Stockfish 5 - repetition evaluation

[lucasart]

The reason SF does what it does is that fixing this has been repeatedly shown to be a small Elo loss.

The question still open : Is it better to have an engine a couple of Elo points stronger but a weird repetition detection ?

Clearly, the emphasis has been on squeezing out the elo points for engine matches, at the cost of veracity in analysis.

Ugh!

There. You know my opinion!

+1 It's the 'ratings rat race' effect at work. Very unfortunate.
SF is also non-commercial, so don't expect this to be fixed any time soon.

Regards,
CL

But SF, unlike commercial engines, is open source. So you can fix it yourself, if this is really important to you.

[Mike S.]

On your distant planet, everyone is a programmer?

[BBauer]

Yeah, the best thing of stockfish is
Open Source.

Could you please provide the patch?

Thanks Bernhard

[Joerg Oster]

Yeah, the best thing of stockfish is
Open Source.

Could you please provide the patch?

Thanks Bernhard

Take a look here: https://github.com/joergoster/Stockfish ... ae4cce5b7d

[carldaman]

The reason SF does what it does is that fixing this has been repeatedly shown to be a small Elo loss.

The question still open : Is it better to have an engine a couple of Elo points stronger but a weird repetition detection ?

Clearly, the emphasis has been on squeezing out the elo points for engine matches, at the cost of veracity in analysis.

Ugh!

There. You know my opinion!

+1 It's the 'ratings rat race' effect at work. Very unfortunate.
SF is also non-commercial, so don't expect this to be fixed any time soon.

Regards,
CL

But SF, unlike commercial engines, is open source. So you can fix it yourself, if this is really important to you.

Thanks for confirming my initial point. Any SF feature wish list discussion usually ends up with the programmers' asking the endusers to do it themselves.

[BBauer]

Thanks for the patch,
but I should perhaps change something else too.

This is what I got from compiling position.cpp:
position.cpp: In member function 'bool Position::is_draw() const':
position.cpp38: error: 'RootPly' is not a member of 'Search'
&& ++rep >= 2 + (gamePly - i < Search::RootPly))
^
make[2]: *** [position.o] Error 1

Kind regards
Bernhard

[gladius]

Yeah, the best thing of stockfish is
Open Source.

Could you please provide the patch?

Thanks Bernhard

Take a look here: https://github.com/joergoster/Stockfish ... ae4cce5b7d

Thanks Joerg. I see this was tested at 15+0.05 before and was neutral, let's give it a run at 60s as well to see what it looks like.

As an aside, open source doesn't mean "please provide the patch". It means the ability for someone to provide the patch. Chess engines in particular are full of tradeoffs, and this isn't the worst one in my opinion. It does cause confusion though, so it's worth giving the fix another try.

[Leto]

There hasn't been a new dev version for Stockfish since the 10th of the month. Is the Stockfish team struggling to find a way to improve the engine, or are members of the team on vacation?

[syzygy]

The reason SF does what it does is that fixing this has been repeatedly shown to be a small Elo loss.

The question still open : Is it better to have an engine a couple of Elo points stronger but a weird repetition detection ?

The answer by the SF maintainers is clear.

Those that do not agree with that answer should know that SF's source is available to all.

[BBauer]

Sorry, I didn't the other changes.
Now it works great.

After Qa7+ Kf8 Qa5 stockfish plays Re3+

So you do not have to be a programer, its enough to be a copier and a paster and a chess patzer.

Kind regards
Bernhard