Frist impressions to SF 1.8.0 JA ...

[Frank Quisinsky]

Hi there,

I saw a lot of the around 200 games I have with w32 (40 in 10). SF have problems vs. Zappa and Hiarcs. Very interesting are the games vs. Hiarcs 13.1 because after my statistic the engine is the most aggressive engine under the TOP-20 so far.

After all I saw ...
I think SF 1.8.0 is stronger but lost many of agressiveness. Perhaps all is a little bit to selective. Nice to see in the Zappchess games. SF comes with advantage in the games but don't see positional moves and Zappa made the points in middlegame.

In endgames SF 1.8 is stronger as version 1.7.1 ... I am sure from version to version stronger ... but not in the early middlegame.

Perhaps code with knowledge for more aggressinveness don't work or is delete?

After a position test a friend make (2 minutes per move for each position) he the same impressions I have.

I am not a very strong chess player and this are only my first impressions.

Other example:
SF 1.7.1 produced a lot of very fast games up to mate. Not the test version and it seems the 1.8.0 can produced it too.

My feeling:
Version 1.8.0 produced an other playing style as the great version 1.7.1.

But let wait of more games ...
SF 1.8.0 is still running in SWCR ... so far I have 200 games only.

Best
Frank

[mcostalba]

Hi Frank,

thanks for your observations, are very interesting indeed. I am a weak player, so I am not able to comment on games. I can only add that the difference with 1.7.1, if any, should be quite small, around 15 ELO, perhaps 20 if we are lucky, so it is very difficult to spot from looking at the games.

I can tell you that we have not removed any "aggressivness" from evaluation, actually evaluation is almost the same of 1.7.1 and very little has changed there.

Instead we have tweaked the search a bit, but my impressions is that we are now at a stall point with search and if we want to gain more we need new evaluation's ideas. Because search is already very aggressive and efficient a small improvment in evaluation could have a nice quite important effect on total strength.

The bottom line is that IMHO the search is already top class, the evaluation instead could be improved...but it is much more easier to say then to do

[Frank Quisinsky]

Hi Marco,

you added the message to the same time I edit my first message.

Have a look what I wrote to middlegame / endgame. I am sure you don't see at because I edit my message.

Look here (one good example):
http://www.talkchess.com/forum/viewtopi ... 56&t=35234

Later I will add the results from SF 1.8.0. I am sure you can directly see what I mean if you look on these Statistics 1/2 only.

Not easy for us to see that (I believe my ELO is around 2.050-2.150), for me too. Statistics can help us.

And only my impressions, must not be right after 200 games only.

But absolutley clear ...
In endgames the Stockfish is stronger with each new version.

Perhaps it have to do with null/move selectiv parameter.

In endgames SF should have more null/move as in middlegame. Not sure what the programmers do here (not looking in sources, I think I will not understand this part of sources).

Thanks for your new version and all the hard work you and the other two Stockfish programmers do. Now it's very hard to improved such a strong program and great engine ... clear for myself.

Perhaps what I wrote can help a little bit.

Best
Frank

[M ANSARI]

Interesting and I will have to take a look. I might try out some LTC games. SF 1.7.1 was much better than it rating at LTC games, which usually means that it has a solid evaluation but maybe lacks search speed. Maybe 1.8 has given up some knowledge to make the engine faster. That is what I think R4 has done vs R3. I did a LTC match using SF 1.71 against R4 and SF did a lot better than in blitz games (I think scored 40%). SF 1.7.1 was superior to R4 in wing attacks or "mass piece" attacks ... seems like R4 is missing code to notice that pieces are being massed against it and will depend on search for that (which in many cases will be outside the radar of danger).