Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time
April 5, 2017 02:54
April 5, 2017 02:54
April 5, 2017 02:54
April 5, 2017 02:54

DeepStack for Leduc Hold'em

DeepStack is an artificial intelligence agent designed by a joint team from the University of Alberta, Charles University, and Czech Technical University. In a study completed in December 2016, DeepStack became the first program to beat human professionals in the game of heads-up (two player) no-limit Texas hold'em, a commonly played poker game.

This project reimplements DeepStack as a player for heads-up no-limit Leduc holdem, a much simpler game than Texas hold'em.

DeepStack is built around two components:

  • An offline component that solves random poker situations (public game states along with probability vectors over private hands for both players) and uses them to train a neural network. After training, this neural network can accurately predict the value to each player of holding each possible hand at a given poker situation.
  • An online component which uses the continuous re-solving algorithm to dynamically choose an action for DeepStack to play at each public state encountered during gameplay. This algorithm solves a depth-limited lookahead using the neural net to estimate values at terminal states.

The strategy played by DeepStack approximates a Nash Equilibrium, with an approximation error that depends on the error of the neural net and the solution error of the solver used in continuous re-solving.


Running any of the DeepStack code requires Lua and torch. Torch is only officially supported for *NIX based systems (i.e. Linux and Mac OS X), and as such, we don't officially support installation of DeepStack on Windows systems. This page contains suggestions for using torch on a Windows machine; if you decide to try this, we cannot offer any help.

Connecting DeepStack to a server requires the luasocket package. This can be installed with luarocks (which is installed as part of the standard torch distribution) using the command luarocks install luasocket. Visualising the trees produced by DeepStack requires the graphviz package, which can be installed with luarocks install graphviz. Running the code on the GPU requires cutorch. Currently only version 1.0 is supported which can be installed with luarocks install cutorch 1.0-0.

The DeepStack player uses the protocol of the Annual Computer Poker Competition (a description of the protocol can be found here) to receive poker states and send poker actions as messages over a network socket connection. If you wish to play against DeepStack, you will need a server for DeepStack to connect to that acts as the dealer for the game; code for a server that fills this role is available through the ACPC here. For convenience, we include a copy of the ACPC dealer in the ACPCDealer/ directory, along with a game definition file (ACPCDealer/ so that the dealer can play Leduc hold'em.

If you would like to personally play against DeepStack, you will need a way of interacting with the server yourself; one such option is the ACPC GUI available from Dustin Morrill here.

Leduc Hold'em

Leduc Hold'em is a toy poker game sometimes used in academic research (first introduced in Bayes' Bluff: Opponent Modeling in Poker). It is played with a deck of six cards, comprising two suits of three ranks each (often the king, queen, and jack - in our implementation, the ace, king, and queen). The game begins with each player being dealt one card privately, followed by a betting round. Then, another card is dealt faceup as a community (or board) card, and there is another betting round. Finally, the players reveal their private cards. If one player's private card is the same rank as the board card, he or she wins the game; otherwise, the player whose private card has the higher rank wins.

The game that we implement is No-Limit Leduc Hold'em, meaning that whenever a player makes a bet, he or she may wager any amount of chips up to a maximum of that player's remaining stack. There is also no limit on the number of bets and raises that can be made in each betting round.


Documentation for the DeepStack Leduc codebase can be found here. In particular, there is a tutorial which introduces the codebase and walks you through several examples, including running DeepStack.

The documentation for code files was automatically generated with LDoc, which can be installed with luarocks install ldoc. To re-generate the docs, run ldoc . in the doc/ directory. If you wish to also generate documentation for local functions, run ldoc . -all instead.


Example implementation of the DeepStack algorithm for no-limit Leduc poker






No releases published


No packages published