Why do chess computers often misevaluate the french defense?

Sort:
MARattigan

Stockfish doesn't necessarily play too well in basic endgames without its closing book (EGTB).

 

If you understand KBNK you can play a series of set up positions both as Black and White and steadily accumulate an advantage in terms of the number of moves played to mate. It is possible this is deliberate when it's playing Black but it also cedes moves as White. I haven't had it long enough to say but I'd guess it also can't learn to play any particular endgame better with practice.

 

But there is apparently Stockfish and Stockfish. I recently downloaded a version and it cedes moves slower than lichess level 8 (which I believe is also Stockfish). It may also be a problem with set up, because no matter how much clock time I give my own version, I can't get it to use more than a minute for the whole endgame.

 

(All I have to do now is find some way of getting through to a basic endgame in the first place.)

BronsteinPawn
mcris wrote:
MARattigan wrote:
mcris wrote:

The (first) problem with this game are in the first 11 moves, theory you say. I am not interested in human theory, but in Stockfish play, so give me a game where Stockfish plays all moves.

The second problem is that Stockfish says it wins with 22...Nf3!  LOL

Stockfish is playing from human theory. It's opening play has been largely determined by generations of grandmasters past.

 

When I first got a computer program that played chess (on a BBC acorn) it hammered me every time until I came up with a non-standard opening with which I could beat it. It wasn't the program that was beating me It was it's book - a product of human understanding.

 

Stockfish plays very well without opening book, especially in the French defense.

You traitor, why do you support the aliens? Support the human race dude, come on.

mcris

Yes, I strongly support the human race and I am in favor of forbidding all chess engines and smartphones.

MARattigan
Optimissed wrote:

Obviously, if a chess programme misevaluates any position then it's programmed wrongly.

End of story.

Can you program it rightly?

MARattigan

As an extreme example of mis-evaluation try the Engine Analysis (Sherlock Holmes chess boardette picture, bottom left) on this theoretically drawn White to play position:


I believe the engine is Stockfish.

RubenHogenhout
MickinMD schreef:

Personally, I favor the Classical Steinitz Variation 4. e5 where Black effectively loses a tempo with the almost automatic 4...Nfd7 or else posts the N awkwardly.  But even with 4 Bb5, I've run both the 32- and 64-bit versions of Stockfish 8 on Lucas Chess and can't find anyplace where Black is favored after 1.e4 e6 2.d4 d5 3.Nc3 Nf6 4.Bg5 Bb4.  Chess.com favors White by 0.18 p's:

 

Using various depths from 6 to 20 on Lucas Chess didn't find any significant difference:

 

This first line is intersting in the Mac Cutcheon variation. But mostly in games I have seen white does not take on f6 and g7 but play 6. Bd2  after 6....Bxc3 and 7.bxc3 Ne4  oft later folows after some time Nxd2 Kxd2 and the position of white looks not so impressive.

 

RubenHogenhout
stassneyking schreef:

And I've heard it's quite common for computers to misevaluate openings at least in the first ten moves or so. Computers hardly ever like King's Indian positions and they hate my favorite Czech Benoni, but really there are many interesting things that can happen.

Interesting thing. Many Gm s also do not play the King Indian and it is by some seen als difficult to play for black. But still very strong players like Radbjov is playing the King -
Indian and winning with it. I saw recently a very strong game from him were he plays ....e4!  an sac you often see in the K I defence to open de diagonal for the g7 bishop. It was a crusching win for black. So the King Indian is still far from easy for white.

 

RubenHogenhout
pfren schreef:

Yet another French, same server (centaur game).

 

 

 

- 11...Qa4 is the engine's top suggestion. It's a logical move, which was played many times in correspondence games- mostly draws, as the players were trusting their engines.

- 11...Qa4 is in all likelihood losing by force, due to the brilliant plan/pawn sac 14.f5!! which former Correspondence World Champion Aleksandr Surenovic Dronov found- and I am pretty sure he found it himself, not his engine! In the game, I did little more than following Dronov's idea, the position just plays itself. Please notice that Dronov declined at his 15th move winning an exchange with Ng3, since it is not consistent with his plan and gives Black more defensive resources (although white did win, after a few inaccuracies by both sides, in the only game I'm aware this happening).

- Old fox and French defence maniac GM Gerald Hertneck has a terific feel of his favorite defence's intricacies, and played the one and only plan that gives Black a decent game (white has at most a slight advantage): 11...c4! and ...Qa5-c7-f7-f5, to stop white's potentially lethal kingside initiative.

 

Needless to say that if I ever have some question about a French defence position, I would certainly ask mr. Hertneck, and not Stockfish, which is rated some 900+ points higher!

 

Impressive game on this Qa4 with I Always hold as a strange move althought it was theorie and considered to be allright.  This f5 is a fine idea.  Whats is by the way the refutation of 10....cxd4  ?

 

Megabyte
logozar escreveu:
mcris wrote:

Just give me a game you won again Stockfish, Houdini or Komodo and I'll believe you.

Analysis with Stockfish:

 


Maybe I'm missing something, but Chess.com's Stockfish says that after 22. ... Nc4! White has no attack. If White insists with something like 23. Qe1?! then 23. ... b5! 24. Ng3?? Nxg3 25. Qxg3 Nd2+ 26. Ka1 Qxd4, meaning white's knight is hanging. 

BronsteinPawn

I would rather spend that mining crypto currency. At least I will get like 2

8 bucks per week with that.

sammyperez

I was looking the exact opposite question. I am watching the top engines championship here in chess.com and I was surprised that StockFish played the French a lot, I mean, a lot.  But not the classical lines. If you look carefully engines misvaluates many black opening. But it is a misinterpretation to say they misvaluate. In chess white has a slight advantage, in all the openings, so it is strange that an engine gives equality at the opening, in general white is a little better in most opening

LogoCzar

Tracking concepts, looking for language anchors, monitoring where i'd be likely to get feedback.

LogoCzar

*quality* feedback... no need to pay attention elsewhere, *with guidelines* like "don't check the forums" and while i did enjoy conversation, when there was a productive purpose, i was less convinced that it was helpful for my development, while my books were giving me concepts, patterns, facts, and so forth!!! social, less so... at times.

and when people noticed my learning oriented approach, they brought quality feedback to me!

LogoCzar

Talk with everyone? I'm not sure.

LogoCzar

it was an experiment, and people sure can be critical "for no reason" it seems

rmz21
Bide Türkçe olsa :)