Posts:
205
Registered:
Dec 16, 2001
From:
Moscow,Russia
Age:
56 Home page
Computer program match in Pente
Posted:
Nov 13, 2001, 11:49 AM
It was finished first game between two Pente Programs. Ninuki (was Black-first)author V. Barykin and lost against Noesis (was white) author M. Mammel. It was played opening e53 from semi-final at PBM dweebo-zabrodin. Because of library Noesis had made first moves without calculation , on the contrary Ninuki seems calculated first moves. After interesting fight Noesis won by captures We will see who win at second game, Noesis have to be first (Black)
This message was originally posted at Yahoo's pente group.
Posts:
260
Registered:
Dec 16, 2001
From:
Maryland
Age:
53 Home page
Re: Computer program match in Pente
Posted:
Nov 15, 2001, 9:55 PM
Here are the transcripts of the Pente computer competition. You can copy the game text between the dashed lines, paste it into notepad or wordpad and save it as a text file. Then you can open it with my program for viewing.
My program is nicknamed Noesis meaning some kind of knowledge and Barykin's program is called Ninuki which is an old name for Pente. Both programs played fairly well, Noesis had the advantage of using an opening book and then using more time to think for later moves.
Game1 Noesis (by M. Mammel) vs Ninuki (by V. Barykin)
M14 by Ninuki was a unusual opening, but after K13 and L14, it is a standard opening, offset 3 spaces vertically. My program uses an opening book and it can recognize offset positions so its moves 3-5 were from the opening book.
Move 5. ... L15 was not in the standard openings, usual is to M15 which might be met with O14. Now my program had to search: the competition allowed 2 hours for each program per game on a 600 MHz computer. That can be a lot of thinking! I had the program set on level 8 with a moderate amount of threat search beyond 8-ply (but not as deep as my standard program).
After 7. ... N13 my program might protect its pair at O13 which may give some initiative to its opponent, but instead it chose to capture at J15, making this game heavy in capture exchanges.
At move 11, the stone at N13 is a keystone that can be readily captured. My program does not immediately capture it but instead threatens it from another angle at L15. I think either approach would work fine.
Move 12 to O12 makes a nice 4-3 threat, Ninuki blocks it with M12. Next M14 threatens to capture M12 and a defending stone at N13! If Ninuki did block with N13, after it is captured and replaced, black wins with N14. Instead he moves on the end at K16.
On move 14, Noesis makes a three and Ninuki blocks with H15. If he blocked on the other side, then 15. N14 threatening the pair would win, the last two captures would not be difficult to force. 16. N15 is a winner, it will now threaten to capture a double keystone pair at N12 and M12 after making the three at N14.
17. ... O16 is a desperate attempt to survive, but the threatened pair is merely protected. 18. ... N16 is no real threat because of the capture of M15 and L16.
Ninuki program seemed to force a lot of capture play but was not able to gain the initiative, and so lost by captures.
This message was originally posted at Yahoo's pente group.
Moves 1-4 were made from my opening book. 6. M13 seems the obvious choice but maybe O10 then L12 is stronger as was done in the PBEM game dweebo vs zabrodin, but zabrodin won the game.
7. ... O11 my program sometimes likes to block a closed three so that it cannot extend. 8. Q8 seems to have too little offensive potential. Maybe K8 or at least N11? 8 ... L9 spoils his capture at N11. 9. J9 mainly to prevent Noesis from moving there. Is it better to capture at N11? Then 9. ... O12 10. M10, J9 11. N10, O10 12. J8!, J11 which is an interesting position. Instead, maybe 9. N10, J9 10. J8 which is similar to the previous position. 10. ... M11 threatens to capture K10 and M10 after moving to L11. 12. ... L11 clinches the game. 13. L12 - how about N11 instead? No, then Noesis moves to O12, H11, and G12 and can force captures.
This message was originally posted at Yahoo's pente group.