Yinsh

An advanced YINSH player - Tournament Results

The AI participated in a tournament where 16 YINSH AI competitors were matched up. After 7 matches (each match implied one game as white and one as black) using the Swiss pairing system, the following rankings were achieved:

Pos Name Score Gms W D L Rtg TPR
1 Pim Nijssen 13,0 7 6 1 0 1060 1350
2 Daniel Hennes 12,0 7 6 0 1 1050 1311
3 Thomas Debray 10,0 7 4 2 1 1030 1159
4 Michael Kaisers 8,0 7 4 0 3 1010 1050
5 Mohsin van Basten-Batenburg 8,0 7 2 3 2 1000 1000
6 Frederik Schadd 7,5 7 2 3 2 1000 1000
7 Michiel Moonen 7,5 7 2 3 2 1000 1000
8 Antonie Valkenberg 7,0 7 3 1 3 1000 1000
9 Sander Arts 7,0 7 3 1 3 1000 1000
10 Bob de Graaf 7,0 7 2 4 1 1010 1050
11 Kalle Fischer 6,5 7 2 3 2 1000 1000
12 Simon Jimenez Rodriguez 5,5 7 1 4 2 990 950
13 Yannick Soldati 5,5 7 2 1 4 980 898
14 Geert Gerritsen 4,5 7 1 3 3 980 898
15 Laurens van der Blom 2,0 7 0 2 5 950 689
16 Bart Mehlkop 1,0 7 0 1 6 940 650

 

It is interesting to see that the advanced AI (in the chart above noted as Thomas Debray) was the only computer player who played a draw against Pim. The only clearly better player against advanced AI was Daniel (partly due to the flaws and the lower depth search). One point was lost in a draw against Mohsin van Basten-Batenburg, where the advanced AI (when 2-2) scored an opponents row in stead of his own row (the 2 flaws combined), therefore making the opponent win that game. All the other games were won very easily, often with a score of 3-0 or 3-1.

 

Nr. White Black Result
1 Thomas Debray Bart Mehlkop 1-0
2 Pim Nijssen Thomas Debray 1/2-1/2
3 Mohsin van Basten-Batenburg Thomas Debray 1/2-1/2
4 Thomas Debray Michael Kaisers 1-0
5 Thomas Debray Daniel Hennes 0-1
6 Michiel Moonen Thomas Debray 0-1
7 Antonie Valkenberg Thomas Debray 0-1

 

We can conclude the AI's heuristics perform very well, and altough the low training time and short evolution, quite effective weights were achieved. If the flaws would be solved, the weights are trained more and an iterative deepening search (if desired using multithreading) is implemented, much better results can be expected.

Finally, the behaviour of the AI looks quite directed and coordinated, meaning they would probably be an interesting opponent for an advanced human YINSH player.

 

<<< Previous | Index | Next >>>