StockFish LS with LimitStrength feature

[stevenaaus]

I wonder if no-one simply has approximations of the skill levels as ELO.
They are meant to be CPU indendpendent i think.

I don't think so. They simply used the MultiPV feature to chose bad moves.

Oh, ok That's a nice idea.
And i guess it means that simplifying the game will lead to worse moves.

[mcostalba]

I wonder if no-one simply has approximations of the skill levels as ELO.
They are meant to be CPU indendpendent i think.

I don't think so. They simply used the MultiPV feature to chose bad moves.

Oh, ok That's a nice idea.
And i guess it means that simplifying the game will lead to worse moves.

I think you both get it wrong

1) Skill levels are meant to be CPU independent and so could be approximated to ELO, it is just that none has done this measuring (it takes time and you should know what to do).

2) Simplifying the game will _not_ necessarily lead to worse moves: there is a "bad move" threshold in the algorithm that prevents picking up real blunders at medium/high skill levels.

[Alexander Schmidt]

1) Skill levels are meant to be CPU independent and so could be approximated to ELO, it is just that none has done this measuring (it takes time and you should know what to do).

Also in SkillLevels Stockfish calculates deeper with a faster CPU. As long as you try to avoid stupid blunders, you can only avoid them when you see it.

After some teste to tune the UCI Levels I measured that SkillLevel 19 with 1750 nps has the same stregth as SkillLevel 20 with 500 nps while SkillLevel 19 with 2000 nps is even stronger than SKillLevel 20 with 500 nps.

Hence the SkillLevels can't be CPU independent or am I wrong?

I guess the difference is marginal when Stockfish calculates deeper, but it is probably still there.

Alex

[mcostalba]

1) Skill levels are meant to be CPU independent and so could be approximated to ELO, it is just that none has done this measuring (it takes time and you should know what to do).

Also in SkillLevels Stockfish calculates deeper with a faster CPU. As long as you try to avoid stupid blunders, you can only avoid them when you see it.

After some teste to tune the UCI Levels I measured that SkillLevel 19 with 1750 nps has the same stregth as SkillLevel 20 with 500 nps while SkillLevel 19 with 2000 nps is even stronger than SKillLevel 20 with 500 nps.

Hence the SkillLevels can't be CPU independent or am I wrong?

I guess the difference is marginal when Stockfish calculates deeper, but it is probably still there.

Alex

You are stretching the argument too much. No human can beat SF at SkillLevel 19 ! And very few at Skill Level higher than 15.

For Skill Levels below 15 the ELO is largely CPU independent, it become to be sensible to CPU speed only at very high skill levels, but is not where a normal or also a strong player wants to set the bar.

Please consider skill levels below 15 that are the ones practically used.

[Don]

Hi Alexander,
the UCI_LimitStrength is not new, this comes first from Shredder, ...

I believe the first good implementation of this was in Rexchess. You could select by ELO the strength you wanted to play and the search was adjusted accordingly.

There have long been programs that have some sort of limit strength feature, such as the Novag machines which would cripple the quies search and/or do other stuff, but not in such a controlled way.

... and i am tried that with Tornado too in the past, but the adjustements of real ELO strength depends on human players is very difficult, i done that also with limit the nodes to search, but how can you know what nodes/sec to search for an ELO about 1800 ?

The base Idea is good, but prepare of near real strength for human players is very hard to find, because if later the engine is changed, you dont know how much the ELO are being changed too.

So it is much easier to make some handfull playing levels for novice players they can try to play, and if they won they can try high levels.

estimation strength with ELO is not a good Idea in my opinion.

[Alexander Schmidt]

You are stretching the argument too much. No human can beat SF at SkillLevel 19 ! And very few at Skill Level higher than 15.

Well, I can

1949 ELO in my hack is SkillLevel 19 at 1750 nps. Thats why the difference is important for my purpose.

Alex