[Lurpoa]

♥♥♥

Just finished Acbl Indy 5683: 4 rounds, and 4 round vs same opponent.
At least the Movement generator could have set him as my partner in the last round, but no....

This might be random ?????, but doesn't seem right to me...

Should be easy to avoid this in the soft .... no ?

♥♥♥

[barmar]

It's just random. I checked a few other players in that tourney, they mostly had different opponents in each round. You just got unlucky this time.

[gordontd]

Lurpoa, on 2012-August-19, 13:34, said:

4 rounds, and 4 round vs same opponent.

Just imagine how that opponent feels about it!

[Lurpoa]

barmar, on 2012-August-19, 18:48, said:

It's just random. I checked a few other players in that tourney, they mostly had different opponents in each round. You just got unlucky this time.

♥ ♥ ♥



Nothing to do with luck or bad luck.
What I wanted to say: this is not good in an individual tournament.
An individual movement is made to have a maximum of different opps and partners.
Such a movement is is NOT Random.
I suggest a correction to the software to take this problem into account.


♥ ♥ ♥

[barmar]

I just checked the code, it's totally random, no attempt is made to prevent replays or repartnerships. In a small game, it could be very difficult to avoid them.

If I've done my calculations correctly, in an 11-table game like the one Lurpoa was playing in, the chance that you'll get the same opponent in all 4 rounds is about 1 in 5,000 (4/43 to get a same opponent in round 2, then 2/43 for this one being repeated in round 3, and 2/43 for round 4). The chance that you'll get the same partner in all 4 rounds is about 1 in 80,000.

[Vampyr]

Why make it random instead of using a proper individual movement?

ANd how many boards in a "round"? Do these tournaments really use the same configuration for more than one board?

[barmar]

ACBL indies are 4 rounds of 3 boards, keeping the same configuration for all boards in the round.

I think the random movement was chosen because a "proper individual movement" wouldn't work well for small tournaments. I believe the standard individual movement is: North stationary, South up 1 table, East up 2 tables, West down 2 tables, boards down 1 table, with a skip in the middle when there are an even number of tables. But I think you need at least 5 tables for this to work.

When we have individual games in our club, we use the above movement, except that between boards in the round South, West, and East rotate clockwise, so you only play with the same partner for 1 hand. But you still have the above problem if there aren't enough tables.

[mycroft]

That "rainbow" movement only works with a prime number of tables (well, for 4 rounds of 3, relatively prime number of tables); they also are technically "4-winner" movements, because you're scoring against your line only on every board (different opponents, including CHO, but you're playing the same cards 1/4 of the players on every hand, and the other 3/4 of the players on no hand). They're good for a fun game, and they're easy to run FtF as no guide cards are required. Generic individual movements exist and should be able to be calculated on the fly. They have fairly "random" movement, but the computer can do guide cards much better than humans.

[barmar]

Can you provide a good reference?

I found this site with a collection of individual movements. But they're all for a fairly large number of boards (the smallest is for 20 boards), I don't know if they can be scaled down to 12 boards without losing the benefits. Also, there's no algorithms given, they're all just lists of who playes where in each round.

[Bbradley62]

barmar, on 2012-August-21, 09:30, said:

I think the random movement was chosen because a "proper individual movement" wouldn't work well for small tournaments.

It works well for everyone who ever has an 8- or 12-player house bridge game. You'd have to hard-code the movement for small tournaments and use the general movement once a threshhold is met.

[Vampyr]

barmar, on 2012-August-21, 09:30, said:

ACBL indies are 4 rounds of 3 boards, keeping the same configuration for all boards in the round.

...

When we have individual games in our club, we use the above movement, except that between boards in the round South, West, and East rotate clockwise, so you only play with the same partner for 1 hand. But you still have the above problem if there aren't enough tables.

Could you not at least move the players round the table, as above? If they are moved randomly (but always swapping partners), you could have a one-winner movement. This seems clearly so much better than keeping the players stationary for 3 boards that I must have missed something.

[mycroft]

Well, most of the movements are based on one set of boards. BBO individuals are simultaneous; we don't have to worry about player X meeting board Y again.

So, really, you could literally randomize players in each round (with a check for collisions) without it being a problem. Or Swiss it, which might be exciting.

Because the boards are all the same, and you don't care all that much about number of comparisons (being that it's 12 boards), you could use any published movement and truncate it after 4 rounds. The balance may be off, but it certainly couldn't be worse than the rainbow movement, which has 4 distinct lines of comparisons that don't mix.

One of the benefits of 3-board rounds are that you don't have the issue of people waiting for each board to be finished at every table before we move to the next one. You could do a rotate-around-North game in each round, and only at the end of the round do movements. Yes, that probably means that there's only one pairing that has to worry about robot-adjusted boards, but it would still work.

I think the only issue would be with small tournaments, where you are constrained by the players rather than the boardsets (which, as I said before, are not a constraint with BBO tournaments). How many of those do we have?

[Bbradley62]

mycroft, on 2012-August-21, 17:05, said:

I think the only issue would be with small tournaments, where you are constrained by the players rather than the boardsets (which, as I said before, are not a constraint with BBO tournaments). How many of those do we have?

There are 18 hourly ACBL Individuals per day. About 6 of these 18 (the first and last few each day) tend to be in the 3-6 tables range. The rest of the games tend to be 10+ tables.

[mycroft]

Okay, then that needs some more thought. I assumed they were more like the speedballs, with 30 tables. It looks like 10+ tables can be handled nicely through the rainbow method:

table 1's players in R1:
R2: N @ 1, S @ 2, E @ 3, W @ 4
R3: N @ 1, S @ 3, E @ 5, W @ 7
R4: N @ 1, S @ 4, E @ 7, W @ 10

This all assumes that everyone's playing the same boards in each round (I can't see why that wouldn't be). Feel free to rotate through each board in the round (and possibly rotate the players' initial seats by table - Tables 1, 5, ... play board 1 of the round in initial order, Tables 2, 6,... play board 1 of the round rotated once, and so on for the 3 and 4 mod 4 lines. I'd have to see that written out to make sure it's okay, but it should work).

Fewer than 10 would require some care, but a good start would be the first four rounds of any individual movement (would have to be 3 boards per partner, though, as there would be some table playbacks). You could go with the first 12 (or 6, with fewer players) rounds of the published individual, but that would mean people are waiting for other tables every round and would likely bog.

[Vampyr]

mycroft, on 2012-August-22, 09:29, said:

Fewer than 10 would require some care, but a good start would be the first four rounds of any individual movement (would have to be 3 boards per partner, though, as there would be some table playbacks). You could go with the first 12 (or 6, with fewer players) rounds of the published individual, but that would mean people are waiting for other tables every round and would likely bog.

I still don't get it -- if four people are staying at a table, why not change their positions each hand?

[barmar]

I've looked closer at the movement code, and I was a bit wrong in my earlier response.

There are two movements for indies, depending on whether it's clocked or unclocked. For an unclocked tourney, the movement is totally random. You can't use a predefined movement, because unclocked tourneys don't move everyone at the same time; they start the new round as soon as enough players are finished with the current round.

For clocked tourneys, which includes the ACBL tourneys, everyone moves at once, so a movement schedule can be used. Here's how it works:

After players get their initial seats, they're assigned player numbers. If T is the number of tables, Souths are numbered 1, 3, 5, 7, ..., T*2-1; Wests are numbered 2, 4, 6, 8, ..., T*2; Norths are numbered T*4, T*4-2, T*4-4, ..., T*2+2; and Easts are T*4-1, T*4-3, T*4-3, ..., T*2+1. When the round changes, 1 North stays stationary, all the rest of the players are moved into the next lower player's seat, skipping over the stationary player. However, if this were all it did, then at all but the first and last tables, opponents N&W would become opponents E&S, and the first and last tables would simply turn both partnerships into opponents. So what we then do is randomly shuffle EW pairs between tables (the new partnerships move together).

This has the following features: 1) You can never have the same partner twice in a row; 2) it works for any number of tables.
It has the misfeatures: 1) A partner can return later (after an even number of rounds); 2) you can have the same opponent several times.

Regarding moving players after each board in the round, I think that would be difficult. The server currently keeps track of player positions by round, not by board, and only calls the movement routine when the round changes. So instead of 4 rounds of 3 boards, we'd have to make it 12 rounds of 1 board. I think there's too much code that's dependent on the current per-round organization to make it feasible to change that.

[barmar]

mycroft, on 2012-August-22, 09:29, said:

Okay, then that needs some more thought. I assumed they were more like the speedballs, with 30 tables. It looks like 10+ tables can be handled nicely through the rainbow method:

table 1's players in R1:
R2: N @ 1, S @ 2, E @ 3, W @ 4
R3: N @ 1, S @ 3, E @ 5, W @ 7
R4: N @ 1, S @ 4, E @ 7, W @ 10

Doesn't this movement have the problem you mentioned earlier, that it's a 4-winner movement because everyone stays in the same direction?

And what if there are more than 4 rounds? Whatever we do should work for all tourneys, not just the way we have ACBL tourneys set up (well, we could set up different movement for them if we wanted, but I don't think it's a good idea).

[Vampyr]

barmar, on 2012-August-22, 18:48, said:

Regarding moving players after each board in the round, I think that would be difficult. The server currently keeps track of player positions by round, not by board, and only calls the movement routine when the round changes. So instead of 4 rounds of 3 boards, we'd have to make it 12 rounds of 1 board. I think there's too much code that's dependent on the current per-round organization to make it feasible to change that.

Wow, who would have thought the software is so feeble.

[mycroft]

mycroft, on 2012-August-22, 09:29, said:

Okay, then that needs some more thought. I assumed they were more like the speedballs, with 30 tables. It looks like 10+ tables can be handled nicely through the rainbow method:

table 1's players in R1:
R2: N @ 1, S @ 2, E @ 3, W @ 4
R3: N @ 1, S @ 3, E @ 5, W @ 7
R4: N @ 1, S @ 4, E @ 7, W @ 10

barmar, on 2012-August-22, 21:42, said:

Doesn't this movement have the problem you mentioned earlier, that it's a 4-winner movement because everyone stays in the same direction?

mycroft said:

This all assumes that everyone's playing the same boards in each round (I can't see why that wouldn't be). Feel free to rotate through each board in the round (and possibly rotate the players' initial seats by table - Tables 1, 5, ... play board 1 of the round in initial order, Tables 2, 6,... play board 1 of the round rotated once, and so on for the 3 and 4 mod 4 lines. I'd have to see that written out to make sure it's okay, but it should work).

barmar said:

And what if there are more than 4 rounds? Whatever we do should work for all tourneys, not just the way we have ACBL tourneys set up (well, we could set up different movement for them if we wanted, but I don't think it's a good idea).

Well, yes, that's an issue. This works for 3(r-1)+1 or more tables only. But it's trivial and simple. Again, you can use the first r rounds of a published individual movement, if the possibility of playing some people twice (either as both opps or as opponent and partner) - potentially 3 times - and not others isn't a big deal (and it could be).

[barmar]

The published movements are just tables. I was hoping for a general purpose algorithm.

Uday thinks the problem I was worried about re moving in the middle of the round may not be so bad, but he's not sure.