Pairing Problem

Running Chess Tournaments
21

You are correct that if one did the “transposition” in the context of the question “Should Charlie be paired with Alpha instead of Baker?”, that transposition is within the 80 point rule since swapping out Baker and swapping in Charlie is less than an 80 point swap.

But that isn’t exactly what I meant. (I might not have said what I meant.) What I meant was that this is not how SwissSys actually does it, as far as I can see. It’s algorithm seems to be “top down”. At the point where the program is pairing Alpha, it is not considering Charlie yet. Colors aside, Baker is the natural pairing for Alpha. The program also isn’t considering colors yet. So colors are, in fact, aside. Colors come later. So, the program pairs Alpha with Baker, Charlie with Foxtrot, Delta with Echo, Golf with Hotel, etc until it has finished what it refers to as the “natural pairings”. (There were actually 10 players in the section in all),

Then it looks for transpositions and interchanges that fall within the 80/20 rules, etc, to improve color allocation (and, for all I know, “drops”). However, when you look for transpositions at that point, it is already too late. You can’t just do a straight transposition of Baker with Charlie because by then Charlie is paired with Foxtrot. Baker has already played Foxtrot. And, apparently there is not any other sequence of two or more transpositions (in which every transposition is an improvement) which gets you back to Alpha v Charlie, Baker v Foxtrot, either. So, the program never evaluates the “transposition” of Baker with Charlie, vs Alpha.

Should we say that its algorithm is “wrong” and produces “wrong” results?

22

Brian,

While it is uninformed speculation on my part (since I am not familiar with the SwissSys source code), I suspect that you are correct that once SwissSys paired Alpha and Bravo, it took those two players “off the table,” considering that a solved problem, and then marched on to pair the remainder of the 2.5 score group in the only way it could (Charlie vs. Foxtrot). I’ve seen other situations where going back and undoing an already “committed” pairing would have improved the overall result, but SwissSys appears not to do so.

23

I think the simulated annealing algorithm which WinTD reportedly uses sounds rather clever, and probably produces better pairings than SwissSys. This does not involve backtracking or look-ahead directly, but the idea of simulated annealing is that by being open to changes from the baseline at the beginning (including changes that initially look worse), you can find regions of the search space which would be inaccessible if you followed a “greedy” approach where only improvements from the baseline are considered. It would be interesting to know how WinTD handles this situation, for example. It would not surprise me to hear that it does better.

I suspect the SwissSys algorithm is probably fairly close to the FIDE Dutch System algorithm (or near-algorithm), allowing for differences between the FIDE and USCF Swiss pairing rules. That is, like the Dutch System, it allows for a certain amount of backtracking. (It has to do so, or else there would be too many situations where it would not find a pairing.)

But are we prepared to say that the algorithm used by SwissSys is “wrong”, and produces “wrong” pairings?

24

Brian, make sure you’ve selected lookahead pairings instead of top down pairings in your pairing options in SwissSys. It might affect the way SwissSys handles situations like this.

SwissSys’s pairing does appear to be wrong in light of the second sentence of rule 29D1a: “Care must be taken in doing this that the odd player can be paired in the next score group …”.

I agree with others, though, that it was too late to change the pairings once games had started.

25

Correction: I should have said “Make sure in-depth searches are enabled.”

26

  1. Is the pairing legal?

  2. Is the pairing best?

Number two may be the point. Getting any one “committee” to agree on what is a best pairing is problematical. It is not a matter of right or wrong but perhaps a matter of judgement regarding what is preferable or not preferable. Perhaps the only “fair” system is “All Vs All” playing both colors (talk about impracticable?!).

27

On the contrary, many human TDs would do it this way – namely, those who use top-down pairings rather than looking ahead properly. Especially likely to make these pairings are those TDs who write down each pairing (on either the pairing sheet and/or the pairing card) as they make it. This atrocious practice makes it difficult for the TD to admit, after seeing the next few pairing cards, that something ought to be changed back there.

Agree with the single drops, disagree with the colors. With either pairing, two of the three colors are bad. One should color-transpose only to reduce the number (or severity) of bad colors, never just to push the bad colors down to a lower board.

Colors are irrelevant here anyway. They take a back seat to avoiding multiple drops.

Excellent point. I always hate it when the TD rips the pairings out of the printer, posts them without a thought, and then changes them only in response to complaints. TDs should look at the pairings before letting the players see them, so they can make any necessary adjustments in peace and quiet.

I’d call the SwisSys pairings “barely legal” – like a person 21.001 years old ordering a drink at a bar.

I hate this concept of “natural pairings”, whatever that means. In this case the SwisSys pairings are far less natural than the suggested pairings, because of the multiple drops. Even if you define “natural pairings” as “only those transpositions made to avoid pairing the same players twice”, there are almost always multiple ways to accomplish this, so the definition becomes ambiguous.

Please don’t confuse “natural pairings” with “top-down” (i.e. amateurish) pairings. A far better concept is “raw” pairings – pairings based on scores and ratings only, with no regard given even to whether players have already played each other. It it far preferable to compare each of two sets of suggested pairings against the raw pairings, rather than against anybody’s concept of “natural” pairings.

On the contrary, the last sentence in 29D says exactly that: “… the first priority (other than avoiding restricted pairings) is to have players play as close to their score group as possible.”

Yes, that’s a reasonable definition.

And that’s a reasonable conclusion.

Actually, that was apparently a ten-player event.

Bill Smythe

28

In the “pairing logic” screen, SwissSys distinguishes between the “natural pairings” BEFORE making allowance for rematches, and AFTER making that allowance. In both cases, transpositions and interchanges for color allocation reasons come AFTER the “natural” pairings.

So a pairing with no variations from the “natural” pairings is merely one where there were no color-related transpositions or interchanges. After doing a pairing, SwissSys tells you many how many changes it made from the “natural pairings”. This is AFTER consideration of rematches. A pairing where there were many multiple drops or violations of strict Top Half Versus Bottom Half pairings because of rematches would still be displayed as having no variations from natural pairings. Only if you went into the pairing logic screen and selected BEFORE would you see the strict THVBH “natural” pairings.

The SwissSys pairings under discussion in this thread were described by the program as the “natural” pairings, and didn’t even explain them. Only when I selected BEFORE did it show them as “unnatural” and explain them (as being necessary to avoid rematches).

29

Hmmm…so how you set the program had some effect on the pairings?

30

Perhaps. However, if only some settings of the program produce pairings which are “correct” pairings according to the USCF, perhaps a program marketed as one which does “USCF pairings” ought to have those as the defaults. Or perhaps the USCF committee for Tournament Directors ought to document what those settings are. Don’t you think?

31

Really? Really? Really (apologies to WWE Superstar "The Miz). Correct and legal are two different things. As one NTD put it: Those pairings programs are great Senior TDs, pairing-wise.

32

Your previous post implied that the programs have to be set up correctly, and that they are not set up correctly out of the box. Now you are saying that the USCF pairing rules are beyond the capabilities of the pairing programs (and Senior TD’s, apparently) to implement properly, and that only NTD’s are capable of the ineffable mysteries of USCF pairing. Whatever these are, they are apparently not susceptible to documentation. (Or, maybe it is that these NTD’s think they are great at pairing but suck at reducing their procedure to a clear algorithm. To me, that would mean they are deceiving themselves and are probably winging it a lot more than they think they are.)

Anyway, which is it? The programs have to be set up correctly, but normal mortals can’t do it – only NTD’s? Or is it that the ineffable mysteries of USCF pairing are beyond being clearly documented so that they can be implemented by software (or apparently even by a human TD short of an NTD)?

33

As usual you are off base. But I suspect that means nothing to you. BTW, is there any progress on the on-line TD testing program you earlier suggested that at you could produce?

34

If I were actually off-base, it would mean a lot to me. It would be bad to be off-base. Oh no, off-baseness is not for me. I definitely want to be on-base.

However, you thinking I am off-base does not equate to my actually being off-base. If you want to say something that would be meaningful to me, you would need to do more than merely say that I am off-base. You would have to demonstrate it, which you apparently can’t be bothered to do.

How would you do that? – By writing down an algorithm which produces legal and correct USCF pairings in all situations.

If you cannot do so, then it means, at least, that you are no good at writing down algorithms and procedures. Considering that you were Editor of the 5th Edition of the USCF Rules of Chess, which included about 80 pages on pairing procedures, that would be a surprise. On the other hand, if you are good at writing down algorithms and procedures clearly, you somehow failed to do it in the 5th Edition. Why was that? Maybe you should start calling other people off-base when you have corrected this problem.

On your other, apparently irrelevant, point: I haven’t done any work on the TD testing program. I could produce it, but I am not doing so at the moment.

35

Whatever?!

36

In other words, SwisSys “natural pairings BEFORE” is what I call “raw pairings”, and SwisSys “natural pairings AFTER” is what you call “natural pairings”.

The latter is a sticky concept, because there are usually multiple ways to get from “raw pairings” to “natural pairings”, some of which are better than others. In other words, “natural pairings” is an ambiguous concept.

For this reason and others, I continue to maintain that any proposed pairings should be compared against “raw” rather than “natural” pairings.

In the following table, the first column is the player number. The second column deals with the SwisSys pairings, and lists the rating difference between each player’s “raw” opponent and his actual opponent. The third column deals with the alternative pairings (suggested by several TDs and others) in the same way.

  • #5 000 057
  • #1 000 220
  • #2 293 220
  • #7 318 057
  • #3 293 000
  • #4 318 000

Thus, the SwisSys pairings result in a total rating distortion of 1222 points, compared with 554 for the alternative suggested pairings.

Of course, it could be argued that rating difference alone is not the best measuring stick, and that score difference should also be taken into account. The following table, for example, considers each 0.5-point score difference to be worth an additional (or subtractional) 100-point rating difference. In other words, if the higher-rated player is also the higher-scoring, the rating difference is adjusted upward by 100 points for each half-point score difference, but if the higher-rated is the lower-scoring, the rating difference is adjusted downward.

  • #5 000 057
  • #1 000 420
  • #2 093 420
  • #7 618 057
  • #3 093 000
  • #4 618 000

Now the rating distortion for the SwisSys pairings is 1422, and for the alternative pairings 954.

The old thread Point-Count Pairings touches a bit on these ideas.

Bill Smythe

37

Bob, in my version of SwissSys (v8.82) there is no “top-down” vs “look-ahead” option. Is it perhaps under some other name? On the “Rules for Pairing” screen, I have it set for “USCF defaults”. There is an option called “In-depth” searches, which is checked. (It is turned when you click “USCF defaults”.)

38

Regarding the suggestions that the 200/80 rules should be turned off.

On p 149 of the rulebook, it states that the 80/20 limits may be exceeded “somewhat” in order to make a transposition or interchange which opens the door to further transpositions and interchanges and makes it possible to achieve a significant improvement in color assignments.

SwissSys lets you adjust these limits up or down, but you can’t turn them off. (If you pick FIDE Defaults, it makes the limits 3000/3000 instead of 80/20, which I guess effectively turns them off.)

However, I am not finding the place in the rule book where it states that these rules may simply be ignored in a tournament conducted according to USCF Swiss rules.

39

The only time I would turn them off would be in a double-round Swiss. Those are generally quick-rated events where each pairing is for two games (one with white and one with black). Since colors are equalized every round there is no reason to try to adjust the pairings to alternate the colors.

The rulebook does not specifically state that th 80/200 rule should be ignored in that particular case, but the rulebook does figure there is some minimal common sense (after all, with double rounds the 200 portion should never come up as virtually all players are equalized every time the pairings are done).

40

Sometimes you just have to use common sense, especially in small tournaments where pairings become difficult and crazy.

In fact, with under 20 players, the best idea may be just to turn the software off altogether and pair by hand.

As to the 80/200 limits, there has been a lot of confusion in this thread. Does turning them off mean changing them to zero, or changing them to infinity (e.g. 3000)? It depends what you’re trying to do.

First of all, these limits apply only to transpositions made to improve color alternation/equalization. It does not apply to transpositions made for other reasons, such as to avoid rematches or to minimize score differences. For that matter, it doesn’t even apply to all color situations – three blacks in a row, for example.

In the problem at hand, colors had nothing to do with the problem, anyway. Both sets of possible pairings ended up with bad color equalizations on two of the three boards. The real issues here were score differences (multiple drops vs single drops) and top-down vs look-ahead.

I have often suggested, in small tournaments, changing the 80-point alternation limit to zero, while leaving the 200-point equalization limit at 200. That’s because of the camp effect. Making colors work too well tends to divide the players into two camps, those who started with white and those who started with black. Good-color pairings are inter-camp pairings. Bad-color pairings are intra-camp pairings. If, in a small tournament, you make too many inter-camp pairings in the early rounds, you’ll run out of available inter-camp pairings in the later rounds. So, since there are going to be a lot of bad-color pairings anyway, let’s put them where they’re least harmful, i.e. in the rounds where mere alternation, rather than equalization, is the issue. That’s achieved by using a zero-point alternation limit.

In fact, I find myself wondering whether, had that advice been followed in round 3 of the tournament in question, round 4 pairings might have been smoother all around.

The opposite concept, setting the limits to 3000, serves a different purpose entirely. It’s for FIDE-like situations where color is king.

Bill Smythe