Gin Rummy EAAI Undergraduate Research Challenge

• EAAI-20 presentation (video, PDF slides, PDF handout, PDF URL tear sheets (cut between URLs))
• Poker AI Resources - Like Gin Rummy, Poker is an imperfect information card game. Unlike Gin Rummy, Poker has received much research attention. Here are potentially useful resources for learning from Poker research approaches:
  • "Depth First Learning: DeepStack" Model AI Assignment
  • "An Introduction to Counterfactual Regret Minimization" Model AI Assignment

• AI player submissions compatible with our base case are due one month before the AAAI-21 paper submission deadline, August 9th. (This will allow time to write up and submit research results.)
• Research papers will be submitted by the AAAI-21 submission deadline, September 9th, to a special undergraduate research paper track of EAAI-21, collocated with AAAI-21, and sharing the same submission system.
Note that authors must register as authors on the AAAI website between July 15th and August 30th
• Papers will be peer-reviewed. Accepted papers will be presented virtually at EAAI-21, February 3-4 and published in the AAAI-21 proceedings.
• Competition results will be shared privately with teams after the submission deadline and before EAAI-21, and will be shared publicly at EAAI-21 with a cash award from SIGAI to be announced.

• 50 faculty mentors, industry mentors, and students formed 14 teams which resulted in 14 Gin Rummy AI submissions, 14 paper submissions, and 13 papers accepted through peer review. This represents a new record participation for an EAAI mentored undergraduate research challenge.
• Accepted papers presented at EAAI:
  • Extracting Learned Discard and Knocking Strategies from a Gin Rummy Bot
    Benjamin Goldstein, Jean Astudillo Guerra, Emily Haigh, Bryan Cruz Ulloa, Jeremy Blum
  • A Heuristic Evaluation Function for Hand Strength Estimation in Gin Rummy
    Aqib Ahmed, Joshua Leppo, Michal Lesniewski, Riken Patel, Jonathan Perez, Jeremy Blum
  • Estimating Card Fitness for Discard in Gin Rummy
    Jacob Gallucci, Sarah Kettell, Richard Bowser, Christian Overton
  • A Deterministic Neural Network Approach to Playing Gin Rummy
    Viet Dung Nguyen, Dung Doan, Todd Neller
  • A Data-Driven Approach for Gin Rummy Hand Evaluation
    Sang Truong, Todd Neller
  • Opponent Hand Estimation in the Game of Gin Rummy
    Peter Francis, Hoang Anh Just, Todd Neller
  • Knocking in the Game of Gin Rummy
    Ryzeson Maravich, Taylor Neller, Todd Neller
  • Heisenbot: A Rule-Based Game Agent for Gin Rummy
    Matthew Eicholtz, Savanna Moss, Matt Traino, Christian Roberson
  • A Highly-Parameterized Ensemble to Play Gin Rummy
    Masayuki Nagai, Kavya Shrivastava, Kien Ta, Steven Bogaerts, Chad Byers
  • Random Forests for Opponent Hand Estimation in Gin Rummy
    Anthony Hein, May Jiang, Vydhourie Thiyageswaran, Michael Guerzhoy
  • Evaluating Gin Rummy Hands Using Opponent Modeling and Myopic Meld Distance
    Phoebe Goldman, Corey Knutson, Ryan Mahtab, Jack Maloney, Joseph Mueller, Richard Freedman
  • Opponent Hand Estimation in Gin Rummy Using Deep Neural Networks and Heuristic Strategies
    Bhaskar Mishra, Ashish Aggarwal
  • Modeling Expert and Folk Knowledge in a Heuristic-Based Reflex Agent for Gin Rummy
    Sarah Larkin, William Collicott, Jason Hiebel
• Gin Rummy AI Contest Results and Next Challenge Preview
• More about the next challenge: AI-Assisted Game Design

Two-Player Rules (video rules presentation and PDF slides):

• Materials: Standard 52-card deck without Jokers, Aces low and not adjacent to King in sequence
• Object: to be the first player to score 100 points from successive hands. The object of each hand is to go out with 10 or fewer card points unmelded, by playing one's hand by melding, laying off, or discarding, and, in doing so, scoring points based on the cards still in the hand(s) of opponent(s).
  • A meld is a set or a run. A set is 3 or 4 of the same rank card, e.g. 3C-3H-3S, or KC-KH-KS-KD. A run is 3 or more cards of the same suit in sequence, e.g. 5C-6C-7C, or 9H-TH-JH-QH-KH.
  • Laying off is adding card(s) to a meld.
  • Card points: An Ace counts 1 point. Each face card counts 10 points. Other cards count their rank number of points.
• Deal: 10 cards are dealt to each player. The first dealer is chosen at random. The dealing player alternates in subsequent hands.
• Players do not openly meld or lay off cards during the game.
• Going out: A player may go out if the total unmelded card points ("deadwood") doesn't exceed 10 points. A player must discard at the end of a turn, and indicates the desire to go out by discarding face down ("knocking").
  • The player going out spreads their cards and groups them in melds.
  • If all cards are melded the player has gone "gin" and scores an additional 25 points
  • Otherwise, if the player that knocked has deadwood, the opponent melds and may lay off cards to the melds of the knocking player. If, after melding and laying off, the opponent of the knocking player has fewer or the same deadwood points, the opponent wins the hand and scores an additional 25 point "undercut" bonus. Otherwise, the knocking player winds the hand.
  • The difference in deadwood is scored by the winner of the hand.
• Stockpile minimum: If two cards remain in the stockpile at the end of a turn where the player has not knocked, the entire hand is replayed with the same dealer.
• Detailed Gin Rummy rules are available at Pagat.com, but our rules differ in the following ways:
  • The deal alternates between players with each hand during a game.
  • North American play tends to award 25 points for gin and undercut bonuses, whereas in the U.K., gin and undercut bonuses are 20 and 10, respectively. To get a feel for Gin Rummy AI play using U.K. scoring, we recommend AI Factory's Gin Rummy Free app

Tiny URL: https://tinyurl.com/ginrummyai

Todd W. Neller