Best Movement?

[bglover]

I run a lot of small tourneys (6 5 4 or 3 tables) and am not sure how the swiss movement works and whether it would not be my best option.

 

Can someone give me a quick summary pleaase? I would obviously like to run it in such a way that the most pairs play vs. one another. I once clicked "swiss" with 4 pairs tables and it seemed to work best but it was honestly just an experiment and I didnt' learn anything from it.

 

So, assuming 6 tables, what would the # of rounds be, etc. (assume 12 boards for now) and how would it work? Any help is much apprecited.

 

Steve

[Dwingo]

I appeciate all the time & effort the Directors volunteer to make competitive bridge enjoyable to others. I like the short tourney versions, especially 8, 10 or 12 board versions.

 

However when the tourney gets really short, the shortest I have seen is 4, I request the Director to make this an MP game and not IMP.

 

Everyone knows the element of luck increases as the duration of the tournament decreases, and in IMP one result is enough to swing you to the 1st spot. These 4 board IMP tourneys are pure lottery in my opinion.

[uday]

None of the momevents guarantee that all the pairs will face each other.

 

Since you want 12 boards, and expect 3 tables, you should probably use

 

Clocked: 4 boards per round, 3 rounds

No point doing anything else, since the 3 NS will play the 3 EW and that is that.

You can make it 2 boards per round, 6 rounds, but that will just mean each EW pair will sweep thru the NS field once, then do it again. Not clear there is any gain.

 

 

Unclocked: 2 boards per round, 6 rounds for unclocked movements

Here, tables are matched up as they complete the previous round. So

while playbacks are possible, EW and NS are not constant so you will

probably see more combinations. If all pairs play at the same speed, 1x12 might work better (more rounds = more possible opps)

 

 

Swiss: Like unclocked. Since the order depends on score (each round, the pairs are scored. Then sorted by score. Then, 1 plays 2 at table 1, 2 plays 3 at table2 , 3 plays 4 at table 3. This movement does not mind issuing repeated playbacks.

If all pairs play at the same speed 1x12 *might** work beter (more rounds = more possible opps - i think. Does a swiss tend to clump the scores together or sread them out? Don't know.)

 

 

 

Survivor: You should not use Survivor for small games like this.

[inquiry]

I am not particularly fond of "survivor" events when a director is available. This has nothing to do with my being booted from an event... heck if I am so far out that I get clipped out, I would WANT TO GET zapped. But, rather, I thnk the fairest competition is when everyone gets a shot at playing against everyone esle. Of course, this isn't usually possible, but I think this is best.

 

This is why my favorite movement is a howell or at least a scrambled howell. The swiss events sort of approach this (kinda) as the directions people sit get scrambled. But in a 3 or 4 table movement, I would love to see a pure howell where everyone plays everyone else. And yes, I know this movement is not implimented on BBO... but such dreams some day may come true... :D

 

The problem with 1x12 events is you have 12 opportunities to sit and wait for the next round. I would much prefer 4X3 (only 3 wait periods) or 3x4 (with four wait periods). As much as I hate "playbacks" uncocked movements are nice as the delays are usually not long at all (unless you get stuck in the slow group movement dark hole, where the clock marches onward to minus 6 and minus 8 minutes each round and you sit waiting for the last slow table to finish)....

 

Ben

[uday]

If someone has an alorithm for implementing Howell movements, I'll look at implementing it.

 

What I need is logic that, either

 

1) given numtables, num rounds, tells me what the assignments should be for each round. Numtables will be anything from 2-40 or so

 

2) given numtables, numrounds, current round number, an EW pair and a NS Pair, tells me where they should go for the next round

 

 

About the unclocked: a few weeks ago we implemented code to alleviate the "slow group" problem ( for instance, we don't allow a new board to start in a round that is already slow) and the excessive playback problem. But this won't help if a table is slow to finish a board already in play.

[inquiry]

I am no expert on howell movements, and there maybe better ways than the sequence I propose for 3, 4, 5 and 6 tables below. Get much more than that, and of course no way to play that many boards on line (well, 1 board per round, but as I said in earlier post, I am not fond of 1x12 or 1x8 things). I

 

I think I got the following movement flow right, but maybe someone like richard who is mathematically inclined can help double check I am not sending everyone to the same table each round... :D

 

Here is the logic for a 3 table howell movement where all six pairs play against each other once. Starting with the "flow" logic, and then showing who is sitting at which table each round.

 

In a three table howell NS table 1 stays stationary

EW table 1 go to table 2 E/W

Table 2 N/S got to table 1 E/W

Table 2 E/W got to table 3 E/W

Table 3 N/S got to table 2 N/S

Table 3 E/W go to table 3 N/S

 

This give us the following five rounds...

 

TABLE NS EW = RD#1

1 6 1

2 2 5

3 3 4

 

TABLE NS EW = RD #2

1 6 2

2 3 1

3 4 5

 

TABLE NS EW = RD #3

1 6 3

2 4 2

3 5 1

 

TABLE NS EW = RD #4

1 6 4

2 5 3

3 1 2

 

TABLE NS EW = RD #5

1 6 5

2 1 4

3 2 3

 

Clearly, I think this is the fairest movement.

 

Next, logic for a 4 table howell... just the flow of the players after each round this time....

 

Table 1 N/S stationary entire time

E/W got to 3 E/W

 

Table 2. N/S move to 3 N/S

E/W move to 4 N/S

 

Table 3. N/S move to 1 E/W

E/W move to 2 E/W

 

Table 4. N/S move to 4 E/W

E/W move to 2 N/S

 

Next logic for 5 table howell

 

Table 1 N/S stationary entire time

E/W got to 2 N/S

 

Table 2. N/S move to 2 E/W

E/W move to 3 E/W

 

Table 3. N/S move to 4 E/W

E/W move to 5N/S

 

Table 4. N/S move to 5 E/W

E/W move to 1 E/W

 

Table 5. N/S move to table 4 N/S

E/W move to tavle 3 N/S

 

 

Next logic for 6 table howell

Table 1 N/S stationary entire time

E/W got to 2 N/S

 

Table 2. N/S move to 5 E/W

E/W move to 3 E/W

 

Table 3. N/S move to 5 N/S

E/W move to 6 E/W

 

Table 4. N/S move to 4 E/W

E/W move to 1 E/W

 

Table 5. N/S move to table 6 N/S

E/W move to tavle 3 N/S

 

Table 6. N/S move to 2 E/W

E/W moves to 4 N/S

[mink]

Hi Ben,

 

I suggested this type of movement here short after the tourneys started last year. I is not really Howell, though. In face to face bridge it is used for barometer tourneys where you have each board duplicated so that it can be played at all tables simultaneously.

 

As in online bridge a board is always played simultaneously at all tables, the movement could be used for any clocked tournament. I would really recommend to replace the existing clocked movement by this one, as it is nicer for small tourneys and has the additional benefit that the pairs do not all stay on NS or EW like today.

 

The only disadvantage this movement would have had in the past, that it would have been more difficult for the director to locate movies of previous hands, is gone with the latest beta.

 

Karl

[Posleda]

I support Ben and Karl very strongly! In addition I'll send to uday all Howell movements I have. It's not a hard work to implement it I think.

 

Dusan

[bglover]

Oh boy! Thanks in advance Uday if we can have Howells!

[mink]

Howell type movements, like all other movements in face to face bridge (except those for barometer tourneys), take care that not only you play against different opps in each round, but also that you play a set of boards that is not used by another table. But that is not what you want in online Bridge Brige, as there each board is played at all tables simultaneously. thereby preventing that you hear information about a board you will play later in the tourney in advance. So the movement in online bridge can be much simpler, like the one Ben presented in this thread. I consider it ideal for clocked tourneys.

 

Karl

[Posleda]

All you wrote is right, Karl, I agree. Nevertheless Howell can't damage anything and is much better balanced. There are couple of pairs In Bens movement which play only 2 times on different lines, e.g. Always better then Mitchell, I don't like Mitchell in small tournaments with one final ranking.

 

Dusan

[inquiry]

There are couple of pairs In Bens movement which play only 2 times on different lines,

 

If I somehow messed up the howell movement, I apologize. But I still like the concept of a howell for small groups, and scrambled howells for one person winners (rather than winner EW and winner NS). Surely someone knows how to put together howell movements.

 

A couple of people point out that howells control movement of boards and people, so that each table plays different hands. This is true, this is also true of the mitchell movement we are currently using in the tournments. But somehow they don't mention the same point, that all tables play the same boards on line at the same time... this is obviously a function of "duplciate boards" used in F2F where you would need the same number of sets of boards as rounds if you were going to play all the same hands at all the tables during a single round. Computers easily overcome this limitation.

 

Ben

[McBruce]

Here is an simple algorithm for producing Howells for online pairs games:

 

(1) Number the pairs from 1 to (number of pairs). If there is an odd number, add a "sitout" pair at the end.

 

(2) In round one, the matchups are:

 

— Highest numbered pair versus lowest numbered pair

— 2nd highest numbered pair vs 2nd lowest number pair

— 3rd highest numbered pair vs 3rd lowest numbered pair

 

...and so on. The sum of each first round matchup adds to (pairs + 1)

 

3) In round two, and all later rounds:

 

— the highest numbered pair stays where they are

— every other pair follows (takes the most recent spot of) the pair with the next lowest number, with the one exception that pair one follows the second highest numbered pair

 

Example: 7 table Howell (14 pairs)

Round 01: 14.vs.01 - 13 vs 02 - 12.vs.03 - 11 vs 04 - 10.vs.05 - 09 vs 06 - 08.vs.07

Round 02: 14 vs 02 - 01.vs.03 - 13 vs 04 - 12.vs.05 - 11 vs 06 - 10.vs.07 - 09 vs 08

Round 03: 14.vs.03 - 02 vs 04 - 01.vs.05 - 13 vs 06 - 12.vs.07 - 11 vs 08 - 10.vs.09

Round 04: 14 vs 04 - 03.vs.05 - 02 vs 06 - 01.vs.07 - 13 vs 08 - 12.vs.09 - 11 vs 10

Round 05: 14.vs.05 - 04 vs 06 - 03.vs.07 - 02 vs 08 - 01.vs.09 - 13 vs 10 - 12.vs.11

Round 06: 14 vs 06 - 05.vs.07 - 04 vs 08 - 03.vs.09 - 02 vs 10 - 01.vs.11 - 13 vs 12

Round 07: 14.vs.07 - 06 vs 08 - 05.vs.09 - 04 vs 10 - 03.vs.11 - 02 vs 12 - 01.vs.13

Round 08: 14 vs 08 - 07.vs.09 - 06 vs 10 - 05.vs.11 - 04 vs 12 - 03.vs.13 - 02 vs 01

Round 09: 14.vs.09 - 08 vs 10 - 07.vs.11 - 06 vs 12 - 05.vs.13 - 04 vs 01 - 03.vs.02

Round 10: 14 vs 10 - 09.vs.11 - 08 vs 12 - 07.vs.13 - 06 vs 01 - 05.vs.02 - 04 vs 03

Round 11: 14.vs.11 - 10 vs 12 - 09.vs.13 - 08 vs 01 - 07.vs.02 - 06 vs 03 - 05.vs.04

Round 12: 14 vs 12 - 11.vs.13 - 10 vs 01 - 09.vs.02 - 08 vs 03 - 07.vs.04 - 06 vs 05

Round 13: 14.vs.13 - 12 vs 01 - 11.vs.02 - 10 vs 03 - 09.vs.04 - 08 vs 05 - 07.vs.06

The other thing you might do would be to flip pair 14 from N-S to E-W, but otherwise this works for any even number of pairs (and for odd numbers you simply make the high pair the stationary pair.)

 

Posleda has pointed out that this means that some pairs play the same boards in the same direction often: for example, Pair 01 and 02 play only two of the thirteen rounds (rounds 02 and 08) in different seats. I'm not 100% sure, but I think the way around this might be to switch positions in a checkerboard pattern, where in odd numbered rounds the 1st 3rd 5th 7th matchups are flipped, and in even numbered rounds the 2nd 4th 6th 8th etc. are flipped. I added dots to the flipped matchups above. It may not be a perfect solution, but if it is a fairly good one and we can do Howells of all sizes with one simple algorithm, that is probably just what Uday wants. :angry:

[uday]

TY. I will look into replacing the standard clocked movement with this one.

Is there a similar simple movement for individuals?

[Posleda]

Here is an simple algorithm for producing Howells for online pairs games:

This is Bens movement in other words. For me it isn't Howell, only Howell-like movement by reason of unbalance (my previous reply). E.g. pairs 1 a 2 play once together, once on different lines and 11 times play the same hand. This is very near to Mitchell, from Howell is only the principle of moving pairs. But still better then Mitchell.

[Gerardo]

(Long post deleted on alternate algorithm, found equivalent to McBruce's, only different arranging. [taking from chess, where alternating colors (directions) is important.] His math is simpler.)

 

Taking a look at McBruce's example, where N is the number of rounds, R the round and T the table:

 

NS[R,T] = N+R-T+1 mod N if T => 2

NS[R,T] = N+1 if T=1

EW[R,T] = R+T-1 mod N

 

(maybe) flip NS and EW if R even

 

[use N instead of 0 in this context, i.e. add N when result is 0]

 

But, this does not scale, and is highly dependent of the number of rounds.

You need to have sections of number of rounds (+1 to make if even if necessary) pairs (half it for number of tables).

 

Posleda: take a closer look. I think in McBruce's example, all (but 14, which you can flip to get that), play 7 rounds in one direction, and 6 in another, being the direction facing 14 the difference.

[Posleda]

Is there a similar simple movement for individuals?

Yes, Rainbow. I'll send it to you with Howells.

[Gerardo]

Very nice description of Rainbow in The Rainbow Movement for Individual Duplicate Games

 

There is a caveat: sections have to have a prime number of tables. (This is needed to guarantee you'll never meet a player twice, as partner or opp)

 

 

Inside a section (in a given round), R round, T table, N size of section:

 

N[R,T] = T

W[R,T] = (T+(R-1)*2) mod N + N

S[R,T] = (T+(R+1)) mod N + N*2

E[R,T] = (T+(R+1)*2) mod N + N*3

 

Individuals could not be 1 board per round, because then you would never have a change of direction for the players.

 

If 2 boards/round, flip South and East.

 

If 3 boards/round, East goes West then South, West goes South then East, South goes East then West.

 

Nice movement!

[McBruce]

For individuals, a OAFA (one algorithm fits all) might be:

 

1) Count the players from 1, adding phantoms to get the total divisible by 4.

 

2) Seat the players for round one as follows:

— highest at table 1 South

— lowest at table 1 North

— 2nd highest at table 2 South

— 2nd lowest at table 2 North

— 3rd highest at table 3 South

 

etc. until you run out of North-South seats, then

 

— next highest player at highest numbered table East

— next lowest player at lowest numbered table West

 

...and so on. Note that the East-West pairs are placed in the reverse order, from highest numbered table down. Adding the numbers of each first-round pair should get you (players + 1), and this adding the numbers of the four players at each table should get you 2 * (players + 1).

 

3) At the end of each round, the highest numbered player stays where he is, and the others take the seat of the next-lowest player (player one takes the seat of the second highest numbered player)

 

4) As we've discovered, it may pay to flip directions in a checkerboard pattern to reduce the frequency of two entries playing the same direction for most of the boards. If you incorporate this into the movement it is probably a good idea to do it last or the progression might be difficult to follow.

 

Here is the 4-table indy complete (15 rounds). The dashes indicate the checkerboard pattern where the N/S and E/W pairs should be switched.

 

            1N 1S  1E 1W    2N 2S  2E 2W    3N 3S  3E 3W    4N 4S  4E 4W
Round 01:   01 16--08 09    02 15  07 10    03 14--06 11    04 13  05 12
Round 02:   02 16  09 10    03 01--08 11    04 15  07 12    05 14--06 13
Round 03:   03 16--10 11    04 02  09 12    05 01--08 13    06 15  07 14
Round 04:   04 16  11 12    05 03--10 13    06 02  09 14    07 01--08 15
Round 05:   05 16--12 13    06 04  11 14    07 03--10 15    08 02  09 01
Round 06:   06 16  13 14    07 05--12 15    08 04  11 01    09 03--10 02
Round 07:   07 16--14 15    08 06  13 01    09 05--12 02    10 04  11 03
Round 08:   08 16  15 01    09 07--14 02    10 06  13 03    11 05--12 04
Round 09:   09 16--01 02    10 08  15 03    11 07--14 04    12 06  13 05
Round 10:   10 16  02 03    11 09--01 04    12 08  15 05    13 07--14 06
Round 11:   11 16--03 04    12 10  02 05    13 09--01 06    14 08  15 07
Round 12:   12 16  04 05    13 11--03 06    14 10  02 07    15 09--01 08
Round 13:   13 16--05 06    14 12  04 07    15 11--03 08    01 10  02 09
Round 14:   14 16  06 07    15 13--05 08    01 12  04 09    02 11--03 10
Round 15:   15 16--07 08    01 14  06 09    02 13--05 10    03 12  04 11

 

Oh dear, follow #1 through the tournament and the poor fellow meets the same four or five people as opponents 90% of the time. Back to the drawing board... :unsure:

 

 

Obviously, this won't work for unclocked movements. I think for unclocked individuals the best idea long term would be to match players by speed, institute an optional penalty to your score for finishing the boards after a certain time, and enforce a hospitality break between boards or rounds for those who are finished very early (to reduce playbacks). I am trying these wrinkles out (by directing players, not by software) in my Alphabet Tournaments and I'll keep you posted on how well it works.

[Posleda]

Posleda: take a closer look. I think in McBruce's example, all (but 14, which you can flip to get that), play 7 rounds in one direction, and 6 in another, being the direction facing 14 the difference.

I don't know if I don't understand you or you don't understand me or both. I try to explain more precisely.

 

Round 01: 1 plays EW, 2 plays EW - same

Round 02: 1 plays NS, 2 plays EW - different

Round 03: 1 plays NS, 2 plays NS - same

Round 04: 1 plays NS, 2 plays NS - same

Round 05: 1 plays NS, 2 plays NS - same

Round 06: 1 plays NS, 2 plays NS - same

Round 07: 1 plays NS, 2 plays NS - same

Round 08: 1 and 2 play together

Round 09: 1 plays EW, 2 plays EW - same

Round 10: 1 plays EW, 2 plays EW - same

Round 11: 1 plays EW, 2 plays EW - same

Round 12: 1 plays EW, 2 plays EW - same

Round 13: 1 plays EW, 2 plays EW - same

 

1 and 2 play 11 times in same directions (NS, EW) and only once in different (round 8 is not taken in consideration). On the other side e.g. 1 and 7 play 11 times in different directions and once in same. This means unbalanced tournament.

 

If you use full Howell with initial position like 5-12 13-10 2-6 4-9 14-11 7-8 1-3, you may choose any two pairs and they will play 5 to 7 times in same direction and in different too. Tournament is balanced as much as possible.

 

That's the difference.

 

Dusan

[hrothgar]

Hi All

 

I'd like to throw out the following "offer".

 

From what I understand, the bible on bridge movements is a recent publication titled "Movements, A Fair Approach", co-authors Olof Hanner and Per Jannersten. English version by Barry Rigal.

 

Hanner is a professional mathematician and used to be the Chief Tournament director in Europe. His goal in producing the book was to create fair/balanced movements. From my perspective, the BEST course of action would be to contact Hanner directly and ask him to tackle the problem. I think that he probably has the best perspective on the issue.

 

If this proves impractical for some reason - for example, Hanner can't be tracked down or is unwilling to help - then I would be happy to buy a copy of the book and lend it to Fred or Uday or whomever would be doing the actual coding.

 

Let me know what folks think.

[Gerardo]

Dusan:

OK, didn't get your point.

Can you prove this (guess so, should be there by design)? Can you provide a few examples (didn't find much)? Few I saw, it had the same principle of N following N-1 (1 following the highest odd), so this properties depends on choosing the (a) right first round. Is this right?

If so, is there a way to find it? Maybe we should have first round hardcoded?

 

It is a strong condition, it limits the number of players (in a section), as it simply does not scale.

Also, it is impossible to meet in swiss pairs in proper time, and probably not at all.

 

Richard:

I'm interested, of course.

However, we should be able to specify what we look in a movement, even if that movement comes from <whoever>.

 

what conditions should a movement meet?

[Posleda]

Can you prove this (guess so, should be there by design)? Can you provide a few examples (didn't find much)? Few I saw, it had the same principle of N following N-1 (1 following the highest odd), so this properties depends on choosing the (a) right first round. Is this right?

If so, is there a way to find it? Maybe we should have first round hardcoded?

 

It is a strong condition, it limits the number of players (in a section), as it simply does not scale.

Also, it is impossible to meet in swiss pairs in proper time, and probably not at all.

The sample movement and all others I have sent to uday was taken from book, written by Andrzej Wachowski, well-known Polish organizer. He says the movements were chosen by computer (brute force I think). I believe him. I can send all movements to you too. There is no need to find other movements.

 

Yes, you are right: N following N-1 is a principle, initial position may change tournament's behaviour (balance).

 

The number of players is not limited. Besides Full Howell there are so called Short Howell od Three Quarter Howell movements. You can use them, when number of rounds is greater than half of players (if equals, you use Mitchell instead). In our example with 14 players you can play from 8 to 13 rounds. Initial position is always different, the less rounds, the more sitting pairs (near to Mitchell).

 

Swiss pairs is another cup of tea.

 

Dusan

[Posleda]

If this proves impractical for some reason - for example, Hanner can't be tracked down or is unwilling to help - then I would be happy to buy a copy of the book and lend it to Fred or Uday or whomever would be doing the actual coding.

 

Let me know what folks think.

Nobody can learn how to organize tournaments only from book without practice, I think. I am no Hanner or Jannersten, but small director from small country with hundreds offline tournaments on account. Some months ago I offered my help on the tournaments field, I am sorry to say without answer.

 

Dusan

[hrothgar]

Nobody can learn how to organize tournaments only from book without practice, I think. I am no Hanner or Jannersten, but small director from small country with hundreds offline tournaments on account. Some months ago I offered my help on the tournaments field, I am sorry to say without answer.

 

Dusan

Funny that. I see this as an example of an essentially "pure" statistics or signal processing problem. I'm unclear why significant amounts of practice with real world tournaments would be required.

 

Once the organizers have agreed to a basic set of design criteria, it simply remains to create an optimal algorithm.

 

Here is my take at a set of design criteria.

 

From my perspective, the primary design goal should be producing a "fair" movement subject to the constraint that the movement is broadly compatible with the number of boards that the organizers want to support. I am defining a fair movement as one in which each pair plays the same number of boards against every other pair in the tournament. I explictly note that I think that other criteria should be held subordinate to this goal.

 

The secondary design goal should be to guaruntee that pairs play a "reasonable" number of boards against one another. For all intents and purposes, a bridge tournament can be modelled as a sampling problem, with each board treated as an independent event. Basic stats suggests that the larger the number of samples, the closer that the sample mean will coverge on the population mean. I'm not sure what the "optimal" number of boards per round is, however, I lean towards three or four.

 

With this said and done...

 

As stage one, I would try to identify the set of "pure" movements that fulfill the selected designed criteria. Ignoring the trivial case in which the number of tables = 1, the simpliest case is when you have 4 pairs competing at two tables.

 

Here, we require 3 rounds to fully mesh the movement. We can create a balanced movement anytime that we are willing to run a tournament in which the total number of boards is an even multiple of 3.

 

In a similar fashion, if we have 6 pairs completing at 3 tables, we require 5 rounds to fully mesh the movement. We can create a balanced movement anytime that we run a tournament where the number of boards is an even multiple of 5.

 

When the number of pairs is relatively small, its fairly easy to create a balanced movement that will roughly accord with the organizers time/board constraints. However, these fully meshed movements start encountering problems when the number of players gets large. For example, suppose that we have 24 pairs, as is normal at an Abalucy event. In this case, supporting a fully meshed movement would require running a tournament where the number of boards was an even multiple of 23. While this movement successfully achieves our first criteria, it going to run into significant problems with criteria 2. No-one is going to bother sticking arround for a 69 board event to ensure that everyone plays an optimal number of boards against all possible opponents.

 

Thus we move to "Stage 2" of the our problem. Ideally, we should be able to partition any given tournament into a series of distinct sections, each of which will support an optimal number of rounds. For example, our set of 24 pairs can be partioned in three sets of eight pairs each. Each set can play 7 rounds of four boards each, requiring a total of 28 boards. Unfortunately, this type of movement doesn't allow us to compare results across sections - logically, you're running three separate tournaments each using the same set of boards. However, I strongly prefer this type of structure to one that "pretends" a false degree of accuracy.