Swiss system: match distribution problem!

[SwHawk]

I’m referring to difference within the Swiss itself nothing to do with Elo.

The problem I see is in multiplayer mode and trying to avoid playing same players twice option. Meeting the criteria strictly quickly becomes impossible, at which point the algorithm throws the bottom players to fill the remaining spots.

So in smaller tournaments this can become impossible by round 2. Take a 6p game with 24 participants. In R2 the 4 winners get paired with 2 last place players because ranks 5-24 all fail the no rematch criteria. So instead of going through the list again to look for best Swiss score match it simply throws the 2 lowest ranks in. By R3 this grows to the bottom 3-4 playing against the tournament leader.

Well this has been stated above, this is an issue with the Swiss system by itself. I already included the excerpt from the Wikipedia article, actually backed by a scientific publication saying that in order to avoid rematches, the hard limit is half the number of players, and that is for 2 players games. So as I said, this does not stem from the BGA implementation, but from the number of rounds set in those tournaments with regards to the number of players...

[sprockitz]

I’m referring to difference within the Swiss itself nothing to do with Elo.

The problem I see is in multiplayer mode and trying to avoid playing same players twice option. Meeting the criteria strictly quickly becomes impossible, at which point the algorithm throws the bottom players to fill the remaining spots.

So in smaller tournaments this can become impossible by round 2. Take a 6p game with 24 participants. In R2 the 4 winners get paired with 2 last place players because ranks 5-24 all fail the no rematch criteria. So instead of going through the list again to look for best Swiss score match it simply throws the 2 lowest ranks in. By R3 this grows to the bottom 3-4 playing against the tournament leader.

Well this has been stated above, this is an issue with the Swiss system by itself. I already included the excerpt from the Wikipedia article, actually backed by a scientific publication saying that in order to avoid rematches, the hard limit is half the number of players, and that is for 2 players games. So as I said, this does not stem from the BGA implementation, but from the number of rounds set in those tournaments with regards to the number of players...

It is absolutely the bga implementation that pairs the top player with the lowest when the group has already met everyone!

[Pinovitch]

Assuming no drawn games, determining a clear winner (and, incidentally, a clear loser) would require the same number of rounds as that of a knockout tournament, which is the binary logarithm of the number of players rounded up. Thus three rounds can handle up to eight players, four rounds can handle up to sixteen players and so on. If fewer than this minimum number of rounds are played, two or more players could finish the tournament with a perfect score, having won all their games but never having faced each other. Due to the fact that players should meet each other at most once and pairings are chosen dependent on the results, there is a natural upper bound on the number of rounds of a Swiss-system tournament, which is equal to half of the number of players rounded up.[9] Should more than this number of rounds be played, the tournament might run into the situation that there is either no feasible round, or some players have to play each other a second time.

On peut également lire dans cet article "Après le premier tour, les joueurs affrontent des adversaires qui comptent le même nombre de points (ou à peu près). Au cours d'un même tournoi, aucun joueur ne rencontre deux fois le même adversaire. "

Et c'est bien ce que je reproche le plus au calcul des tournois en 1 contre 1.

Alors, au-delà des considérations théoriques, pourquoi ce qui était possible avant ne l'est plus maintenant ?

https://boardgamearena.com/tournament?id=74335 : aucun doublon
https://boardgamearena.com/tournament?id=36239 : aucun doublon
https://boardgamearena.com/tournament?id=36325 : aucun doublon

Et pourtant le nombre de concurrents était similaire pour 20 parties également.

[SwHawk]

It is absolutely the bga implementation that pairs the top player with the lowest when the group has already met everyone!

It's actually the creator's fault, for selecting such a great number of steps (which, I'll say it AGAIN, will force rematches as evidenced by scientific studies) and selecting TRY not to create rematches. The algorithm will then create games with players that haven't been playing together, and since the highest ranking players have already played together, it matches them against low ranking players to avoid rematches.

On peut également lire dans cet article "Après le premier tour, les joueurs affrontent des adversaires qui comptent le même nombre de points (ou à peu près). Au cours d'un même tournoi, aucun joueur ne rencontre deux fois le même adversaire. "

That's what a normal Swiss tournament should do. But you seem to be disregarding that setting too many rounds will either create the situation with no possible pairings (since a player with a suitable level AND that hasn't been met before isn't possible) or forcing rematches. Which is what you're encountering here... Because BGA has to put safeguards against ridiculous settings such as those you chose. Granted they could have placed checks when creating the tournament to prevent the tournament from ever being created so that such issues wouldn't arise, but they choose to permit those settings, so that the creators may force rematches... That's their choice, and I won't argue it here.

Alors, au-delà des considérations théoriques, pourquoi ce qui était possible avant ne l'est plus maintenant ?

It was certainly possible before because players would be matched against players that weren't of matching level... I mean, if there's been a v2 of the Swiss system, that's basically because some things were broken with v1... Swiss-system is about matching players with players of similar level, not for matching everyone with everyone... So if it did before, it wasn't working as expected. You're trying to replicate that behaviour by trying to enforce no rematches even though the number of steps is too large for it being possible.

Also, both of the configurations having everyone play everyone and everyone play with players of matching level are not reconciliable... You can either have matches with players of the same level (as long as the pool is large enough, so half of the players present for 2p matches, or, from what I can figure, the total number of matches in one round for multiplayer matches...) and have rematches if the number of steps is too large, or have no rematches but accept to play with low level players...

Otherwise, as I said before, Swiss system tournament is not the system you're looking for. If you're adamant about every player playing against every other player, then round-robin is what your looking for... Which, incidentally, for the number of players you've provided would basically need approximately the same number of steps, at least for a 2p game. If you actually want players to be matched with players of a similar level, then you have to reduce the number of steps to be at maximum the number of possible matches during a step to avoid being in the situations you described...

I'll state it again, you find yourself in this situation because of the settings you've used and not because BGA's implementation is broken... Either you change the settings, or you accept to be put in this situation. But understand that the situation stems from the settings you've chosen, rather than on anything BGA does. If you don't understand why I strongly urge you to read the scientific publication that's referenced in the Wikipedia article, and that for once is free to download and read: https://arxiv.org/abs/2007.10704 (click PDF under Download in the top right hand corner of the page).

[Pinovitch]

La Tour Ronde me limite à 15 joueurs et on peut soit balancer toutes les parties d'un coup ou bien les programmer avec dates fixes. Impossible de sélectionner l'option "commencer dès que match précédent fini". C'est donc trop restrictif !

Le Système Suisse était donc le seul moyen de faire des événements à plus de 15 et sans être limité dans le nombre... car qui peut prédire le nombre de participants au lancement du tournoi ? Et je me refuse de fermer la porte aux gens. La V1 faisait ce boulot sans redondance.
Et c'est avec cet esprit que j'ai crée ces tournois, l'algorithme de Matchmaker, proposé par BGA dès l'instant, devait pouvoir le faire. Pourquoi ne devrait-il plus le faire d'ailleurs ?
Ma priorité : faire le plus de rencontres possibles en évitant les doublons comme je le faisais auparavant. Finalement je m'en fiche qu'on ait un système suisse.

Je constate juste que ce n'est plus possible et c'est ce qui me dérange, moi, et les participants des tournois.

Et je ne parle même pas des tournois multijoueurs qui font se rencontrer le meilleur contre les plus mauvais, systématiquement après quelques journées... là j'en ai fait mon deuil et ai choisi l'autre option.

[SwHawk]

Le Système Suisse était donc le seul moyen de faire des événements à plus de 15 et sans être limité dans le nombre... car qui peut prédire le nombre de participants au lancement du tournoi ? Et je me refuse de fermer la porte aux gens. La V1 faisait ce boulot sans redondance.
Et c'est avec cet esprit que j'ai crée ces tournois, l'algorithme de Matchmaker, proposé par BGA dès l'instant, devait pouvoir le faire. Pourquoi ne devrait-il plus le faire d'ailleurs ?

Swiss-system tournaments are based on players facing opponents of the same level... If the v1 did allow for players to be matched even though they were not of the same level, then it was a serious default in the matchmaking algorithm, which BGA admins chose to address by releasing this v2 which seems to be more coherent with the spirit of Swiss-system tournaments...

If you want to handle a large number of players and have players encounter the maximum number of opponents, then you could do a group stages tournament, with at least the first step as a round-robin for each group, then another round robin (or swiss-system but with a known number of participants since you indicate how many players will be qualified in each group) for the best 15 (or higher if swiss) players.

Another alternative, which is done for some tournaments, is to actually forego using the tournament system offered by BGA, and handle matchmaking by yourself. I've seen several articles (this one for instance) which will allow you to indicate all the player names and create all the matches that can be created with them regardless of how many players should be present in that game. While it is easy to keep scores for a 2p game, you will have to create some kind of score scale for multiplayer games. You should then have players report the link to the table the match was played, as a means to check the reported results. This way, you can have all the matches set up in advance, and players can start games as soon as their opponent is available. Granted, that will make your life that much harder as you'll have to schedule and check for everything... There are websites that can help you organize this, I'll let you search on the web, I've found one that seems to be able to accommodate for 256 players round-robins...

[Pinovitch]

J'ai déjà fait ce genre de chose et crois-moi, on passe plus de temps à relancer les gens et chercher les résultats non publiés qu'à jouer.
J'ai déjà donné, merci !
L'automatisation me convient bien.

Et je pense que la V1 respectait bien le système suisse dans son esprit : trouver un système de classement où il est impossible de faire se rencontrer tout le monde tout en évitant de se faire rencontrer des joueurs deux fois. Qu'on fasse se rencontrer un fort contre un faible de me dérange pas, ce que je cherche c'est de m'approcher le plus possible d'une tour ronde.
La V2 ne respecte plus le deuxième paramètre "Au cours d'un même tournoi, aucun joueur ne rencontre deux fois le même adversaire.", qui lui aussi est essentiel au système suisse. Que les paramètres de création choisis soient cohérents ou non, à partir du moment où on est plus de joueurs que de matchs à jouer. Et c'est bien ce qui embête mes joueurs !

On est là pour s'amuser, pas dans un tournoi d'échecs officiel... et l'amusement frôle l’écœurement quand on rencontre cinq fois de suite ou plus le même adversaire. Car une fois de plus, qui peut prédire le nombre de participants lors de la création pour des paramètres idéaux ?

[sprockitz]

It is absolutely the bga implementation that pairs the top player with the lowest when the group has already met everyone!

It's actually the creator's fault, for selecting such a great number of steps (which, I'll say it AGAIN, will force rematches as evidenced by scientific studies) and selecting TRY not to create rematches. The algorithm will then create games with players that haven't been playing together, and since the highest ranking players have already played together, it matches them against low ranking players to avoid rematches.

This isn't what is happening, but we clearly can't get through to you...so continuing to argue is fruitless. They are playing bottom people they have already played (not bottom people they haven't played)...because the algorithm is broken...but you aren't listening and continue spouting things that aren't true.

[SwHawk]

Et je pense que la V1 respectait bien le système suisse dans son esprit : trouver un système de classement où il est impossible de faire se rencontrer tout le monde tout en évitant de se faire rencontrer des joueurs deux fois. Qu'on fasse se rencontrer un fort contre un faible de me dérange pas, ce que je cherche c'est de m'approcher le plus possible d'une tour ronde.

I'll state it one last time, a Swiss-system tournament is about opponents having roughly the same level. And is not supposed to have low level players matching against high ranking players! It just wasn't designed that way! As I said before, it the v1 algorithm allowed such matchmaking, then it was critically flawed with regards to what a Swiss-system tournament should be. You seem to hava an incomplete understanding of the different types of tournaments offered by the platform... If you want a round-robin tournament, for now you're limited to 15 players. You can create a suggestion in the bugs & suggestion system to ask for that limit to be increased, but there's certainly a reason it was designed that way. Other method is to organize it yourself...

La V2 ne respecte plus le deuxième paramètre "Au cours d'un même tournoi, aucun joueur ne rencontre deux fois le même adversaire."

The option states: Try to prevent same players to be opponents twice. Operative keyword being TRY. If it isn't possible then it will make a rematch... And contrary to v1, it takes into account the level difference, which means that if the level difference (score) is too big, it will organize a rematch...

Car une fois de plus, qui peut prédire le nombre de participants lors de la création pour des paramètres idéaux ?

Agree to disagree on that part, you can use statistics to have a rather good idea of the number of players that will be available for a specific game tournament.

Finally, understand that the issue you're encountering, which stems from your misunderstanding of how to set parameters for the Swiss system tournament and restrictions you choose to place upon your tournament organization are, let's be honest, quite an edge case. Basically, the vast majority of people that have taken part in this discussion are either players from your group, or myself... So asking BGA to change settings for roughly 200 players, while they have to satisfy a player base that can be counted in hundreds of thousands to a couple millions, you understand that you're not really a priority to BGA admins... They don't have to cater to your particular problem...

I've offered you an alternative that allows you to organize tournaments for 30 players, were players can meet 22 different players (around 73% of participants), while limiting rematches to 7 players from the group the player has been assigned to, using Group stages tournaments with both of phases being round robin tournaments. Up to 45 players, players will meet at most 25 different opponents (around 55% of participants), with 5 rematches...
Other alternative include resorting to external tools, some of them which can help you remind players through apps or emails that they have to organize matches, and/or post results...

[SwHawk]

This isn't what is happening, but we clearly can't get through to you...so continuing to argue is fruitless. They are playing bottom people they have already played (not bottom people they haven't played)...because the algorithm is broken...but you aren't listening and continue spouting things that aren't true.

Neither are you willing to entertain the idea that these situations arise because selecting 20 rounds for 24 players with 4 player games is obviously going to create rematches after 3 to 4 steps... Even though there are more than ten thousand possible matches, that is not what the Swiss-system is about...

I'll be willing to admit that the algorithm is flawed when I'll have conducted an analysis that proves it, or when someone presents a complete and thorough analysis proving it. What you present for now are claims and nothing more, unless you're able to present otherwise. For now, I'm still analyzing the Patchwork tournament I mentioned, and doing it by hand is quite time consuming... Are you willing to admit that you've been in these situations because of the settings the tournament creator used ? That doesn't seem to be the case seeing your answers...