Depth-Limited Solving for
A fundamental challenge in imperfect-information games is that states do not have well-defined values. As a result, depth-limited search algorithms used in single-agent settings and perfect-information games do not apply. This paper introduces a principled way to conduct depth-limited solving in imperfect-information games by allowing the opponent to choose among a number of strategies for the remainder of the game at the depth limit. Each one of these strategies results in a different set of values for leaf nodes. This forces an agent to be robust to the different strategies an opponent may employ. We demonstrate the effectiveness of this approach by building a master-level heads-up no-limit Texas hold’em poker AI that defeats two prior top agents using only a 4-core CPU and 16 GB of memory. Developing such a powerful agent would have previously required a supercomputer.
Depth-Limited Solving for
Noam Brown, Tuomas Sandholm, Brandon Amos Computer Science Department Carnegie Mellon University email@example.com, firstname.lastname@example.org, email@example.com
noticebox[b]Preprint. Work in progress.\end@float
Imperfect-information games model strategic interactions between agents with hidden information. The primary benchmark for this class of games is poker, specifically heads-up no-limit Texas hold’em (HUNL), in which Libratus defeated top humans in 2017 . The key breakthrough that led to superhuman performance was nested solving, in which the agent repeatedly calculates a finer-grained strategy in real time (for just a portion of the full game) as play proceeds down the game tree [5, 26, 6].
However, real-time subgame solving was too expensive for Libratus in the first half of the game because the portion of the game tree Libratus solved in real time, known as the subgame, always extended to the end of the game. Instead, for the first half of the game Libratus pre-computed a fine-grained strategy that was used as a lookup table. While this pre-computed strategy was successful, it required millions of core hours and terabytes of memory to calculate. Moreover, in deeper sequential games the computational cost of this approach would be even more expensive because either longer subgames or a larger pre-computed strategy would need to be solved. A more general approach would be to solve depth-limited subgames in real time even in the early portions of a game.
The poker AI DeepStack does this using a technique similar to nested solving that was developed independently . However, while DeepStack defeated a set of non-elite human professionals in HUNL, it never defeated prior top AIs despite using over one million core hours to train the agent, suggesting its approach may not be sufficiently practical or efficient in domains like poker. We discuss this in more detail in Section 7. This paper introduces a different approach to depth-limited solving that defeats prior top AIs and is computationally orders of magnitude less expensive.
In perfect-information games, the value that is substituted at a leaf node of a depth-limited subgame is simply an estimate of the state’s value when all players play an equilibrium [34, 32]. For example, this approach was used to achieve superhuman performance in backgammon , chess , and Go [35, 36]. The same approach is also widely used in single-agent settings such as heuristic search [29, 24, 30, 15]. Indeed, in single-agent and perfect-information multi-agent settings, knowing the values of states when all agents play an equilibrium is sufficient to reconstruct an equilibrium. However, this does not work in imperfect-information games, as we demonstrate in the next section.
2 The Challenge of Depth-Limited Solving in Imperfect-Information Games
In imperfect-information games (also referred to as partially-observable games), an optimal strategy cannot be determined in a subgame simply by knowing the values of states (i.e., game-tree nodes) when all players play an equilibrium strategy. A simple demonstration is in Figure 0(a), which shows a sequential game we call Rock-Paper-Scissors+ (RPS+). RPS+ is identical to traditional Rock-Paper-Scissors, except if either player plays Scissors, the winner receives 2 points instead of 1 (and the loser loses 2 points). Figure 0(a) shows RPS+ as a sequential game in which acts first but does not reveal the action to . The optimal strategy (Minmax strategy, which is also a Nash equilibrium in two-player zero-sum games) for both players in this game is to choose Rock and Paper each with 40% probability, and Scissors with 20% probability. In this equilibrium, the expected value to of choosing Rock is , as is the value of choosing Scissors or Paper. In other words, all the red states in Figure 0(a) have value in the equilibrium. Now suppose conducts a depth-limited search with a depth of one in which the equilibrium values are substituted at that depth limit. This depth-limited subgame is shown in Figure 0(b). Clearly, there is not enough information in this subgame to arrive at the optimal strategy of , , and for Rock, Paper, and Scissors, respectively.
In the RPS+ example, the core problem is that we incorrectly assumed would always play a fixed strategy. If indeed were to always play Rock, Paper, and Scissors with probability , then could choose any arbitrary strategy and receive an expected value of . However, by assuming is playing a fixed strategy, may not find a strategy that is robust to adapting. In reality, ’s optimal strategy depends on the probability that chooses Rock, Paper, and Scissors. In general, in imperfect-information games a player’s optimal strategy at a decision point depends on the player’s belief distribution over states as well as the strategy of all other agents beyond that decision point.
In this paper we introduce a method for depth-limited solving that ensures a player is robust to such opponent adaptations. Rather than simply substituting a single state value at a depth limit, we instead allow the opponent one final choice of action at the depth limit, where each action corresponds to a strategy the opponent will play in the remainder of the game. The choice of strategy determines the value of the state. The opponent does not make this choice in a way that is specific to the state (in which case he would trivially choose the maximum value for himself). Instead, naturally, the opponent must make the same choice at all states that are indistinguishable to him. We prove that if the opponent is given a choice between a sufficient number of strategies at the depth limit, then any solution to the depth-limited subgame is part of a Nash equilibrium strategy in the full game. We also show experimentally that when only a few choices are offered (for computational speed), performance of the method is extremely strong.
3 Notation and Background
In an imperfect-information extensive-form game there is a finite set of players, . A state (also called a node) is defined by all information of the current situation, including private knowledge known to only one player. A unique player acts at state . is the set of all states in the game tree. The state reached after an action is taken in is a child of , represented by , while is the parent of . If there exists a sequence of actions from to , then is an ancestor of (and is a descendant of ), represented as . are terminal states for which no actions are available. For each player , there is a payoff function . If and , the game is two-player zero-sum. In this paper we assume the game is two-player zero-sum, though many of the ideas extend to general sum and more than two players.
Imperfect information is represented by information sets (infosets) for each player . For any infoset belonging to player , all states are indistinguishable to player . Moreover, every non-terminal state belongs to exactly one infoset for each player .
A strategy (also known as a policy) is a probability vector over actions for player in infoset . The probability of a particular action is denoted by . Since all states in an infoset belonging to player are indistinguishable, the strategies in each of them must be identical. We define to be a strategy for player in every infoset in the game where player acts. A strategy is pure is all probabilities in it are or . All strategies are a linear combination of pure strategies. A strategy profile is a tuple of strategies, one for each player. The strategy of every player other than is represented as . is the expected payoff for player if all players play according to the strategy profile . The value to player at state given that all players play according to strategy profile is defined as , and the value to player at infoset is defined as , where is player ’s believed probability that they are in state , conditional on being in infoset , based on the other players’ strategies and chance’s probabilities.
A best response to is a strategy such that . A Nash equilibrium is a strategy profile where every player plays a best response: , . A Nash equilibrium strategy for player is a strategy that is part of any Nash equilibrium. In two-player zero-sum games, if and are both Nash equilibrium strategies, then is a Nash equilibrium.
An imperfect-information subgame, which we refer to simply as a subgame, is a contiguous portion of the game tree that does not divide infosets. Formally, a subgame is a set of states such that for all , if and for some player , then . Moreover, if and and , then . If but no descendant of is in , then is a leaf node. Additionally, the infosets containing are leaf infosets. Finally, if but no ancestor of is in , then is a root node and the infosets containing are root infosets.
4 Multi-Valued States in Imperfect-Information Games
In this section we describe our new method, which we refer to as multi-valued states. For the remainder of this paper, we assume that player is attempting to find a Nash equilibrium strategy in a depth-limited subgame.
We begin by considering some unknown Nash equilibrium , and discuss what information about is required to reconstruct a Nash equilibrium strategy in a depth-limited subgame . Later, we will approximate this information and therefore be able to construct an approximate Nash equilibrium strategy in . We also assume for now that played according to prior to reaching .
As explained in Section 2, knowing the values of leaf nodes in the subgame when both players play according to (that is, for state and player ) is insufficient to construct the portion of that is in the subgame (unlike in perfect-information games), because it assumes would not adapt. But what if we do let adapt? Specifically, suppose we allow to choose any strategy in the entire game, while constraining to play according to outside of the subgame. Since is a Nash equilibrium strategy and can choose any strategy in the game, so clearly cannot do better than playing in the subgame. Thus, would play (or a different Nash equilibrium) in .
To simplify this process, upon reaching a leaf infoset , we could have choose a pure strategy for the remainder of the game that follows . So if there are possible pure strategies, would choose among actions upon reaching , where action would correspond to playing pure strategy for the remainder of the game. Since the choice of action would define a strategy for the remainder of the game and since is known to play according to outside , so the chosen action could immediately reward the expected value to , where is the state the players are in. Therefore, knowing these expected values would be sufficient to have play according to a Nash equilibrium in . This is stated formally in Proposition 1.
Proposition 1 adds the condition that we know for every root infoset . This condition is used if does not begin at the start of the game. Knowledge of is needed to ensure that any strategy that computes in cannot be exploited by changing their strategy earlier in the game. Specifically, we add a constraint that for all root infosets . This makes our technique safe:
Assume has played according to Nash equilibrium strategy prior to reaching a depth-limited subgame of a two-player zero-sum game. In order to calculate the portion of a Nash equilibrium strategy that is in , it is sufficient to know for every root infoset and for every pure strategy and every leaf node .
Other safe subgame solving techniques have been developed in recent papers, but those techniques require solving to the end of the full game [7, 17, 27, 5, 6] (except one , which we will compare to in Section 7). For ease of understanding, when considering the intuition for our techniques in this paper we suggest the reader first focus on the case where a subgame is rooted at the start of the game.
While we do not know , we can estimate it. And while it is impractical to know the value resulting from every pure strategy against this estimate of , in practice we may only need to know the values resulting from just a few strategies (that may or may not be pure). Indeed, in many complex games, the possible opponent strategies at a decision point can be approximately grouped into just a few “meta-strategies,” such as which of three lanes to choose in the real-time strategy game Dota 2. In our experiments, we find that excellent performance is obtained in poker with fewer than ten opponent strategies. In part, excellent performance is possible with a small number of strategies because the choice of strategy beyond the depth limit is made separately at each leaf infoset. Thus, if the opponent chooses between ten strategies at the depth limit, but makes this choice independently in each of leaf infosets, then the opponent is actually choosing between different strategies. This raises two questions. First, how do we estimate ? Second, how do we determine the set of strategies? We answer each of these in turn.
To estimate , we construct a blueprint strategy profile , which is a rough approximation of a Nash equilibrium for the entire game. The agent will never actually play according to . The blueprint is only used to estimate . There exist several methods for constructing a blueprint. One option, which achieves the best empirical results and is what we use, involves first abstracting the game [19, 12] and then applying the iterative algorithm Monte Carlo Counterfactual Regret Minimization . Several alternatives exist that do not rely on a distinct abstraction step [3, 16, 10].
We now discuss two different ways to select a set of strategies. Ultimately we would like the set of strategies to contain a diverse set of intelligent strategies the opponent might play, so that ’s solution in a subgame is robust to possible adaptation. One option is to bias the blueprint strategy in a few different ways. For example, in poker the blueprint strategy should be a mixed strategy involving some probability of folding, calling, or raising. We could define a new strategy in which the probability of folding is multiplied by (and then all the probabilities renormalized). If the blueprint strategy were an exact Nash equilibrium, then would still be a best response to . Thus, certainly qualifies as an intelligent strategy to play. In our experiments, we use this biasing of the blueprint strategy to construct a set of four opponent strategies on the second betting round. We refer to this as the bias approach.
Another option is to construct the set of strategies via self-play. The set begins with just one strategy: the blueprint strategy . We then solve a depth-limited subgame rooted at the start of the game and going to whatever depth is feasible to solve, giving only the choice of this strategy at leaf infosets. That is, at leaf node we simply substitute for . Let the solution to this depth-limited subgame be . We then approximate a best response assuming plays according to in the depth-limited subgame and according to in the remainder of the game. Since plays according to this fixed strategy, approximating a best response is equivalent to solving a Markov Decision Process, which is far easier to solve than an imperfect-information game. This approximate best response is added to the set of strategies that may choose at the depth limit, and the depth-limited subgame is solved again. This process repeats until the set of strategies grows to the desired size. This self-generative approach bears some resemblance to the double oracle algorithm  and recent work on generation of opponent strategies in multi-agent RL . In our experiments, we use this self-generative method to construct a set of ten opponent strategies on the first betting round. We refer to this as the self-generative approach.
One practical consideration is that since is not an exact Nash equilibrium, a generated strategy may do better than against . In that case, may play more conservatively than in a depth-limited subgame. To correct for this, one can “weaken” the generated strategies so that they do no better than against . Formally, if , we uniformly lower for by . (An alternative (or additional) solution would be to simply reduce for by some heuristic amount, such as a small percentage of the pot in poker.)
Once a strategy and a set of strategies have been generated, we need some way to calculate and store . Calculating the state values can be done by traversing the entire game tree once. However, that may not be feasible in large games. Instead, one can use Monte Carlo simulations to approximate the values. For storage, if the number of states is small (such as in the early part of the game tree), one could simply store the values in a table. More generally, one could train a function to predict the values corresponding to a state, taking as input a description of the state and outputting a value for each strategy. Alternatively, one could simply store and the set of strategies. Then, in real time, the value of a state could be estimated via Monte Carlo rollouts. We will present results for both of these approaches.
5 Nested Solving of Imperfect-Information Games
We use the new idea discussed in the previous section in the context of nested solving, which is a way to repeatedly solve subgames as play descends down the game tree . Whenever an opponent chooses an action, a subgame is generated following that action. This subgame is solved, and its solution determines the strategy to play until the next opponent action is taken.
Nested solving is particularly useful in dealing with large or continuous action spaces, such as an auction that allows any bid in dollar increments up to $10,000. To make these games feasible to solve, it is common to apply action abstraction, in which the game is simplified by considering only a few actions (both for ourselves and for the opponent) in the full action space. For example, an action abstraction might only consider bid increments of $100. However, if the opponent chooses an action that is not in the action abstraction (called an off-tree action), the optimal response to that opponent action is undefined.
Prior to the introduction of nested solving, it was standard to simply round off-tree actions to a nearby in-abstraction action (such as treating an opponent bid of $150 as a bid of $200) [14, 33, 11]. Nested solving allows a response to be calculated for off-tree actions by constructing and solving a subgame that immediately follows that action. The goal is to find a strategy in the subgame that makes the opponent no better off for having chosen the off-tree action than an action already in the abstraction.
Depth-limited solving makes nested solving feasible even in the early game, so it is possible to play without acting according to a precomputed strategy or using action translation. At the start of the game, we solve a depth-limited subgame (using action abstraction) to whatever depth is feasible. This determines our first action. After every opponent action, we solve a new depth-limited subgame that attempts to make the opponent no better off for having chosen that action than an action that was in our previous subgame’s action abstraction. This new subgame determines our next action, and so on.
We conducted experiments on the games of heads-up no-limit Texas hold’em poker (HUNL) and heads-up no-limit flop hold’em poker (NLFH). Appendix B reminds the reader of the rules of these games. HUNL is the main large-scale benchmark for imperfect-information game AIs. NLFH is similar to HUNL, except the game ends immediately after the second betting round, which makes it small enough to precisely calculate best responses and Nash equilibria. Performance is measured in terms of mbb/g, which is a standard win rate measure in the literature. It stands for milli-big blinds per game and represents how many thousandths of a big blind (the initial money a player must commit to the pot) a player wins on average per hand of poker played.
6.1 Exploitability Experiments in No-Limit Flop Hold’em (NLFH)
Our first experiment measured the exploitability of our technique in NLFH. Exploitability of a strategy in a two-player zero-sum game is how much worse the strategy would do against a best response than a Nash equilibrium strategy would do against a best response. Formally, the exploitability of is , where is a Nash equilibrium strategy.
We considered the case of betting 0.75 the pot at the start of the game, when the action abstraction only contains bets of 0.5 and 1 the pot. We compared our depth-limited solving technique to the randomized pseudoharmonic action translation (RPAT) , in which the bet of 0.75 is simply treated as either a bet of 0.5 or 1. RPAT is the lowest-exploitability known technique for responding to off-tree actions that does not involve real-time computation.
We began by calculating an approximate Nash equilibrium in an action abstraction that does not include the 0.75 bet. This was done by running the CFR+ equilibrium-approximation algorithm  for 1,000 iterations, which resulted in less than 1 mbb/g of exploitability within the action abstraction. Next, values for the states at the end of the first betting round within the action abstraction were determined using the self-generative method discussed in Section 4. Since the first betting round is a small portion of the entire game, storing a value for each state in a table required just 42 MB.
To determine a strategy in response to the 0.75 bet, we constructed a depth-limited subgame rooted after the 0.75 bet with leaf nodes at the end of the first betting round. The values of a leaf node in this subgame were set by first determining the in-abstraction leaf nodes corresponding to the exact same sequence of actions, except initially bets 0.5 or 1 the pot. The leaf node values in the 0.75 subgame were set to the average of those two corresponding value vectors. When the end of the first betting round was reached and the board cards were dealt, the remaining game was solved using safe subgame solving.
Figure 1 shows how exploitability decreases as we add state values (that is, as we give more best responses to choose from at the depth limit). When using only one state value at the depth limit (that is, assuming would always play according to the blueprint strategy for the remainder of the game), it is actually better to use RPAT. However, after that our technique becomes significantly better and at 16 values its performance is close to having had the 0.75 abstraction in the abstraction in the first place.
While one could have calculated a (slightly better) strategy in response to the 0.75 bet by solving to the end of the game, that subgame would have been about 10,000 larger than the subgames solved in this experiment. Thus, depth-limited solving dramatically reduces the computational cost of nested subgame solving while giving up very little solution quality.
6.2 Experiments Against Top AIs in Heads-Up No-Limit Texas Hold’em (HUNL)
Our main experiment uses depth-limited solving to produce a master-level HUNL poker AI called Modicum using computing resources found in a typical laptop. We test Modicum against Baby Tartanian8 , the winner of the 2016 Annual Computer Poker Competition, and against Slumbot , the winner of the 2018 Annual Computer Poker Competition. Neither Baby Tartanian8 nor Slumbot uses real time computation; their strategies are a precomputed lookup table. Baby Tartanian8 used about 2 million core hours and 18 TB of RAM to compute its strategy. Slumbot used about 250,000 core hours and 2 TB of RAM to compute its strategy. In contrast, Modicum used just 700 core hours and 16GB of RAM to compute its strategy and can play in real time at the speed of human professionals (an average of 20 seconds for an entire hand of poker) using just a 4-core CPU. We now describe Modicum and provide details of its construction in Appendix A.
The blueprint strategy for Modicum was constructed by first generating an abstraction of HUNL using state-of-the-art abstraction techniques [12, 20]. Storing a strategy for this abstraction as 4-byte floats requires just 5 GB. This abstraction was approximately solved by running Monte Carlo Counterfactual Regret Minimization for 700 core hours .
HUNL consists of four betting rounds. We conduct depth-limited solving on the first two rounds by solving to the end of that round using MCCFR. Once the third betting round is reached, the remaining game is small enough that we solve to the end of the game using an enhanced form of CFR+ described in the appendix.
We generated 10 values for each state at the end of the first betting round using the self-generative approach. The first betting round was small enough to store all of these state values in a table using 240 MB. For the second betting round, we used the bias approach to generate four opponent best responses. The first best response is simply the opponent’s blueprint strategy. For the second, we biased the opponent’s blueprint strategy toward folding by multiplying the probability of fold actions by 10 and then renormalizing. For the third, we biased the opponent’s blueprint strategy toward checking and calling. Finally for the fourth, we biased the opponent’s blueprint strategy toward betting and raising. To estimate the values of a state when the depth limit is reached on the second round, we sample rollouts of each of the stored best-response strategies.
The performance of Modicum is shown in Table 1. For the evaluation, we used AIVAT to reduce variance . Our new agent defeats both Baby Tartanian8 and Slumbot with statistical significance. Moreover, based on results from the previous experiment, our new agent is likely far less exploitable than Baby Tartanian8 and Slumbot. For comparison, Baby Tartanian8 defeated Slumbot by mbb/g, Libratus defeated Baby Tartanian8 by mbb/g, and Libratus defeated top human professionals by mbb/g. While our new agent is probably not as strong as Libratus, it was produced with less than of the computing resources and memory, and is not vulnerable to off-tree opponent actions on the first two betting rounds.
|Blueprint (No real-time solving)|
|Naïve depth-limited solving|
While the rollout method used on the second betting round worked well, rollouts may be significantly more expensive in deeper games. To demonstrate the generality of our approach, we also trained a deep neural network (DNN) to predict the values of states at the end of the second betting round as an alternative to using rollouts. The DNN takes as input a 34-float vector of features describing the state, and outputs four floats representing the values of the state for the four possible opponent strategies (represented as a fraction of the size of the pot). The DNN was trained using 180 million examples per player by optimizing the Huber loss with Adam , which we implemented using PyTorch . In order for the network to run sufficiently fast on just a 4-core CPU, the DNN has just 2 hidden layers and 64 nodes per layer. This achieved a Huber loss of 0.03. Using a DNN rather than rollouts resulted in the agent losing to Baby Tartanian8 by mbb/g. Thus, performance is slightly worse (which is to be expected). Nevertheless, this is still strong performance overall and demonstrates the generality of our approach.
7 Comparison to Prior Work
Section 2 demonstrated that in imperfect-information games, states do not have unique values and therefore the techniques common in perfect-information games and single-agent settings do not apply. This paper introduced a way to overcome this challenge by assigning multiple values to states. A different approach is to modify the definition of a “state” to instead be all players’ belief probability distributions over states, which we refer to as a joint belief state. This technique was previously used to develop the poker AI DeepStack . The evidence suggests that in the domain we tested on, using multi-valued states leads to better performance. This is exemplified by the fact that our approach defeats two prior top AIs with less than 1,000 core hours of computation. In contrast, while DeepStack defeated human professionals who were not specialists in HUNL, it was never shown to defeat prior top AIs even though it used over 1,000,000 core hours of computation. Still, there are benefits and drawbacks to both approaches. The right choice may depend on the domain and future research may change the competitiveness of either approach.
A joint belief state is defined by a probability (belief) distribution for each player over states that are indistinguishable to the player. In poker, for example, a joint belief state is defined by each players’ belief about what cards the other players are holding. Joint belief states maintain some of the properties that regular states have in perfect-information games. In particular, it is possible to determine an optimal strategy in a subgame rooted at a joint belief state independently from the rest of the game. Therefore, joint belief states have unique, well-defined values that are not influenced by the strategies played in disjoint portions of the game tree. Given a joint belief state, it is also possible to define the value of each root infoset for each player. In the example of poker, this would be the value of a player holding a particular poker hand given the joint belief state.
One way to do depth-limited subgame solving, other than the method we describe in this paper, is to learn a function that maps joint belief states to infoset values. When conducting depth-limited solving, one could then set the value of a leaf infoset based on the joint belief state at that leaf infoset.
One drawback is that because a player’s belief distribution partly defines a joint belief state, the values of the leaf infosets must be recalculated each time the strategy in the subgame changes. With the best domain-specific iterative algorithms, this would require recalculating the leaf infosets about 500 times. Monte Carlo algorithms, which are the preferred domain-independent method of solving imperfect-information games, may change the strategy millions of times in a subgame, making them incompatible with the joint belief state approach. In contrast, our multi-valued state approach requires only a single function call for each leaf node regardless of the number of iterations conducted.
Moreover, evaluating multi-valued states with a function approximator is cheaper and more scalable to large games than joint belief states. The input to a function that predicts the value of a multi-valued state is simply the state description (for example, the sequence of actions), and the output is several values. In our experiments, the input was 34 floats and the output was 4 floats. In contrast, the input to a function that predicts the values of a joint belief state is a probability vector for each player over the possible states they may be in. For example, in HUNL, the input is more than floats and the output is more than floats. The input would be even larger in games with more states per infoset.
Another drawback is that learning a mapping from joint belief states to infoset values is computationally more expensive than learning a mapping from states to a set of values. For example, Modicum required less than 1,000 core hours to create this mapping. In contrast, DeepStack required over 1,000,000 core hours to create its mapping. The increased cost is because a joint belief state value mapping is learning an inherently more complex function. The multi-valued states approach is learning the values of best responses to a particular strategy (namely, the approximate Nash equilibrium strategy ). In contrast, a joint belief state value mapping is learning the value of all players playing an equilibrium strategy given that joint belief state. As a rough guideline, computing an equilibrium is about 1,000 more expensive than computing a best response in large games .
On the other hand, the multi-valued state approach requires knowledge of a blueprint strategy that is already an approximate Nash equilibrium. A benefit of the joint belief state approach is that rather than simply learning best responses to a particular strategy, it is learning best responses against every possible strategy. This may be particularly useful in self-play settings where the blueprint strategy is unknown, because it may lead to increasingly more sophisticated strategies.
Another benefit of the joint belief state approach is that in many games (but not all) it obviates the need to keep track of the sequence of actions played. For example, in poker if there are two different sequences of actions that result in the same amount of money in the pot and all players having the same belief distribution over what their opponents’ cards are, then the optimal strategy in both of those situations is the same. This is similar to how in Go it is not necessary to know the exact sequence of actions that were played. Rather, it is only necessary to know the current configuration of the board (and, in certain situations, also the last few actions played).
A further benefit of the joint belief state approach is that its run-time complexity does not increase with the degree of precision. In contrast, for our algorithm the computational complexity of finding a solution to a depth-limited subgame grows linearly with the number of values per state.
We introduced a principled method for conducting depth-limited solving in imperfect-information games. Experimental results show that this leads to stronger performance than the best precomputed-strategy AIs in HUNL while using orders of magnitude less computational resources. Additionally, the method exhibits low exploitability.
This material is based on work supported by the National Science Foundation under grants IIS-1718457, IIS-1617590, and CCF-1733556, and the ARO under award W911NF-17-1-0082, as well as XSEDE computing resources provided by the Pittsburgh Supercomputing Center. We thank Thore Graepel, Marc Lanctot, and David Silver of Google DeepMind for helpful inspiration, feedback, suggestions, and support.
-  Michael Bowling, Neil Burch, Michael Johanson, and Oskari Tammelin. Heads-up limit hold’em poker is solved. Science, 347(6218):145–149, January 2015.
-  Noam Brown, Sam Ganzfried, and Tuomas Sandholm. Hierarchical abstraction, distributed equilibrium computation, and post-processing, with application to a champion no-limit texas hold’em agent. In Proceedings of the 2015 International Conference on Autonomous Agents and Multiagent Systems, pages 7–15. International Foundation for Autonomous Agents and Multiagent Systems, 2015.
-  Noam Brown and Tuomas Sandholm. Simultaneous abstraction and equilibrium finding in games. In Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI), 2015.
-  Noam Brown and Tuomas Sandholm. Baby Tartanian8: Winning agent from the 2016 annual computer poker competition. In Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence (IJCAI-16), pages 4238–4239, 2016.
-  Noam Brown and Tuomas Sandholm. Safe and nested subgame solving for imperfect-information games. In Advances in Neural Information Processing Systems, pages 689–699, 2017.
-  Noam Brown and Tuomas Sandholm. Superhuman AI for heads-up no-limit poker: Libratus beats top professionals. Science, page eaao1733, 2017.
-  Neil Burch, Michael Johanson, and Michael Bowling. Solving imperfect information games using decomposition. In AAAI Conference on Artificial Intelligence (AAAI), pages 602–608, 2014.
-  Neil Burch, Martin Schmid, Matej Moravčík, and Michael Bowling. AIVAT: A new variance reduction technique for agent evaluation in imperfect information games. 2016.
-  Murray Campbell, A Joseph Hoane, and Feng-Hsiung Hsu. Deep Blue. Artificial intelligence, 134(1-2):57–83, 2002.
-  Jiří Čermák, Branislav Bošansky, and Viliam Lisy. An algorithm for constructing and solving imperfect recall abstractions of large extensive-form games. In Proceedings of the 26th International Joint Conference on Artificial Intelligence, pages 936–942. AAAI Press, 2017.
-  Sam Ganzfried and Tuomas Sandholm. Action translation in extensive-form games with large action spaces: axioms, paradoxes, and the pseudo-harmonic mapping. In Proceedings of the Twenty-Third international joint conference on Artificial Intelligence, pages 120–128. AAAI Press, 2013.
-  Sam Ganzfried and Tuomas Sandholm. Potential-aware imperfect-recall abstraction with earth mover’s distance in imperfect-information games. In AAAI Conference on Artificial Intelligence (AAAI), 2014.
-  Sam Ganzfried and Tuomas Sandholm. Endgame solving in large imperfect-information games. In International Conference on Autonomous Agents and Multi-Agent Systems (AAMAS), pages 37–45, 2015.
-  Andrew Gilpin, Tuomas Sandholm, and Troels Bjerre Sørensen. A heads-up no-limit Texas hold’em poker player: discretized betting models and automatically generated equilibrium-finding programs. In Proceedings of the Seventh International Joint Conference on Autonomous Agents and Multiagent Systems-Volume 2, pages 911–918. International Foundation for Autonomous Agents and Multiagent Systems, 2008.
-  Peter E Hart, Nils J Nilsson, and Bertram Raphael. Correction to "a formal basis for the heuristic determination of minimum cost paths". ACM SIGART Bulletin, (37):28–29, 1972.
-  Johannes Heinrich and David Silver. Deep reinforcement learning from self-play in imperfect-information games. arXiv preprint arXiv:1603.01121, 2016.
-  Eric Jackson. A time and space efficient algorithm for approximately solving large imperfect information games. In AAAI Workshop on Computer Poker and Imperfect Information, 2014.
-  Eric Jackson. Targeted CFR. In AAAI Workshop on Computer Poker and Imperfect Information, 2017.
-  Michael Johanson, Nolan Bard, Neil Burch, and Michael Bowling. Finding optimal abstract strategies in extensive-form games. In Proceedings of the Twenty-Sixth AAAI Conference on Artificial Intelligence, pages 1371–1379. AAAI Press, 2012.
-  Michael Johanson, Neil Burch, Richard Valenzano, and Michael Bowling. Evaluating state-space abstractions in extensive-form games. In Proceedings of the 2013 International Conference on Autonomous Agents and Multiagent Systems, pages 271–278. International Foundation for Autonomous Agents and Multiagent Systems, 2013.
-  Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014.
-  Marc Lanctot, Kevin Waugh, Martin Zinkevich, and Michael Bowling. Monte Carlo sampling for regret minimization in extensive games. In Proceedings of the Annual Conference on Neural Information Processing Systems (NIPS), pages 1078–1086, 2009.
-  Marc Lanctot, Vinicius Zambaldi, Audrunas Gruslys, Angeliki Lazaridou, Julien Perolat, David Silver, Thore Graepel, et al. A unified game-theoretic approach to multiagent reinforcement learning. In Advances in Neural Information Processing Systems, pages 4193–4206, 2017.
-  Shen Lin. Computer solutions of the traveling salesman problem. The Bell system technical journal, 44(10):2245–2269, 1965.
-  H Brendan McMahan, Geoffrey J Gordon, and Avrim Blum. Planning in the presence of cost functions controlled by an adversary. In Proceedings of the 20th International Conference on Machine Learning (ICML-03), pages 536–543, 2003.
-  Matej Moravčík, Martin Schmid, Neil Burch, Viliam Lisý, Dustin Morrill, Nolan Bard, Trevor Davis, Kevin Waugh, Michael Johanson, and Michael Bowling. Deepstack: Expert-level artificial intelligence in heads-up no-limit poker. Science, 2017.
-  Matej Moravcik, Martin Schmid, Karel Ha, Milan Hladik, and Stephen Gaukrodger. Refining subgames in large imperfect information games. In AAAI Conference on Artificial Intelligence (AAAI), 2016.
-  John Nash. Equilibrium points in n-person games. Proceedings of the National Academy of Sciences, 36:48–49, 1950.
-  Allen Newell and George Ernst. The search for generality. In Proc. IFIP Congress, volume 65, pages 17–24, 1965.
-  Nils Nilsson. Problem-Solving Methods in Artificial Intelligence. McGraw-Hill, 1971.
-  Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in pytorch. 2017.
-  Arthur L Samuel. Some studies in machine learning using the game of checkers. IBM Journal of research and development, 3(3):210–229, 1959.
-  David Schnizlein, Michael Bowling, and Duane Szafron. Probabilistic state translation in extensive games with large action sets. In Proceedings of the Twenty-First International Joint Conference on Artificial Intelligence, pages 278–284, 2009.
-  Claude E Shannon. Programming a computer for playing chess. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 41(314):256–275, 1950.
-  David Silver, Aja Huang, Chris J Maddison, Arthur Guez, Laurent Sifre, George Van Den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda Panneershelvam, Marc Lanctot, et al. Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587):484–489, 2016.
-  David Silver, Julian Schrittwieser, Karen Simonyan, Ioannis Antonoglou, Aja Huang, Arthur Guez, Thomas Hubert, Lucas Baker, Matthew Lai, Adrian Bolton, et al. Mastering the game of Go without human knowledge. Nature, 550(7676):354, 2017.
-  Oskari Tammelin, Neil Burch, Michael Johanson, and Michael Bowling. Solving heads-up limit texas hold’em. In Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI), pages 645–652, 2015.
-  Gerald Tesauro. Programming backgammon using self-teaching neural nets. Artificial Intelligence, 134(1-2):181–199, 2002.
Appendix: Supplementary Material
Appendix A Details of How We Constructed the Modicum Agent
In this section we provide details on the construction of our new agent and the implementation of depth-limited subgame solving, as well as a number of optimizations we used to improve the performance of our agent.
The blueprint abstraction treats every poker hand separately on the first betting round (where there are 169 strategically distinct hands). On the remaining betting rounds, the hands are grouped into 30,000 buckets [2, 12, 20]. The hands in each bucket are treated identically and have a shared strategy, so they can be thought as sharing an abstract infoset. The action abstraction was chosen primarily by observing the most common actions used by prior top agents. We made a conscious effort to avoid actions that would likely not be in Baby Tartanian8’s and Slumbot’s action abstraction, so that we do not actively exploit their use of action translation. This makes our experimental results relatively conservative. While we do not play according to the blueprint strategy, the blueprint strategy is nevertheless used to estimate the values of states, as explained in the body of the paper.
We used unsafe nested solving on the first and second betting rounds, as well as for the first subgame on the third betting round. In unsafe solving , each player maintains a belief distribution over states. When the opponent takes an action, that belief distribution is updated via Bayes’ rule assuming that the opponent played according to the equilibrium we had computed. Unsafe solving lacks theoretical guarantees because the opponent need not play according to the specific equilibrium we compute, and may actively exploit our assumption that they are playing according to a specific strategy. Nevertheless, in practice unsafe solving achieves strong performance and exhibits low exploitability, particularly in large games .
In nested unsafe solving, whenever the opponent chooses an action, we generate a subgame rooted immediately before that action was taken (that is, the subgame starts with the opponent acting). The opponent is given a choice between actions that we already had in our action abstraction, as well as the new action that they actually took. This subgame is solved (in our case, using depth-limited solving). The solution’s probability for the action the opponent actually took informs how we update the belief distribution of the other player. The solution also gives a strategy for the player who now acts. This process repeats each time the opponent acts.
Since the first betting round (called the preflop) is extremely small, whenever the opponent takes an action that we have not previously observed, we add it to the action abstraction for the preflop, solve the whole preflop again, and cache the solution. When the opponent chooses an action that they have taken in the past, we simply load the cached solution rather than solve the subgame again. This results in the preflop taking a negligible amount of time on average.
To determine the values of leaf nodes on the first and second betting round, whenever a subgame was constructed we mapped each leaf node in the subgame to a leaf node in the blueprint abstraction (based on similarity of the action sequence). The values of a a leaf node in the subgame (as a fraction of the pot) was set to its corresponding blueprint abstraction leaf node. In the case of rollouts, this meant conducting rollouts in the blueprint strategy starting at the blueprint leaf node.
As explain in the body of the paper, we tried two methods for determining state values at the end of the second betting round. The first method involves storing the four opponent approximate best responses and doing rollouts in real time whenever the depth limit is reached. The second involves training a deep neural network (DNN) to predict the state values determined by the four approximate best responses.
For the rollout method, it is not necessary to store the best responses as 4-byte floats. That would use bits per abstract infoset, where is the number of actions in an infoset. If one is constrained by memory, an option is to randomize over the actions in an abstract infoset ahead of time and pick a single action. That single action can then be stored using a minimal number of bits. This means using only bits per infoset. This comes at a slight cost of precision, particularly if the strategy is small, because it would mean always picking the same action in an infoset whenever it is sampled. Since we were not severely memory constrained, we instead stored the approximate best responses using a single byte per abstract infoset action. In order to reduce variance and converge more quickly, we conduct multiple rollouts upon reaching a leaf node. We found the optimal number of rollouts to be three given our memory access speeds.
For the DNN approach, whenever a subgame on the second round is generated we evaluate each leaf node using the DNN before solving begins. The state values are stored (using about 50 MB). This takes between 5 and 10 seconds depending on the size of the subgame.
Starting on the third betting round, we always solve to the end of the game using an improved form of CFR+. We use unsafe solving the first time the third betting round is reached. Subsequent subgames are solved using safe nested solving (specifically, Reach subgame solving where the alternative payoffs are based on the expected value from the previously-solved subgame ).
To improve the performance of CFR+, we ignore the first 50% of iterations when determining the average strategy. Moreover, for the first 30 iterations, we discount the regrets after each iteration by where indicates the iteration. This reduces exploitability in the subgame by about a factor of three.
The number of CFR+ iterations and the amount of time we ran MCCFR varied depending on the size of the pot. For the preflop, we always ran MCCFR for 30 seconds to solve a subgame (though this was rarely done due to caching). On the flop, we ran MCCFR for 10 to 30 seconds depending on the pot size. On the turn, we ran between 150 and 1,000 iterations of our modified form CFR+. On the river, we ran between 300 and 2,000 iterations of our modified form of CFR+.
Appendix B Rules of the Poker Games
We experiment on two variants of poker: heads-up no-limit Texas hold’em (HUNL) and heads-up no-limit flop hold’em (NLFH).
In the version of HUNL we use in this paper, and which is standard in the Annual Computer Poker Competition, the two players ( and ) in the game start each hand with $20,000. The position of the two players alternate after each hand. There are four rounds of betting. On each round, each player can choose to either fold, call, or raise. Folding results in the player losing and the money in the pot being awarded to the other player. Calling means the player places a number of chips in the pot equal to the opponent’s share. Raising means that player adds more chips to the pot than the opponent’s share. A round ends when a player calls (if both players have acted). Players cannot raise beyond the $20,000 they start with, so there is a limited number of actions in the game. All raises must be at least $100, and at least as larger as the most recent raise on that round (if there was one).
At the start of each hand of HUNL, both players are dealt two private cards from a standard 52-card deck. must place $100 in the pot and must place $50 in the pot. A round of betting then occurs. When the round ends, three community cards are dealt face up that both players can ultimately use in their final hands. Another round of betting occurs, starting with this time. After the round is over, another community card is dealt face up, and another round of betting starts with acting first. Finally, one more community card is revealed to both players and a final betting round occurs starting with . Unless a player has folded, the player with the best five-card poker hand, constructed from their two private cards and the five community cards, wins the pot. In the case of a tie, the pot is split evenly.
NLFH is similar to HUNL except there are only two rounds of betting and three community cards.
Appendix C Proof of Proposition 1
Consider the augmented subgame structured as follows. contains and all its descendants. Additionally, for every root node (that is, a node whose parent is not in ), contains a node belonging to . If and are root nodes in and and share an infoset, then and share an infoset. begins with an initial chance node that reaches with probability proportional to the probability of reaching if tried to do so (that is, the probability of reaching it according to ’s strategy and chance’s probabilities).
At node , has two actions. The “alt” action leads to a terminal node that awards . The “enter” action leads to . From Theorem 1 in , a solution to is part of a Nash equilibrium strategy in the full game.
Now consider the depth-limited augmented subgame that is similar to but does not contain the descendants of . We show that knowing for every pure strategy and every leaf node is sufficient to calculate the portion of a Nash equilibrium strategy for that is in . That, in turn, gives a strategy in that is a Nash equilibrium strategy in the full game.
We modify so that, after ’s strategy is chosen, chooses a probability distribution over the pure strategies where the probability of pure strategy is represented as . This mixture of pure strategies defines a strategy . In this way, can pick any strategy because every strategy is a mixture of pure strategies. Upon reaching a leaf node , receives a reward of . Clearly can do no better than playing , because it is a Nash equilibrium and can play any strategy. Thus, any strategy plays in , when combined with outside of , must do at least as well as playing in the full game. ∎