Re: Help Bot Advice adopted.

Sanny wrote:

> That game was played by EASY level. Heres a game by Advance Level
> where Zebediah gave lots of sacrifices and win the game in just 34
> moves.
> 
> In this game Zebediah was a knight ahead but later sacrificed its
> knight to bring a quick end.

You mean that Getclub gifted him a knight and then played like a patzer. 
This play is basically insane. There is something wrong with either the 
evaluation function or the search algorithm.

> 1. d2-d4{8} Ng8-f6{0}
> 2. c2-c4{2} c7-c5{0}
> 3. d4-d5{2} b7-b5{0}
> 4. f2-f3{22} e7-e5{0}

4. ... e6 looks like the best bet here, but b4, d6 or amusingly going 
against HelpBots earlier (and generally true) advice Qa5+ is quite 
playable here! But e5 isn't all that bad.

> 5. c4-b5{572} Bf8-d6{1506}

Words fail me. It is hard to imagine how any half decent evaluation 
function can see this as a good move even at ply 1. Blocking in your own 
QP with a bishop is doing whites job for him.

Bd6 is Shredders 15th ranked choice of 26 moves so pick a move at random 
and you have better than even chances of improving Getclubs play here.

5. ... d6 and black is pretty well on track for a strong game.
5. ... e4 looks interesting too (not sure I would risk it).


> 6. e2-e4{104} Nf6-e4{1978}

And now the final coup de grace. Suicide chess anyone ???

This clinches it! There is something fundamentally wrong with GetClub 
chess. It is happy to give away a N for a P. This move is so bad I had 
to extend the search window to find the move ranking which was 20th out 
of 28. So a random choice of move would be better 75% of the time!!

Shredder so dislikes the position of the Bd6 that it plays Bc7 as first 
preference, with O-O or a6 both close behind.

> 7. f3-e4{14} Qd8-h4{2114}
> 8. Ke1-e2{8} Qh4-e4{2258}
> 9. Ke2-f2{6} c5-c4{2254}

I reckon it is playing for these spite checks. P for N swap is useless.
And c4 is pretty bad. O-O or Qg6 would be playable.

> 12. Nb1-d2{28} f7-f5{2234}

O-O here is needed.

> 17. Nc4-d2{108} Ke8-g8{1284}

And O-O here is pretty well fatal. Why castle now when it doesn't help?
17. ... d6 consolidates a badly lost position.

 > 31. Qd7-e8{Q}{14} Kh8-h7{1954}

This lousy half baked notation is another serious fault which you refuse 
to do anything about. If you want to be taken seriously at least try to 
implement some of the helpful suggestions for minor improvements.

It is a pawn when it moves to the 8th rank and becomes a queen.

31. d7-e8Q{Q}{14} Kh8-h7{1954}

Would do at a pinch. Otherwise engine annotation ignores the illegal 
moves and the game after a promotion is complete gibberish.

Humans would also like to have castles shown as O-O not Ke1-g2 and 
captures shown with the standard "x" capture notation Nf6xe4 not Nf6-e4 
and mark checks with +.

> What were the wrong moves that Zebediah was able to win despite it was
> Advance Level which thinks for 1/2 an hour?

It doesn't matter how long it thinks.
What matters is how well and how deep!

It might be instructive to hobble a copy to play with a pure 1 ply 
search and show the evaluation for all moves from a few test positions 
then we can see what misconceptions it has about the position.

It would be handy if you could give an indication of how long it takes 
your program to make a 1,2,3 etc ply search starting from the initial 
board position then we can see how its search strategy scales.

The current crop of top flight engines can get one ply deeper with a 
2-3x increase in computation time in most positions (this is a lot 
better than the 4-6x you can get with a basic alpha-beta search).

Regards,
Martin Brown
** Posted from http://www.teranews.com
**