in the same division. What will change is which division the champion teams will be in.

BruceM -- why don't you adopt a method like the one I used? Wouldn't it make the assignment process simpler and more fair to all at the same time? You've already got the data posted on FIRST's web site, that's where I got it. If you want to avoid the aggregation and the math, I can send you my spreadsheet with all the average QP's already calculated.

I have carefully avoided the responsibility of doing the assignments myself, leaving that to FIRST staff to do (and absorb the flak) However, you are welcome to send the spreadsheet to me (and a text file with the methodology) and I will be glad to share this with the folks doing the work. I think their results and yours would be similar..

I am considering adding this functionality in future years. It will be an interesting discussion to decide if it should be based purely on regional performance, or if there should be some weighting based on home city/state/country (either keep together, or spread around), if we should normalize based on some factor of total regional performance, etc. Sounds like a job for Tim Jump/Fred Rose's team in MN). Should divisional assignments be retained over successive years? Lots of cool idea here, but in the big picture, quite frankly, if the strongest teams get well distributed, the rest could be assigned randomly and still have a good chance of being strong competitors. With a sufficiently larger number of teams, we have the law of large numbers on our side.

pls send a copy of the spreadsheet to kenny@sigmacat.com
thanks.

Also, I'm wondering if their can be a check (someday) to evenly distribute goal-only teams and ball-only teams. Could you imagine if a "random" division existed where ALL of the robots were goal-only robots?

Also, someone asked about # of matches. Well, typically any given regional (or division) will play 120 matches (give or take). 120 X 4 / # teams = matches played.

So, EPCOT is slated for 292 teams / 4 divisions = 73 teams per division. 120 X 4 / 73 = 6.5 matches. So, maybe we'll all get to play in 7.

If all the stages move along like UTC, then maybe 8 or 9, but that'd be really pushing it...

(Bruce, pls correct if wrong)

KA-108 :cool:

p.s. glad to see you join the forum B.!

Re: Predicted Division Seedings
 

My simple question - what happened to the teams numbered between 400 and 800??

Actually, even better, there's something wrong with the way that page is displaying information (at least, here at school, and maybe at my house), 'cause it lists our team info. for team number 398.

Hmm.

Michael,
You're right, what's posted on the web site is messed up. I checked my original spreadsheet, and it's not the same. I'll ask our webmaster to fix it on Sunday. As an alternative, I've posted the original spreadsheet in the White Papers section.

THe CDI way...
 

Bruce,

At the Chief Delphi Invitational, we (I) had to come up with some way of picking teams for the seeding matches.

Basically, I violated some of the constraints that you (Bruce) put up in order to get a better (in my view) set of pairings.

One thing I did was relax the contraint that every team had to play their Nth match before another any other team could play their (N+1)th match. Without relaxing this contraint, I always got to the end of a round of matches and was stuck with some pretty crumby pairings.

Here is what I replaced that contraint with: No team could play their (N+2)th match before every team played their Nth match. This gave the program the freedom to have teams play well paired matches at the end of a round of matches (though it does make the end of a round a but fuzzier to see because other than at the beginning and end of the seeding, there is never a point when every team has played the same number of matches).

Here is how I would pick the pairing:

For NNN = 1 to Nmatches
[list=1][*]Find the set of all teams that have played the fewest number of matches*
[*]Randomly pick from the above set assign it as Team1
[*]Find the set of all teams that been Team1's partner the least often AND have not already played 2 more matches than Team1*
[*]Randomly pick from the above set assign it as Team2
[*]Find the set of all teams that have played against Team1 and Team2 least often AND have not already played 2 more matches than Team1*
[*]Randomly pick from the above set assign it as Team3
[*]Find the set of all teams that have played against Team1 and Team2 least often AND have partnered with Team3 least often AND have not already played 2 more matches than Team1*
[*]Randomly pick from the above set assign it as Team4
[/list=1]
Next NNN

*If you want you can put other constraints here such as "not having played in the last 3 matches"

This process does not ASSURE good seeding round pairings but it is pretty good.

What I do after the seeding round is set is to compute some metrics that let me judge whether it is a good set of matches or not. For example I compute the number of times a team plays with or against the same team. If there are too many teams playing with or against the same teams too often, I hit the "recalculate" button and compute another set of pairings. You can also compute other things like how often a team has matches too close or whatever other criteria you like and, again, just keep hitting the recalculate button until you get there pairing configuration you like.

Anyway, in 6 years of Chief Delphi Invitationals, I have not had too many complaints.

Joe J.

Joe --

I'll take a look at your scheduler in great detail after Chamionships. It looks very attractive. My previous attempt down this line was probably over-constrained. I was randomly selecting teams, but weighting them based on factors like how many times they have played. I like the method of allowing overlap between successive rounds. I don't suppose you have this implemented ;-) ? Can I peak at the code.

BTW, the software we are using to do scoring allows importing of schedules generated elsewhere, so it should be easy to use alternate scheduling for post season play..

BruceM

i am sorry, a little too lazy to read every reply.


in one sentence.....when will they announce the divisions



also, MOEs very own Bill Enslen has done some calculations to predict the divisions.(THEY ARE NOT OFFICIAL BUT BILL USED THE CALCULATIONG SYSTEM UED IN THE PAST)

MOE Predictions

K. -- You are correct in your calculation. It looks like 7 rounds would be a very good guess.

B.

Basically,
Step 1. (still in progress, as far as i know) All teams that won regionals have to decide if they can attend. Some teams have not paid yet and are in danger of not being allowed to compete.
Step 2. Divide up the teams
Step 3. Announce the divisions.

KA-108

Eric was saying there was difficulty in finding Green material for the Pit.... so probably White...maybe black. But ya never know.

KA-108

Coded but only for 24 teams, 2X2X2
 

Bruce,

I have implemented the code. I tried to make it general, but infact, I think I hardcoded a few places where I depended on either the number of matches being 24 or the number of teams being 24 or the matches being 2 on 2 on 2 (which was the last Chief Delphi Invitationals game format).

I think I have a 3 on 3 version as well, but again, I think I have hardcoded some items that should be variables.

It is not that I intentionally hard coded these options, I actually have variables defined for many of them, but I THINK that there are bugs that don't quite work right if the variable is changed. You know how it is, time was tight and I didn't have the time to beat on the code and find all the bugs for variable changes that I was not really planning on changing (at least that year).

We can talk off season about various random pairing strategies and what code I have or could generate or could get Nate to code for us.

Until then...

Joe J.

Re: Coded but only for 24 teams, 2X2X2
 

The offer is there, for ScoreFIRST(Bruce and I have already discussed this somewhat, and it looks like part of my code is making it into the nationals version of the system this year), CDI, and anyone else who needs custom scoring...ask anyone from OCCRA if you need a reference...

Any chance of changing 2002 code?
 

Hey Nate & Bruce,

If I/we wrote something to make random pairings for a general number of teams and a general number of matches, could we still get it inserted into the 2002 code?

If so, what would you need? Would it be enough to take in a list of X teams and put out a random, well-formed set of Y matches in a table?

I am game to try if it can still be put into practice this year. Otherwise, 2003 will have to be soon enough.

Joe J.

Sure, I'm game to try. There are a couple of options...

We could use a loosely coupled approach -- I have an "Import Schedule" function that allows an externally generated schedule to come in -- I'll send you the format when I sit down on that computer again. Your schedule could then be standalone.

Alternatively, I could directly couple it into the VBA code behind the rest of the scoring system. It would take the number of teams and the number of rounds, and would return the number of matches and a 5 column table with rows consisting of a match number, and a team index (1 to numTeams) for each of the 4 positions. In the case of an extra match, I indicate this by making the team index negative. I will also send the exact call when I sit down with the software again. My code then picks up the table, gets a randomly ordered list of teams, and assigns them to matches based on the index values.

So, did I understand right that at ChiefDelphi you have done alternative pairings (2x2x2, etc) and game rules? What are your plans for this year?

B.