I'm in a position to bring back this discussion...

http://www.rcgroups.com/forums/showthread.php?t=1514164

To bring this conversation back to the forefront, has anyone come up with anything better?

It almost seems that there should be a permanent "reasonable" matrix for each of the combinations of number on a team and number of groups...

It seems like a lot of work to actually create a brand new draw for every event. For instance, if you pre-create a matrix for 5 teams of 4 for 10 rounds, then you can store that, and if the event you have has 5 teams of 4, your system just uses that matrix, filling in the blanks...Just do that a bunch of times for the different possible event types and permanently store the matrices.

That's essentially how they do most other tournament draws. They just pick the initial random order, and follow the lines...

And yes, I realize that gliderscore does it already, and does it pretty good. I'm in a similar situation as jim was in that I've created an on-line scoring system, and this is the last piece...

Any comment?

Tim

This is also interesting as the J TS is coming up this year and my system may be used again. I have focused on scripts to load Gerry's MDB and get the data back out for my system. I think you are right that we could use his program to generate a complete set of matrix and reference them. This is for the simple assumptions that there are no frequency conflicts to deal with and we have team protection. In reality we never get more than 16 rounds in so a set of matrices for 16 rounds for teams of 4, 5, 6 pilots and numbers of teams between 3 and 14 would fill all of my possible needs.

It could take some work but I could use a script to create a set of inputs for gliderscore, run the matrix, export it and process it into a format suitable for storage to be used by another program to populate the matrix...

Hmmm brain churning - sounds like an exercise for the plane ride tonite...
Jim

Hey Jim,

Yeah, it kind of seems like a lot of extra programming cycles to be creating these matrices every time, for every event, no matter how small.

We should probably just develop a system to build out those matrices that become static for those permutations and then we can all just use those matrices when creating our events.

That way you can run 10,000 iterations if you want for one matrix, and once you're happy with the pilot to pilot distribution, you pick one and that becomes the defacto standard for that combination of teams and groups.

On a side note, I'm just about to launch my own scoring and database web site. I kind of developed it without really knowing about everyone else's existing stuff, so it does essentially the same thing as your site and gliderscore and f3kscore, etc...It just has them all in one place (i.e. it will do F3B, F3F, F3J, F3K and TD events)

Tim

GliderScore gives you evenly sized groups regardless of the number of pilots in each team and optimises the draw in all circumstances including taking into account team and frequency issues.

If all teams have the max pilots you will be fine with a default matrix.

But if, for example, you have a max team size of 6, and some of those teams have only 4 or 5 pilots then your idea of a fixed default matrix will end up with some groups full, and some short a few pilots. It won't be pretty and it certainly won't be optimised.

Gerry

Gerry,

How would that be different in the gliderscore case?

If you have say teams of 5, but occasionally a team has 4 or 3, won't you have groups that have one or two less pilots on occasion?

t

Originally Posted by ttraver

Gerry,

How would that be different in the gliderscore case?

If you have say teams of 5, but occasionally a team has 4 or 3, won't you have groups that have one or two less pilots on occasion?

t

No. GliderScore always has a maximum variation of 1 in group size. In any round, the earlier groups drawn are always the larger size, and if necessary to balance the numbers, one or more of the later groups drawn will be one less pilot in size.

so, actually, the answer is yes...occasionally you will have groups with one less pilot in them. The difference appears that if you have a group with two less, that you spread it out over two groups with each only having one less...

I'm not trying to bash your program Gerry. Its certainly in wide spread use and is awesome...

I see your point about a default matrix though, because depending on the team sizes, there might be a group that has 2 less pilots...

t

Originally Posted by ttraver

so, actually, the answer is yes...occasionally you will have groups with one less pilot in them. The difference appears that if you have a group with two less, that you spread it out over two groups with each only having one less...

I'm not trying to bash your program Gerry. Its certainly in wide spread use and is awesome...

I see your point about a default matrix though, because depending on the team sizes, there might be a group that has 2 less pilots...

t

The point really is that with a default matrix and teams that are incomplete, you will have groups with gaps in them. That is, no pilot where a pilot should or could be.

GliderScore does not do this. It fills every group as best it can.
You can't always have the same number of pilots in every group because the numbers don't allow it. So you get groups that are different is size by one pilot. That's not the same as having a gap. It just happens to be as close as you can get to having the same number of pilots in each group.

I think what we are saying here is that the unfilled slots in a team are considered BYE pilots. You do not want a bunch of the BYEs to line up in a group - shorting the pilot count for the group. It is a good point. The static matrix only works well for events that are mostly full.

I knew that seemed to easy...

Jim

couldn't you then even have statics for unfilled slots as well?

For instance, you could have a static matrix for 4 teams of 4 and 1 team of 3, or 3 teams of 4 and one team of 3, etc...

t

I think this is not really worth it. Assuming that you need a separate matrix based on the number of teams you have to get the statistics right then for an event with 4 pilots per team and 14 teams and never less than 4 teams, you have 396 permutations of the matrix. Add in matrixes for 5 and 6 member teams and you have 1485 permutations.

If the number of rounds affects the actual matrix generation then this number jumps dramatically higher. For me it is easier to use Jerry's program by exporting my pilot data in a format that is usable in his DB, then running the matrix and then exporting the data in a format I can import back into my system.

I also don't agonize over this too much. For club events with less than 20 pilots I just do an easy random selection. You don't get that much flexibility in the matrix with smaller numbers so it does not seem to me to make a lot of difference. I just do this for the bigger events.

Jim

Originally Posted by jimsoars

I also don't agonize over this too much. For club events with less than 20 pilots I just do an easy random selection. You don't get that much flexibility in the matrix with smaller numbers so it does not seem to me to make a lot of difference. I just do this for the bigger events.

Would have to agree with you, Jim, after running so many matrices over the years ... sometimes the random draw, even picking golf balls from a bucket works as well or better than the coded solution

Originally Posted by tkallev

Would have to agree with you, Jim, after running so many matrices over the years ... sometimes the random draw, even picking golf balls from a bucket works as well or better than the coded solution

If that is true then the coded solution is particularly poor.
A structured draw produced by GliderScore will always be superior to a random draw regardless of contest size.
And random draws can produce awful results. For small contests it is better to manually attempt to replicate a coded structured draw than to use a random draw.

Originally Posted by Gerry.Carter

If that is true then the coded solution is particularly poor.

This needs a large ROFLMAO smiley to go with it.

TK, I'd be firing your code writer...

We have found Gerrys program to be really quite nice in running contests here. A laptop and a printer that runs on a 12V inverter, and one can show up on the field, sort out the contest entries, make a rational non-random draw, print out score sheets, and run the comp.

BTW, I hate flying against one guy 90% of the comp and not flying against another guy for the full comp! This tends to occur repeatedly with the random draw concept.

BTW#2 4 Gerry I think that the NZ soaring SIG may have a small token to deliver to you at the upcoming Jerilderie comp...

A Hobbit?