Thanks much Bill! - especially the note about the compound pairing. I had been acting under that assumption but with a fair bit of uncertainty, so I appreciate it.
As I mentioned, I didnt think I’d have time to check back. I ended up using the computer pairings - I was a little afraid that personal bias - I wanted to see fairer/more even pairings for the players (also considering prize impacts of a possible change) - was creeping into my extra effort to find an alternative to the computer.
Aww. Your instincts seem sound. I was hoping you’d go with your gut.
I’m not too sure about “more even pairings for the players” or “considering prize impacts of a possible change”, though. You had plenty of good reasons – better colors, etc – to make the change.
In my earlier post, I said:
Now I’d like to modify that – or maybe even retract it (somewhat). As written, it seems to suggest that the proposed pairing should have been rejected if the rating difference in the third pairing had been 82 points instead of 78. And that would be unfortunate, as it is clear that the proposed pairing accomplishes a lot.
So it would seem better, when considering Point-Count Pairings (or anything like it), to assign, to each player in each pairing, two scores – a positive score for what is accomplished for that player, and a negative score to reflect the rating difference between the player’s raw and proposed opponents.
In this case, in the third pairing, player 34 would be assigned 80 positive points because it corrects color alternation, and 78 negative points for the rating difference, for a net of plus 2.
The overall “value” of a proposed pairing would be the greater (the more positive, or the less negative) of the net scores of the two players in the pairing. This is analogous to the rule for a single transposition, where the smaller of the two rating differences (between each player’s proposed opponent and his raw opponent) is the one that counts.
If the rating difference had been 82 instead of 78, the net would be minus 2 – but that should not automatically cause the pairing to be rejected! If the net effect is increase the total value of all the pairings, then go for it.
Yes (me too) on the gut Like I posted, I didn’t feel like I’d have time to check back on the forums. Having these things on a platform with push notifications might help the instant feedback / response we’ve gotten used to in a Facebook-esque world - but that’s a whole another thread.
Interesting idea - of assigning a value to the improvement for each player and basing some decisions on that… Although I could see a player considering getting his due color as a detriment… if he’s due black.
Another item came up today with the pairings. Unbeknownst to me, players had been forwarding my pairings over to their new coach, who is a frequent reader and sometime contributor to these forums. He took issue to my pairings of the top score group and suggested I ask here about it, since I had already shown a willingness to do so. (part of me wanted to avoid this precendent of crying to you guys each time SwissSys or someone has a problem with my pairings, but in the interest of enlightenment…)
Here’s the scenario:
4.5 6 1825 BWBWW
4 2 2120 WBWBW
3 1960 BWBWB
5 1879 --BWB
7 1813 WBWBB
3.5 1 2189 BWBWB
9 1754 WBWBW
13 1698 B-WWB
23 1409 BWBWW
illegals (all prior matches):
relevant: 2 v 6 1 v 7
not relevant: 1 v 6, 1 v 2, 2 v 13, 3 v 9
After droping 6 into the top score group, there are 3 players due white, 2 players due black.
When deciding who to drop, lowest ranked (7) is about as strongly due white as possible.
I see no reason to drop a different player.
The first natural pairing is illegal (6 vs 2), so top score group goes
6 vs 3
5 vs 2
In the next score group, the first natural pairing is also illegal.
7 vs 9
1 vs 13 (yes, 13 is due for white, but for alternation only and there is no transposition anywhere close here)
23 drops to the 3.0 score group (which helps because there were 5 due white, 4 black in that score group)
So… I think this probably has been addressed here a lot of times - but the issue the person took with the pairings was sort of along this line, expressed by Bill:
Two factors in my methodology in pairing it (and SwissSys’s as well, aparently) that caused the issue:
I was considering the score groups discretely. Once I determined that dropping the lowest player would not hurt color allocation, I then considered the second score group as its own entity. The fact that the dropped player vs the highest ranked player in this score group was illegal did not enter into my decision on who to drop. The next player in line to play the dropped player was due the opposite color.
I was also considering the avoidance of an illegal matchup as discrete from any transpositions/interchanges. In fact, I would consider that I did no transpositions in pairing those groups. This fits in with the quote above - I wasn’t considering avoiding an illegal match-up as “avoiding a horrible pairing”.
The player suggested I pair it as follows:
Looking ahead and seeing that player 7 was going to get a huge “transposition” to avoid the illegel pairing (-435 points). Drop the other player due white (5) into the next score group instead of 7. This would represent only a 79 point increase in the rating of 5’s opponent (and a 79 point decrease in the rating of 7’s natural (but illegal) pairing.
so:
6 vs 3
7 vs 2
5 vs 1 ( color problem of equalization)
13 vs 9
23 drops
So ive moved the board 4 single color alternation problem to a higher board 3 equalization problem. This seems wrong, but - that being said - the pairings are much “fairer” with less deltas between players. Also it has a couple of side benefits:
Player 7 has had a pretty easy ride relatively. Pairing him with the 1750 only continued this.
Player 13 and Player 9 can play directly for the U1800 glory and cash.
oh and because i dont have my rule book right here, who gets due color in 6th edition --BWB vs BWBWB. I think its BWBWB, but that seems counterintive as 4 blacks and 2 whites doesnt seem as harsh a treatment as 3 blacks and 1 white. I used higher ranked in my example above, which I believe was the older 4th edition rule (you dont go all the way back in the history to determine due color). Either way, the pairings above produce an equalization problem.
Comments? Thoughts?
So again - thanks much in advance, gentlemen!
If there are any females that regular follow this forum, please excuse my ignorance
Sorry. But the player is correct. Those are clearly superior to yours. You are (in effect) doing a 435 point switch (in upfloating 9 rather than 1) rather than a 66 point switch in the downfloat. You don’t do that to correct colors. (What did the computer come up with—yours or the player’s alternative?
Assuming the two players have the same difference in colors (+1 Black for both here), work back to the first point where they differ. That’s round 2, - vs W. So the second player gets B. It’s as simple as that. The point is that color matters more as you go deeper into the tournament, so the later a player gets a B (W), the more she “deserves” W (B).
Yes, I didnt do anything to correct colors. The inclusion of this sentence seems to imply that I went astray in trying to correct colors. As there were an odd number of players, I simply dropped the lowest player in the top score group and i paired him with the highest ranking non-illegal player in the next score group. The 435 point movement was to avoid an illegal matchup and I gave the player the next legal opponent. The colors happened to work out as best as could be expected (1 alternation problem across both score groups). This seemed to be a proper application of the rules for pairing included in the USCF rulebook. When the computer (SwissSys) agreed with them, I didnt question them.
Did I miss in the guidelines that there is a concept of “upfloating” - bringing the top rated player up from a lower score group and then pairing the upfloated player with the lowest rated player to which he is eligible? if this is an available option, is there any preference given to downfloating vs upfloating? I suppose whichever one I “feel” produces the best pairings? (I can use the point-count method to try to produce an unbiased view of that.)
where 1 (the upfloated player) would be obligated to play 5 (the lowest legal pairing) and the rest of the pairings are legal.
Here a color problem of equalization is inevitable, so I would not have endeavored to avoid it and this method would have produced the alternative pairings.
An “upfloat” is simply a player who is paired against the odd man dropped from a higher score group. There is no concept of raising a player up from a lower score group into a higher score group.
Whenever you downfloat a player, you upfloat another.
You can look at it as downfloating one player into a lower score group, or as upfloating the other player into a higher score group. I prefer a more balanced viewpoint: you are creating an additional score group, between the other two, that contains just two players.
For example, when pairing a 2.0 against a 1.5, think of these two players as forming a new two-player 1.75 score group.
The decision whom to downfloat is equally important as the decision whom to upfloat. In this case, when downfloating player 7 turned out to force a 400-point transposition in the lower group, it would have made a lot of sense to undo the pairings in the upper group and look for a different player to downfloat. (Sometimes the “look ahead” method requires that you look ahead into the next score group.)
Let’s apply Point Count Pairings, or at least the incarnation of it suggested in this thread:
raw pairings:
6 vs 2 (already played each other, and bad equalization)
3 vs 5 (bad equalization)
7 vs 1 (already played each other, and bad equalization)
9 vs 13 (no problems)
proposed pairings 1:
6 vs 3 (value for 6: 9999 minus 160) (value for 3: 200 minus 54) (net pairing value 9839)
2 vs 5 (value for 2: 9999 plus 200 minus 54) (value for 5: 0) (net pairing value 10145)
7 vs 9 (value for 7: 9999 minus 435) (value for 9: minus 113) (net pairing value 9564)
1 vs 13 (value for 1: 9999 plus 200 minus 491) (value for 13: minus 80 minus 115) (net pairing value 11598)
proposed pairings 2:
6 vs 3 (value for 6: 9999 minus 160) (value for 3: 200 minus 54) (net pairing value 9839)
2 vs 7 (value for 2: 9999 plus 200 minus 12) (value for 7: 0) (net pairing value 10145)
5 vs 1 (value for 5: minus 229) (value for 1: 9999 minus 200 minus 66) (net pairing value 9733)
9 vs 13 (value for 9: 0) (value for 13: 0)
If my arithmetic is correct (and it well may not be), pairings 2 defeats pairings 1 by a score of 29717 to 21146.
A few minor comments:
A player due black but getting white is just as detrimental to the pairings as a player due white but getting black.
As long as you downfloat a player due white (which this situation calls for), I see no reason to worry about how strongly the downfloated player is due white.
It wouldn’t surprise me if Akzidenz (Maret Thorpe) were reading along. She posts sometimes (and she has had a lot of excellent posts) but lurks other times, I believe.
Well… i thought i understood the other thread - but then this application of it threw me a bit. i will have to examine it - as i thought we would generate a raw undesirability score and then compare the two options to it to determine which option made it “more better”.
Here’s also something to note. There is checkbox in SwissSys under Rules for Pairing called “Weikel Oddman Pairings”. When i check this, it produces the alternate pairings. It turns out that it does indeed “upfloat” a player into the higher score group:
Google is mum on the subject, but SS help shows:
Weikel oddman pairings - Allows for specialized treatment of the oddman in a score group. Brings up an oddman from below and pairs this player as if in the higher score group. Non-standard.
The Weikel pairings (aka Harkness) produce the “right” results only by chance—it’s a very different approach to dealing with floats than the standard USCF methods.
The ideal under USCF rules is for the lowest rated player in the higher score group to play the highest rated player in the lower score group. That’s subject to a whole string of caveats (not making colors worse, not leaving either remainder scoregroup unpairable, …). Obviously the first and most important caveat is whether you can actually pair the two players involved. In your case, you can’t. If the ideal pair of floats doesn’t work, you have to change the upfloat, or the downfloat, or (possibly) both. Here you have that massive delta between the first and second players in the lower score group, and a rather small one in the higher score group, so the cheaper fix is to change the downfloat.
The Weikel pairings are (roughly) equivalent to downfloating the middle player in the higher score group. With three players in that (after one is floated up to play the top scorer), that would be that second player from the bottom. If there were five in the score group, it would be the third from the bottom, which probably wouldn’t be the one you would choose to float down under USCF rules.
I’m horrified by Mr. Cassidy’s apparent suggestion that if you can get the pairing program to do something, it must be a legitimate way of pairing the section. Then again, it is not the first time I’ve been horrified.
hah! no - i wasnt saying that at all - I just thought it funny that I asked a question about the concept of “upfloating” - i.e. I didnt remember this concept being covered in the rules. Then I found the exact option, with quite the obscure name, in the pairing program to upfloat a player instead of downfloat. I understand that this method arrived at the same pairings by chance! It also screwed up lower pairings - and it wasnt like I was going to accept that just because there was a check box for it
A couple of items to note:
I said I wouldnt have time to check back here and I didn’t - I sent the pairings out on Monday. But over the past day, the fact that there were better, more balanced, pairings out there kept eating at me. So this morning, I sent out revised pairings and made the switch.
Of course, the player, whom I like a lot and have a lot of personal bias towards, that received the 435-point downfloat from his raw pairing was disappointed to see that he was now paired down only 69 points. In fact, he didnt see it that way at all - he saw it as a 356 point increase. His problem was also that the had spent 3 days preparing against a 1750 player instead of against a 2120. I was ready to defend the switch and defend all the reasons why this was not a subjective judgement, but I was not ready to defend why it took me almost 3 days to make it happen
Before sending the pairings, I did the math of the point-count system in a slightly different way. Assuming I am not going to make an illegal pairing, I made a simple spreadsheet defining the ratings differences + color penalties as follows. As suggested, this system pointedly does not distinguish between positive and negative benefit - just the overall difference between the raw pairing opponent’s rating and the adjusted rating. I included some penalties for the raw pairings (3 illegal matchups and 3 color equalization problems), but they are irrelevant.
For each player, penalties consist of the rating difference between the raw pairing opponent and the proposed opponent.
Rematches are worth a 5000 point penalty. Penalties for both players combined in the raw penalties column. Penalties split between players a and b in the revisions. The order the players are listed in does not reflect color. If a resultant color problem would occur, penalties added as such:
A bad color alternation is worth 80 (only case of this is applied to player 1698 in revision 1)
A bad color equalization is worth 200 (only case of this is applied to player 2189 in revision 2)
raw pairings:
penalties
1825 2120 10200
1960 1879 10200
1813 2189 10200
1754 1698 0
total undesirable penalties: 30600
revision 1: penalties
a b
1825 1960 160 54
2120 1879 54 160
1813 1754 435 115
2189 1698 115 515
total undesirable penalties: 1608
revision 2: penalties
a b
1825 1960 160 54
2120 1813 12 69
2189 1879 266 229
1754 1698 0 0
total undesirable penalties: 640
I got a trial version of WinTD to run this tournament through It will expire before I’m ready to run another tournament again, but maybe the gods will take pity on me and send me a free permanent key.
Float and its various derivatives are used in the description of the FIDE pairing rules. It’s a very convenient term since it can be used as both a verb and a noun, and with the “up” and “down” prefixes can be used to describe both directions of movement. The USCF rule book talks about the “odd player” (from the higher score group) “dropping”, but doesn’t really have a word to describe the person with whom the dropped player is paired.
I think they’re really the same thing. Either you regard the raw pairings as the standard against which others are measured, or you attach an undesirability score to all pairing sets, including the raw pairings. It’s like increasing X instead of decreasing Y before you compute X - Y.
No matter how you select the players for the cross-group pairing, you are always downfloating one player and upfloating another, no two ways about it. Standard USCF pairings attempt to downfloat the lowest rated and upfloat the highest rated. Harkness attempts to downfloat the middle player and upfloat the highest. As for Weikel, I can’t quite figure out what it attempts to do based on the passage you quoted. Apparently it selects (somehow) a player to upfloat, then downfloats the player who would have played the upfloater if the upfloater had been in the top group to begin with.
Once again, with feeling: It is not a question of downfloating “versus” upfloating. You’re doing both, whenever it is necessary to pair two players not in the same score group. The player from the higher group is called the downfloater, and the player from the lower is the upfloater.
That’s the problem with changing the pairings after they’re posted (or announced). There will always be somebody who preferred the original pairing. It’s probably best not to change the pairings, once announced, except in the case of a really egregious error.
I think he meant pairing two players who have already met. Perhaps he also meant Really Bad Colors, such as three in a row, or giving a player the wrong color who is already out of balance by 2 or more.
By the way, before you take this “point count” business too seriously, please remember that it is strictly my own invention, intended mainly as a conversation starter. No official chess organization has specifically endorsed it.
Nevertheless, it probably isn’t too far from what WinTD already does.
FIDE doesn’t allow it, no matter how egregious the error is. I recall a norm swiss when players withdrew after pairings had been posted. The opponents were given house games.
I’ve done the former. Certainly heard about the latter. Those of us who have been around long enough recall Mike Nolan talking about a tournament where Black won every game in the first two rounds.
Like I posted, I didn’t feel like I’d have time to check back on the forums. Having these things on a platform with push notifications might help the instant feedback / response we’ve gotten used to in a Facebook-esque world - but that’s a whole another thread.
and basing any moves (illegal or color or otherwise) off of them.