Bayesian pattern ranking for move prediction in the game of Go

  title={Bayesian pattern ranking for move prediction in the game of Go},
  author={David H. Stern and Ralf Herbrich and Thore Graepel},
  journal={Proceedings of the 23rd international conference on Machine learning},
We investigate the problem of learning to predict moves in the board game of Go from game records of expert players. In particular, we obtain a probability distribution over legal moves for professional play in a given position. This distribution has numerous applications in computer Go, including serving as an efficient stand-alone Go player. It would also be effective as a move selector and move sorter for game tree search and as a training tool for Go players. Our method has two major… 

Figures from this paper

Computing "Elo Ratings" of Move Patterns in the Game of Go

  • Rémi Coulom
  • Computer Science
    J. Int. Comput. Games Assoc.
  • 2007
A new Bayesian technique for supervised learning of move patterns from game records, based on a generalization of Elo ratings, which outperforms most previous pattern-learning algorithms, both in terms of mean log-evidence, and prediction rate.

Move Prediction in the Game of Go

A novel move prediction system is created, based on a naive Bayes model, which builds upon the work of several previous move prediction systems and achieves competitive results in terms of move prediction accuracy when tested on professional games and high-ranking amateur games.

Move Prediction in Go with the Maximum Entropy Method

This work uses the relative frequencies of local board patterns observed in game records to generate a ranked list of moves, and then applies the maximum entropy method (MEM) to the list to re-rank the moves.

Factorization Ranking Model for Move Prediction in the Game of Go

A Stochastic Gradient Decent algorithm is proposed to train a FBT model using expert game records, and two methods for fast computation of the gradient in order to speed up the training process are provided.

K-means Pattern Learning for Move Evaluation in the Game of Go

This paper proposes a method that reduce the complexity of Go by learning and extracting patterns from game records, which is more efficient and stronger than the baseline method chosen.

Only-One-Victor Pattern Learning in Computer Go

A supervised learning method, by introducing a new generalized Bradley-Terry model, named Only-One-Victor to learn patterns from game records, which outperforms Elo rating, a well-known best supervised pattern learning method.

Mastering the game of Go with deep neural networks and tree search

Using this search algorithm, the program AlphaGo achieved a 99.8% winning rate against other Go programs, and defeated the human European Go champion by 5 games to 0.5, the first time that a computer program has defeated a human professional player in the full-sized game of Go.

Global Context Influences Local Decisions

By mining thousands of positions online, it is shown that at some player levels the sequence of plays leading up to a local position is a stronger determinant of the next move than the position alone.

On Move Pattern Trends in a Large Go Games Corpus

A large corpus of game records of the board game of Go is processed and a way of extracting summary information on played moves is proposed, and statistically significant mappings to player attributes such as playing strength or informally perceived "playing style" are shown.

Bayesian Learning of Generalized Board Positions for Improved Move Prediction in Computer Go

Cortigo is described, a system that efficiently and autonomously learns useful generalizations for large state-space classification problems such as Go by using a hierarchical generative model loosely related to the human visual cortex to recognize Go board positions well enough to suggest promising next moves.



Modelling Uncertainty in the Game of Go

This work presents a simple conditional Markov random field model for predicting the pointwise territory outcome of a game and indicates that the model successfully learns to predict territory despite its simplicity.

Local Move Prediction in Go

A system that learns to predict local strong expert moves in the game of Go at a level comparable to that of strong human kyu players is presented and experiments suggest that local move prediction will be a significant factor in enhancing the strength of Go programs.

Computer Go: An AI oriented survey


This work has generated pattern databases for the game of Go that are associated to conditions that are external to the pattern, and believes that patterns associated to external conditions can be useful in other games.

Bayesian Generation and Integration of K-nearest-neighbor Patterns for 19x19 Go

The generation and utilisation of a pattern database for 19x19 go with the Knearest-neighbor representation and the improvement brought about by the use of this pattern database is estimated at 15 points on average, which is significant on go standards.

A family of algorithms for approximate Bayesian inference

This thesis presents an approximation technique that can perform Bayesian inference faster and more accurately than previously possible, and is found to be convincingly better than rival approximation techniques: Monte Carlo, Laplace's method, and variational Bayes.

Automatic Acquisition of Tactical Go Rules

Gogol is a rule-based computer Go program that uses a lot of reliable tactical rules and has metainformation about the correct use of the rules in global Go positions.

Computer Go

Space/time trade-offs in hash coding with allowable errors

Analysis of the paradigm problem demonstrates that allowing a small number of test messages to be falsely identified as members of the given set will permit a much smaller hash area to be used without increasing reject time.

Local move prediction in Go. 3rd International Conference on Computers and Games

  • Local move prediction in Go. 3rd International Conference on Computers and Games
  • 2002