TalkChess.com

I mostly use engines for playing partners and analysis partners for my games and for when I am analyzing the games of others. I do like to run little mini-matches from time to time. So I do have quite a few engines and the following are boring me for various reasons. Correct me if I am missing something that would help me get some juice out of these engines:

Ktulu 8 and 9: This was a waste of money. It is good but there is no limit strength for use as a playing partner. And it can't really beat any other engine. I thought it might be good for endgame analysis, but ehh.

Onno: Was excited at first but no limit strength feature AND the MP version is taking forever!

Deep Sjeng 3 and WC2008: I am not sure why I have lost interest in it. It has the limit strength feature so I do play against it but it never gives me something new in analysis. Never! Not once! Every other engine that I use for analysis will come with a different opinion then the other engines but Deep Sjeng never has, EVER!

Stockfish 1.5: Deep Junior 2010 beat it handily and it has no limit strength feature. But it was free! (The limit strength thing is big for me...so I don't really deal with free engines that much...Glaurung eventually bored me as well.

Naum 4: Haven't bought it yet but no limit strength and Shredder 12 seems to be it's equal in strength.

Rybka 4: This is more of a consequence of all the rental rybka ideas! I love Rykba 3's analysis, and it has limit strength! I can't tell if Rybka 4 will have the limit strength or the different personalities as well. The waiting game and the uncertainty have made me come to the conclusion that I might just have to be content with Rybka 3 with no bug fixes. Kind of a shame actually!

Hi Tom

If you want a engine with a human style of play & analysis then try hiarcs or pro deo.

Hiarcs probably has the most realistic limit strength engines out there.

Even those engines not having the limit strength option can be used for producing different moves: just give the engines one/two seconds of thinking time and deactiivate the permanent brain. BTW: I do not understand your chess behaviour. For me only the best (deep) analysis is good enough to work with engines when studying my and other games.
Kurt

Question.

How does one implement limit strength? I am thinking of implementing this in GnuChess
as it is an often requested feature (and I like it myself very much).

The most obvious thing I can see is

(1) measure the relation of your engine's strength to its nps and put this in a lookup table.
(2) at startup do a quick self test to measure the nps on the current computer.
(3) in the function that gives wall clock time insert a time odds factor equal to the actual nps and the nps corresponding to the desired elo.

I have never tried Hiarcs but I tend to believe your statement as most people have said similar things about it. I would also recommend crafty. The old versions of crafty have a feature where you can limit the search depth and thereby limit its playing strength. I would absolutely recommend Pro Deo. I don't know if there is a more configurable engine out there. In the past I have had a good deal of fun making different personalities for it. It comes prepacked with many personalities and you can modify all of them or create new ones. Try creating a personality that is ultra aggressive and another that is very positional with a very high pawn value and a very high pawn structure value then play against them. Your style of play will definitely be challenged. Finally, the latest versions of Glaurung and Fruit are also very configurable. You can change the engine settings and then save the engine as a new and different engine. I am sure that I am forgetting some other configurable engines. Those I mentioned are just a few.

Michel wrote:Question.

How does one implement limit strength? I am thinking of implementing this in GnuChess
as it is an often requested feature (and I like it myself very much).

The most obvious thing I can see is

(1) measure the relation of your engine's strength to its nps and put this in a lookup table.
(2) at startup do a quick self test to measure the nps on the current computer.
(3) in the function that gives wall clock time insert a time odds factor equal to the actual nps and the nps corresponding to the desired elo.

This approach yields personalities that are tactically much stronger than desired.

Matthias.

Michel wrote:Question.

How does one implement limit strength? I am thinking of implementing this in GnuChess
as it is an often requested feature (and I like it myself very much).

The most obvious thing I can see is

(1) measure the relation of your engine's strength to its nps and put this in a lookup table.
(2) at startup do a quick self test to measure the nps on the current computer.
(3) in the function that gives wall clock time insert a time odds factor equal to the actual nps and the nps corresponding to the desired elo.

Here is what is done in Crafty ( and Pupsi):

Set SkillLevel to a number between 0 and 100.
When returning from the evaluation function do this:

reducedVal = (val * SkillLevel)/100 + Random(100 - SkillLevel)

where Random(x) is a random number between 0 and x.

SkillLevel 0 gives random values
SkillLevel 100 gives full strength.

So the lower the skill level, the more "noise" is added to the evaluation.

Works remarcably well!

Kind regards,
Jesper

jesper_nielsen wrote: Here is what is done in Crafty ( and Pupsi):

Set SkillLevel to a number between 0 and 100.
When returning from the evaluation function do this:

reducedVal = (val * SkillLevel)/100 + Random(100 - SkillLevel)

where Random(x) is a random number between 0 and x.

SkillLevel 0 gives random values
SkillLevel 100 gives full strength.

So the lower the skill level, the more "noise" is added to the evaluation.

Works remarcably well!

Kind regards,
Jesper

If you don't want to invest far more effort, the Crafty/Pupsi approach is certainly very good, far better than fiddling with time (alone).

Matthias.

Kurt Utzinger wrote:Even those engines not having the limit strength option can be used for producing different moves: just give the engines one/two seconds of thinking time and deactiivate the permanent brain. BTW: I do not understand your chess behaviour. For me only the best (deep) analysis is good enough to work with engines when studying my and other games.
Kurt

Of course I do not limit the strength of the engine when doing analysis. I only use the limit strength function as a playing partner. When analyzing games I set the engines to their full strength.

Why don't you just use engines which are around your strength at _full_
strength for that purpose? IMO this gives a much better match
than all those dumb down tries.
BTW in your original post you say that the new Junior is stronger
than Stockfish. Actually it is not, by a lot and I am sure it is much more
than 100 rating points below Stockfish.

Guenther