Ring Games

The Annual Computer Poker Competition started with heads-up (i.e., two player) poker games. As techniques in the two player game improved and as the popularity of poker rose, interest in ring games (games with more than two players) also increased. In 2008, the Annual Computer Poker Competition introduced its first ring game: 6-player limit Texas Hold'em. Subsequent years switched to 3 players.

The competition uses standard rules for its ring games. Generally the rules for each hand of a ring game are similar to a heads-up game. The most notable difference is the betting order. More specifically, there will be a button, a "small blind" to the left of the button, and a "big blind" to the left of the small blind. As per their names, these players make forced bets of the small blind and big blind, respectively. The action starts to the left of the big blind pre-flop, and it starts to the left of the button on all other betting rounds.

Ring Limit Texas Hold'em

As in heads-up limit Texas Hold'em, bets are of a fixed size and there is a maximum of 4 bets per round, even if it is heads-up.

No-limit Games

In contrast to limit games, no-limit games have no limit on the number or size of bets that can take place, except for the size of a player's stack. Players can bet any amount from the minimum bet size to the maximum "all-in" bet, where they bet the remainder of their stack.

No-limit Texas Hold'em

In No-limit Texas Hold'em, the nominal minimum bet size is either the size of the big blind at the beginning of a round or the size of the most recent bet. If the nominal minimal bet is larger than the maximum bet, then the minimum bet is equal to the maximum bet. If someone has already made the maximum "all-in" bet, then you cannot raise. For the competition, raises must be of an integer number of chips.

Doyle's Game

Doyle's Game is a variant of no-limit Texas Hold'em where players can make an "all-in" bet, but a player's stack size is reset to some default size after every hand and the score is kept separately. For example, when playing Doyle's Game with stacks 1000 times the size of the small blind, the maximum amount to be put in the pot by a player in a hand is 1000 times the small blind. Regardless of a player's current stack size, upon starting the next hand, their stack size is set to be 1000 times the size of the small blind. This is akin to both players buying in for 1000 small blinds every hand, cashing in their chips at the end of the hand, and buying in for 1000 small blinds again the next hand.

Resetting stacks in this manner prevents two problems:

  • an agent going bankrupt while we want to continue evaluating them against other opponents
  • an agent having a larger stack than their opponents, potentially creating all-in bets of unequal sizes between the players

2014 Rules

The 2014 AAAI Computer Poker Competition will have one heads-up limit competition, two heads-up no-limit competitions, and two 3-player ring limit competitions.

If you have comments or suggestions, please post to the forums in the rules discussion section.

This year we will have the following competitions:

  • Total bankroll heads-up limit Texas Hold'em competition
  • Total bankroll and instant run-off no limit Texas Hold'em competitions
  • Total bankroll and instant run-off 3-player ring limit Texas Hold'em competitions
  • Total bankroll 3-player Kuhn competition

The basic rules for this year's events will be mostly unchanged from last year, with a few notable exceptions.  First, the 3-player events will have 3000 hand matches, like the events in both 2-player divisions.  Second, in no-limit we will average across all future board cards when both players go all-in.  Much of the variance in no-limit comes from the randomly drawn future board cards at all-in showdowns.  Using the average of all possible future outcomes allows us greatly reduce the variance, and thus we can compute more accurate results in fewer hands.  Finally, the largest change is the elimination of the heads-up limit instant run-off event.  One-on-one play in an instant run-off competition does not correspond well to exploitability, the proper measurement of how close to an equilibrium a strategy is.  With improvements in hardware and algorithms, it is now reasonable to directly measure exploitability heads up limit.  Combined with a greatly expanded field of high-quality instant run-off entries, the ACPC 2-player limit instant run-off competition now is of questionable value given the increasing amount of resources that it requires.  Like last year, we will be seeding the 2-player limit competition with some intentionally weak bots ("chumps") to provide an incentive for bots to try exploiting other players.

Teams may enter one agent per competition (i.e., one agent for each winner determination rule in each of the three game divisions). By default, if a team submits only one agent to a division with both a total bankroll and instant run-off event, then it will be entered into both competitions for that division. Teams may opt-out of this if they wish, but the competition would like to encourage as many entries as possible in all competitions. The winner of a competition will be determined by matches between bots that were submitted to that specific competition.

We will use the same server code as the last few years: ACPC 2011 Server Code. The protocol remains unchanged. Please check the technical details page for agent implementation details.

Advancing our Objective

Pursuant to the competition objective, the details of each hand will be made completely public and available to anyone who wants to view them. This includes cards (both the community cards and each agent's hole cards, regardless of whether the hand made it to showdown), betting sequences, and so forth.

We will endeavor to make a benchmark server available after the competition. With this server, agents will be available to test against (for teams who competed in 2014), but the code will not be available to download. This will help competitors try alternative techniques to find out how they would have done.

Finite Computational Resources

Unfortunately we only have a limited amount of computing time available to generate the results for this year's competition. If we have insufficient resources to evaluate all pairs of competitors in a given competition, then we may need to take measures to ensure we get sufficient data to properly evaluate the competition. This will likely take the form of a reduced number of matches between some combinations of entrants.  We will compute a round robin using all the competitors with a small number of matches to get an initial estimate of agent performance. Top agents based on these initial results will then be given further computation time.

Heads-up Limit Texas Hold'em

Details of the event rules are listed below. Please see the links for further details of this particular variant.

Game Limit Texas Hold'em
Competition Format Series of heads-up duplicate matches
Hands Per Match 3000
Stack Sizes Infinite
Bet Sizes 10/20
Blind Sizes 5/10
Blind Structure Reverse blinds, no ascending blinds
Showdown Mucking No
Illegal Actions

Any illegal action is interpreted as a call.

Winner Determination

total bankroll.

Heads-up No-limit Texas Hold'em

Details of the event rules are listed below. Please see the links for further details of this particular variant.

Game No-Limit Texas Hold'em (Doyle's Game)
Competition Format Series of heads-up duplicate matches
Hands Per Match 3000
Stack Sizes 200 big blinds (400 small blinds)
Bet Sizes No limit
Blind Sizes 50/100
Blind Structure Reverse blinds, no ascending blinds
Showdown Mucking No
Illegal Actions

Any illegal action is interpreted as a call if raise is illegal. If it is a raise for an illegal amount, it is interpreted as the closest possible raise amount.

Winner Determination

Bankroll instant run-off and total bankroll

Ring Limit Texas Hold'em

In both 3 player competitions all entrants will play at least one match against every possible combination of opponents. This means there will be m choose 3 matches, where m is the number of entrants. Unless there is an abundance of computing resources available there will be only one match per configuration. This is to give the maximum number of hands possible in between each memory reset (as described in the duplicate section above) for opponent modeling purposes. Likely this will still result in less hands in between each memory reset than in the two player competitions.

Details of the event rules are listed below. Please see the links for further details of this particular variant.

Game 3 player ring Limit Texas Hold'em
Competition Format Series of 3-player duplicate matches
Hands Per Match 3000
Stack Sizes Infinite
Bet Sizes 10/20
Blind Sizes 5/10
Blind Structure No ascending blinds
Showdown Mucking No
Illegal Actions

Any illegal action is interpreted as a call.

Winner Determination

Bankroll instant run-off and total bankroll

3-player Kuhn Poker

Details of the event rules are listed below. Please see the links for further details of this particular variant.

Game Limit Kuhn Poker
Competition Format Series of 3-player duplicate matches
Hands Per Match 3000
Stack Sizes Infinite
Bet Size 1 chip
Ante Size 1 chip
Showdown Mucking No
Illegal Actions

Any illegal action is interpreted as a check/call.

Winner Determination

total bankroll.

Arbitration

If collusion or abuse of the network is detected, a Chair will present the issue to the arbiter of the competition (to be determined). In such cases, the arbiter's decision will be final. Abuse of the network is defined as any ethernet traffic that is not a communication of actions to the server. Collusion is cooperation between agents.

Agents that crash during matches may be disqualified from the competition at the discretion of the chairs in consultation with the arbiter. The rationale for this rule is that a high degree of variance could be injected into the total bankroll competitions if an agent crashes frequently against some opponents but not others.

Card Generation

The teams will collaborate to generate cards. Specifically, seeds from each team will be used to compose the random number seeds for the competition.

2017 Rules

The 2017 AAAI Computer Poker Competition will have two heads-up no-limit competitions, total bankroll and instant run-off no-limit Texas Hold'em.

Teams may enter one agent per competition (i.e., one agent for each winner determination rule). By default, if a team submits only one agent to a division with both a total bankroll and instant run-off event, then it will be entered into both competitions for that division. Teams may opt-out of this if they wish, but the competition would like to encourage as many entries as possible in all competitions. The winner of a competition will be determined by matches between bots that were submitted to that specific competition.

We will use the same server code as the last few years: ACPC 2011 Server Code. The protocol remains unchanged. Please check the technical details page for agent implementation details.

Advancing our Objective

Pursuant to the competition objective, the details of each hand will be made completely public and available to anyone who wants to view them. This includes cards (both the community cards and each agent's hole cards, regardless of whether the hand made it to showdown), betting sequences, and so forth.

We will endeavor to make a benchmark server available after the competition. With this server, agents will be available to test against (for teams who competed in 2017), but the code will not be available to download. This will help competitors try alternative techniques to find out how they would have done.  Agents will be available for two years after the competition concludes.

Finite Computational Resources

Unfortunately we only have a limited amount of computing time available to generate the results for this year's competition. If we have insufficient resources to evaluate all pairs of competitors in a given competition, then we may need to take measures to ensure we get sufficient data to properly evaluate the competition. This will likely take the form of a reduced number of matches between some combinations of entrants.  We will compute a round robin using all the competitors with a small number of matches to get an initial estimate of agent performance. Top agents based on these initial results will then be given further computation time.

Heads-up No-limit Texas Hold'em

Details of the event rules are listed below. Please see the links for further details of this particular variant.

Game No-Limit Texas Hold'em (Doyle's Game)
Competition Format Series of heads-up duplicate matches
Hands Per Match 3000
Stack Sizes 200 big blinds (400 small blinds)
Bet Sizes No limit
Blind Sizes 50/100
Blind Structure Reverse blinds, no ascending blinds
Showdown Mucking No
Illegal Actions

Any illegal action is interpreted as a call if raise is illegal. If it is a raise for an illegal amount, it is interpreted as the closest possible raise amount.

Winner Determination

Bankroll instant run-off and total bankroll

Arbitration

If collusion or abuse of the network is detected, a Chair will present the issue to the arbiter of the competition (to be determined). In such cases, the arbiter's decision will be final. Abuse of the network is defined as any ethernet traffic that is not a communication of actions to the server. Collusion is cooperation between agents.

Agents that crash during matches may be disqualified from the competition at the discretion of the chairs in consultation with the arbiter. The rationale for this rule is that a high degree of variance could be injected into the total bankroll competitions if an agent crashes frequently against some opponents but not others.

Card Generation

The teams will collaborate to generate cards. Specifically, seeds from each team will be used to compose the random number seeds for the competition.

2013 Rules

The 2013 AAAI Computer Poker Competition will have two heads-up limit competitions, two heads-up no-limit competitions, and two 3-player ring limit competitions.

If you have comments or suggestions, please post to the forums in the rules discussion section.

This year we will have the following competitions:

  • Two heads-up limit Texas Hold'em competitions
  • Two heads-up no limit Texas Hold'em competitions
  • Two 3-player ring limit Texas Hold'em competitions

Most of the rules for this year's competition will be the same as last year. In particular, we will use the same two winner determination rules: Bankroll instant run-off and total bankroll.

Teams may enter one agent per competition (i.e., one agent for each winner determination rule in each of the three game divisions). By default, if a team submits only one agent to a division then it will be entered into both competitions for that division. Teams may opt-out of this if they wish, but the competition would like to encourage as many entries as possible in all competitions. The winner of a competition will be determined by matches between bots that were submitted to that specific competition.

We will use the same server code as last year and 2011: ACPC 2011 Server Code. The protocol remains unchanged. Please check the technical details page for agent implementation details.

Advancing our Objective

Pursuant to the competition objective, the details of each hand will be made completely public and available to anyone who wants to view them. This includes cards (both the community cards and each agent's hole cards, regardless of whether the hand made it to showdown), betting sequences, and so forth.

We will endeavor to make a benchmark server available after the competition. With this server, agents will be available to test against (for teams who competed in 2013), but the code will not be available to download. This will help competitors try alternative techniques to find out how they would have done.

Finite Computational Resources

Unfortunately we only have a limited amount of computing time available to generate the results for this year's competition. If we have insufficient resources to evaluate all pairs of competitors in a given competition, then we may need to take measures to ensure we get sufficient data to properly evaluate the competition. This will likely take the form of some kind of qualifying round where competitors will be partitioned into smaller "mini-competitions". The top agents in the mini-competitions will be allowed to move forward. Another alternative is that we will compute a round robin using all the competitors but fewer matches to get an initial estimate of agent performance. Top agents based on these initial results can then be given further computation time.

Heads-up Limit Texas Hold'em

Details of the event rules are listed below. Please see the links for further details of this particular variant.

Game Limit Texas Hold'em
Competition Format Series of heads-up duplicate matches
Hands Per Match 3000
Stack Sizes Infinite
Bet Sizes 10/20
Blind Sizes 5/10
Blind Structure Reverse blinds, no ascending blinds
Showdown Mucking No
Illegal Actions

Any illegal action is interpreted as a call.

Winner Determination

Bankroll instant run-off and total bankroll.

Heads-up No-limit Texas Hold'em

Details of the event rules are listed below. Please see the links for further details of this particular variant.

Game No-Limit Texas Hold'em (Doyle's Game)
Competition Format Series of heads-up duplicate matches
Hands Per Match 3000
Stack Sizes 200 big blinds (400 small blinds)
Bet Sizes No limit
Blind Sizes 50/100
Blind Structure Reverse blinds, no ascending blinds
Showdown Mucking No
Illegal Actions

Any illegal action is interpreted as a call if raise is illegal. If it is a raise for an illegal amount, it is interpreted as the closest possible raise amount.

Winner Determination

Bankroll instant run-off and total bankroll

Ring Limit Texas Hold'em

In both 3 player competitions all entrants will play at least one match against every possible combination of opponents. This means there will be m choose 3 matches, where m is the number of entrants. Unless there is an abundance of computing resources available there will be only one match per configuration. This is to give the maximum number of hands possible in between each memory reset (as described in the duplicate section above) for opponent modeling purposes. Likely this will still result in less hands in between each memory reset than in the two player competitions.

Details of the event rules are listed below. Please see the links for further details of this particular variant.

Game 3 player ring Limit Texas Hold'em
Competition Format Series of 3-player duplicate matches
Hands Per Match 1000 (though we will see what our resources allow)
Stack Sizes Infinite
Bet Sizes 10/20
Blind Sizes 5/10
Blind Structure No ascending blinds
Showdown Mucking No
Illegal Actions

Any illegal action is interpreted as a call.

Winner Determination

Bankroll instant run-off and total bankroll

Arbitration

If collusion or abuse of the network is detected, a Chair will present the issue to the arbiter of the competition (to be determined). In such cases, the arbiter's decision will be final. Abuse of the network is defined as any ethernet traffic that is not a communication of actions to the server. Collusion is cooperation between agents.

Agents that crash during matches may be disqualified from the competition at the discretion of the chairs in consultation with the arbiter. The rationale for this rule is that a high degree of variance could be injected into the total bankroll competitions if an agent crashes frequently against some opponents but not others.

Card Generation

The teams will collaborate to generate cards. Specifically, seeds from each team will be used to compose the random number seeds for the competition.

Subcategories