Logistello

The othello program Logistello, written by Michael Buro as PhD thesis work for the University of Paderborn in Germany, has been considered the strongest othello player in the world, although the newer program Hannibal seems to be about as strong as of 1997. He is currently at the NEC Research Institute in New Jersey.

In August 1997, Logistello defeated the human world champion, Takeshi Murakami, by 6-0 in a match at long time controls. The Othello Match of the Year describes the event.

In January 1998, Logistello retired from active tournament play, leaving Hannibal as the strongest active program.

references

Logistello
Michael Buro
This web page lists facts about Logistello.

An overview of Logistello (1997, 3 pages)
Michael Buro
Compressed postscript. Concisely describes Logistello's evaluation function, selective search, opening book learning, and development history. This is a good place to start learning about the program.

Statistical feature combination for the evaluation of game positions (1995, 10 pages)
Michael Buro
Compressed postscript. There is an online abstract at JAIR; you can also download the article from there. Compares different statistical regression techniques to find the best one for constructing Logistello's evaluation function.

ProbCut: An effective selective extension of the alpha-beta algorithm (1995, 7 pages)
Michael Buro
Compressed postscript. Despite the title, this is about selective pruning, not search extension. Subtrees which are shown, by shallow search, to have little probability of affecting the search are pruned away. This is similar to the null-move heuristic popular in chess programs, except that the pruning cutoffs are estimated from data by a statistical technique.

Techniken für die Bewertung von Spielsituationen anhand von Beispielen (1994)
Michael Buro
Compressed postscript. Written in German. Buro's PhD dissertation. The title means "Methods for the evaluation of game positions using examples".

An evaluation function for othello based on statistics (1997, 6 pages)
Michael Buro
Compressed postscript. A description of Logistello's old evaluation function, since superceded by the stronger one described below.

Experiments with Multi-ProbCut and a new high-quality evaluation function for othello (1997, 9 pages)
Michael Buro
Compressed postscript. Describes a new table-estimation technique for constructing Logistello's evaluation function. The technique takes into account correlations between the features, resulting in a much stronger evaluator. Also gives improvements on the ProbCut pruning algorithm.

Toward opening book learning (1997, 5 pages)
Michael Buro
Compressed postscript. Learning an opening book so your program doesn't repeat its mistakes. This is an offline method that searches for promising deviations from lines that lost in the past.