Wasp 3.00 available ...

[carldaman]

Thanks John & Frank!

Cheers,
CL

[Frank Quisinsky]

Hi,

in the first of four 1.000 games test-runs 38% vs. stronger opponents (20 opponents from CEGT Rating List places 6-30).

Again and again ...
Nice to see that Wasp won vs. stronger opponents more fast games as to lost very fast.
Typical for Wasp.

A good style with all the different opening systems we have and many pieces on board.
Only one game lost undo move 50 vs. Fizbo.

Second test-run started ...

After the first 1.000 games around 30-35 Elo more as Wasp 2.60.
That's indeed very nice!

First 1.000 games can be download in "Wasp Beta Selection" on my site.

Strong work JOHN !!

Best
Frank

[matejst]

Frank,

I followed your testing of Wasp 3.0 (and previous testing too), and I would like to suggest that you change the opponents for further testing especially since you tested Wasp exclusively against stronger opponents, and it could deflate its rating. I believe that testing Wasp against a wider range of opponents could give a more precise picture of its strength. Anyway, thanks for the great work you do!

[Frank Quisinsky]

Hi,

1.
after my experience it can be 5-10 Elo.
Means, Wasp can be 5-10 Elo stronger if the opponents are more balanced to the Elo Wasp have.
So you are right. I made a lot of test-runs in times I am working on my own FCP Rating List and I know that it's really 5-10 Elo.

2.
CEGT 40 / 20 is around 40 / 4 with my conditions / hardware … can be also 5-10 Elo if Wasp is stronger with faster time controls?! Can be possible because Wasp lost more points in the very late middlegames as engines with around the same playing level or better … Wasp switched not often with more time the moves in that playing phase. Complete other situation if many pieces on board. Wasp found more as engines with around the same playing strength and switched more often to a better move. Most computer chess games building the end point (most blunders) for the final game result with move number 54 - 59 (chess programs for around 2900 - 3000 Elo). Move average goes higher with a higher average of opponents Elo. Or … the higher the Elo strength the higher the move average for lost and draw games.

Wasp strength and weaknesses are very own and not comparable to other engines I tested more in detail.

Believe me, I know the style of our TOP-50 since I made so many statistics in times FCP Rating List is running. For the moment Wasp is clearly my favorite engine because I like the style of Wasp a lot. The reason I wrote all the time … Wasp is the optimal chess program for a chess computer. With a fixed ply undo 12 Wasp is stronger as many TOP-10 programs. So with max. 100 Mhz Wasp produced for humans the most interesting style and the strength goes very high.

Not a wonder / mircales that programs by Ed Schröder are extremly strong with lesser fixed ply.
And fans of the older chess computers are often fans by Ed Schröder, all that is fully OK.



Thats the situation today. Stockfish, Houdini and Komodo produced most of the high strength in the very late middlegame and transformation into endgame.

3.
For Xhpos, Andscacs, Arasan and Ethereal, I haven't the current Elo in the table (not tested in CEGT). Can be 5 Elo too.

Not sure at all!
John wrote 50 Elo more (more right as in my testing).
After 800 CEGT games Wasp should be stronger as in my testing.

To test vs. stronger opponents have a reason!!

I try since years, in combination with a tool Ferdinand create for me, to find out an automatically system for weaknesses. For that idea I need many lost games. A bit comparable is a new Chessbase 14 feature : Tactical strength! Klaus Wlotzka is using for my chess computer tourney TMET the new feature from Chessbase.

I made a lot of experiments with my system (only with lost games). The problem is, that I can't public the idea because it's not good enough for the moment and maybe my idea is wrong. Working on it!

Best
Frank

[matejst]

Frank,

I read your post carefully and there are things I agree with. I use Wasp mostly to analyze openings and early middlegame, and I find that its analyses are sound. In many openings, it's the engine the most likely to find book moves and to display a very human-like game.

I do find the latest version a bit tactically weak, but it seems to be a problem John's programs have since the last version of Zarkov. Wasp's pawn play in the middlegame seems to be a bit too shy, it often avoids to play with its pawns, to open the position, and starts moving pieces around without plan.

But the biggest problem seems to be a of lack endgame knowledge, which is crucial when playing against engines who reach 6, even 10 plies deeper in the late middlegame and which can use some kind of tablebases. That's why Wasp transition game is so bad.

At fixed depth, Wasp 2.6 seems to be the strongest Wasp so far, while Wasp 2.01 seems a bit weaker than the others.