lucasart wrote:I can make a few trivial changes to Stockfish and pass the similarity tests, any day!
Then please do it and release the source.
It may be interesting to know how much elo do you lose for it and if the engine that you get is stronger than DiscoCheck(note that 60% is not enough and you need similarity that is smaller than 55%)
The similarity is less than 55% and the elo fall is about 70-80 only.
This version of TwinFish is based on Stockfish dev 14 01 29 6:02PM (TimeStamp : 1391014933 )
The only changes made to reach a "<55%" similarity are a complete asymetric PST (based on Adam Hair values).
If you want to see the changes just search for "Robber" in the sources files.
There is only the source code, no binary.
If someone wants to compile good binaries it should be nice.
Don't forget : this version is only a joke and I don't steal Stockfish!
I'm very interested to see the result in similarity dendogram now !
Twinfish 0.07 is more related with Toga Hair than with Stockfish with the Don's Similarity Tester. And there is no code from Toga in it !!!
The similarity is less than 55% and the elo fall is about 70-80 only.
Although this a laudable attempt it is not really conclusive since cloners would never accept a 70-80 elo loss. The challenge is to do it without elo loss.
And so we are witnessing the death of similarity tester. Now that the cat is out of the bag I can confirm Ben's findings. During the PST-thread in the programmers forum I did some experiments with the several posted PST's and Piece Values and indeed they dreadfully bring down the similarity percentage without too much elo loss (20-30).
My point is that even the least competent developper can do a SF clone and pass similarity test by butchering the eval. I never said anything about not losing elo or losing no more than 70-80 elo. Of course if you butcher carelessly the eval just you lose elo, because SF is very well tuned.
With this TwinFish you not only prove my point, but prove an even stronger version of my point, quantifying the maximum elo loss. Basically using the similarity test to declare a closed source engine clean is simply naive.
Until the source code is revealed, nothing proves that a closed source engine contains no foreign code.
At least going open source means you have nothing to hide. It still puzzles me why people develop private engines (so you can't even run the similarity test?) or closed source engines when they are hundreds of elo below the top engines. Why do they fear to show us their code?
Theory and practice sometimes clash. And when that happens, theory loses. Every single time.
Rebel wrote:And so we are witnessing the death of similarity tester. Now that the cat is out of the bag I can confirm Ben's findings. During the PST-thread in the programmers forum I did some experiments with the several posted PST's and Piece Values and indeed they dreadfully bring down the similarity percentage without too much elo loss (20-30).
So folks be aware, cloners will find out anyway.
That is okay. There are more productive things that can be done using the tester and Polyglot. After all, the tester is just a mechanism to send positions and UCI commands to an engine. You can use the tester (with Polyglot to create logs) to evaluate large sets of positions with different engines. There are many things that can be measured, not just engine similarities.
I don't know if it's a rhetoric question, but some months ago I opened a poll asking the same question.
right now I'm on my tablet, but layer I can search for it and paste the link.
Two first meanings of the dutch word "leren":
1. leren [vc] (learn, larn, acquire) acquire or gain knowledge or skills.
2. leren [v] (teach, learn, instruct) impart skills or knowledge to.
Rebel wrote:And so we are witnessing the death of similarity tester. Now that the cat is out of the bag I can confirm Ben's findings. During the PST-thread in the programmers forum I did some experiments with the several posted PST's and Piece Values and indeed they dreadfully bring down the similarity percentage without too much elo loss (20-30).
So folks be aware, cloners will find out anyway.
Still, no false positives with Sim, only false negatives.