New Windows Shredder on the way with......

[fern]

I hope this will not be considered a break of confidentiality, but today, in a chat with Stephan, father of Shredder, I was acquainted with the great novelty -great for me- that the new W version into which he works will have a new and very developed feature that never has been adequately fulfilled -in my opinion- by any engine at the present: I mean the weakened levels designed to imitate such and such kind of player.
To date, a supposedly -say- 1600 Elo player in fact play as a unthinkable monster that play one move as GM and another as an idiot. That is not a 1600 player. A 1600 is always an idiot. (joke)
I am sure Stephan will deliver. I am fed up of supposedly 1800 GrandMaster personalities kicking asslike Fisher until they give for nothing a rook as a moron.
just wait regards
Fern.

[Nimzovik]

Yawn............... I truly do hope that Shredder can beat the snot out of Rybka........ Ain't gonna happen tho............ Sigh...................... I truly hope it comes with a unique to computer style that understands blocked postions and fortresses. Yah. Right.........

[GothicChessInventor]

I am fed up of supposedly 1800 GrandMaster personalities kicking asslike Fisher until they give for nothing a rook as a moron.

Yes, it is a tough problem to decide how to remove I.Q. from a program without executing obvious errors.

One thing I programmed into my program (The Sniper, http://www.uschess.org/msa/MbrDtlMain.php?12528567 )was the following:

1. Quanitify the amount of "weakening", assign it 2 numbers, call them A and B.
2. Count legal moves at the root, call this N.
3. Perform the search, slightly less deeply.
4. At a leaf node, change the score by +/- B, randomly
5. For a % of the N moves determined by A, change the sign of the score.

Slight changes to the B parameter lets the program play strong, just finding 2nd and 3rd and 4th best moves, with usually results that are not catastrophic.

Changes to the A parameter are more effective at weakening the program so it won't give away pieces, but it will allow you to set up tactical shots if you can see through the inferior play.

So, by tuning these weights, you can simulate that which Shredder may be attempting. I was doing this waaaaaaay back in 1988, and it had some good results.

[Gian-Carlo Pascutto]

Deep Sjeng had this feature for a long time. It's fairly advanced in that it tries to overlook the same kind of moves that a weak player often overlooks.

[fern]

I hope Stephan -and others- see your ideas. Or perhaps you should talk to Stephan. Just e-mail him to his mail support address you can find in his site.
My own pet idea about this is simple: just make the engine see less deep and with less width. It is precisely why we humans commit mistakes.

My best
fern

[fern]

You are right your engine is the best in that, currently. And I believe some of the reason is that in lesser levels it see less deep. No more than 4/5 ply or 6/7 according how much it is weakened.
nevertheless, still does not give the real feeling of facing a human 1800 player or so.
Of course you approach is zillions times better than in grandmaster, where you see personalities marked as 2300 doing very stupid things
Fern

[Gian-Carlo Pascutto]

My own pet idea about this is simple: just make the engine see less deep and with less width. It is precisely why we humans commit mistakes.

My best
fern

It's important to make the engine not miss simple moves. For example, it should never miss a recapture. But it might miss the fact that it doesn't HAVE to recapture. That's the typical kind of mistakes from a weaker player.

I'm curious how it works in Shredder.

[M ANSARI]

I have yet to see a program that can properly implement "weakening". The main problem is that humans and engine programs are weak in different fields. I would estimate that a 2000 ELO human player is about 1800 ELO tactically strong and 2200 ELO positionally strong ... at least in comparison to engines. Engines are the opposite ... a 2000 ELO program is usually 2400 ELO tactically strong and 1600 ELO positionally strong.

This ofcourse can make playing a program which is "weakened" a frustrating experience.

[Marek Soszynski]

One obviously poor implementation is in time taken. Usually, human players can find positionally sound moves relatively quickly but are slowed down considerably in tactical situations (including simple endgames). It's completely unrealistic for reduced-strength engines to play their moves very quickly (unless it's early in the opening, or a straightforward recapture or check response).

[Matthias Gemuh]

I have yet to see a program that can properly implement "weakening". The main problem is that humans and engine programs are weak in different fields. I would estimate that a 2000 ELO human player is about 1800 ELO tactically strong and 2200 ELO positionally strong ... at least in comparison to engines. Engines are the opposite ... a 2000 ELO program is usually 2400 ELO tactically strong and 1600 ELO positionally strong.

This ofcourse can make playing a program which is "weakened" a frustrating experience.

I agree to 99.99 % .

Matthias.