ShashChess

[AndrewGrant]

Ray might of course maintain the list outside of CCRL.

With no Torch, Ethereal and Dragon that were supplied to CCRL testers for use only in CCRL testing, not possible in its current format. It would be free engines only, no private or commercial.

For Ethereal at least, I have no problem with you using your CCRL copy in your own personal capacity. And I'll be sure to include you in any future releases.

[amchess]

Statistically, it does change, as at 1m+1s much more pruning is done.
In addition, ShashChess dynamically evaluates position quality to direct the search: it should be obvious that it needs more time...
The 200 matches in our match are all from unbalanced positions, not dead draws.
Indeed, your result shows that, to try to beat ShashChess, you have to use Stockfish at its peak performance (bullet monster) and ShashChess at its minimum, but not even that!
In fact, if you try to run the same test between Stockfish and itself you will see obvious randomness in the final result.
In fact, the original search algorithm of Stockfish (and of any decent modern chess engine) inherently has its own randomness.

It is scary how much you actually know about proper testing despite refusing to do it yourself. Run a 50k game test against Stockfish at your time control of choice and share the PGNs. The claim that "It should be obvious that it needs more time..." is speculation that needs to be backed up by evidence and the burden of proof is on you.

ShashChess, basically, is Stockfish + a selectivity technique given precisely by Shashin's theory. This technique consists of a classification of each root position to direct the search algorithm.
The basis of computational theory teaches that if there is more processing to be done, the program needs more time, having to execute, trivially, more instructions.
One simply has to compare the code to see that, and I am the author of the engine.
The test is only to verify that, above a threshold time cadence of play, this qualitative analysis compensates for the lower speed. This is precisely the purpose of the test performed.
I also reiterated that my purpose is not to make a bullet monster, which is completely useless to both OTB and Corr chess players.
The purposes are different.
I respect the community, but this is not, unfortunately, reciprocal.

[Pali]

Statistically, it does change, as at 1m+1s much more pruning is done.
In addition, ShashChess dynamically evaluates position quality to direct the search: it should be obvious that it needs more time...
The 200 matches in our match are all from unbalanced positions, not dead draws.
Indeed, your result shows that, to try to beat ShashChess, you have to use Stockfish at its peak performance (bullet monster) and ShashChess at its minimum, but not even that!
In fact, if you try to run the same test between Stockfish and itself you will see obvious randomness in the final result.
In fact, the original search algorithm of Stockfish (and of any decent modern chess engine) inherently has its own randomness.

It is scary how much you actually know about proper testing despite refusing to do it yourself. Run a 50k game test against Stockfish at your time control of choice and share the PGNs. The claim that "It should be obvious that it needs more time..." is speculation that needs to be backed up by evidence and the burden of proof is on you.

ShashChess, basically, is Stockfish + a selectivity technique given precisely by Shashin's theory. This technique consists of a classification of each root position to direct the search algorithm.
The basis of computational theory teaches that if there is more processing to be done, the program needs more time, having to execute, trivially, more instructions.
One simply has to compare the code to see that, and I am the author of the engine.
The test is only to verify that, above a threshold time cadence of play, this qualitative analysis compensates for the lower speed. This is precisely the purpose of the test performed.
I also reiterated that my purpose is not to make a bullet monster, which is completely useless to both OTB and Corr chess players.
The purposes are different.
I respect the community, but this is not, unfortunately, reciprocal.

I have no interest in Shashin theory until it's proven to work.
2 + 2 = 4?
You are not the author of the engine - It is the entire SF development team + you.
Your test is 200 games long, it only verifies that neither engine wrecks the other.
You cannot claim long time control superiority until you SPRT ShashChess against Stockfish / Do a fixed games test with Elo difference well outside the error margin and share the PGNs for reproducibility.

If you are not willing to provide conclusive evidence such as the one above, then let me know and I will stop posting on the thread.

[amchess]

It is obvious that I am the author of Shashin's theory in Stockfish and the rest is from the community.
A long time sprt test on 15000 games is simply impractical.
Therefore, I adopted a different strategy I think more intelligent since it also makes use of the chess concept of positional traits.
200 LTC games out of 100 different positional tracts ended with that gap is largely sufficient evidence that ShashChess is superior at least at that time cadence.
Any person with rudiments of statistics can confirm this, but there is none so deaf as those who will not hear and blind as those who will not see.
Unfortunately, most programmers do not understand this because they are not rated chess players
Now, I also find myself having to restate obviousness.

[Viz]

It is obvious that I am the author of Shashin's theory in Stockfish and the rest is from the community.
A long time sprt test on 15000 games is simply impractical.
Therefore, I adopted a different strategy I think more intelligent since it also makes use of the chess concept of positional traits.
200 LTC games out of 100 different positional tracts ended with that gap is largely sufficient evidence that ShashChess is superior at least at that time cadence.
Any person with rudiments of statistics can confirm this, but there is none so deaf as those who will not hear and blind as those who will not see.
Unfortunately, most programmers do not understand this because they are not rated chess players
Now, I also find myself having to restate obviousness.

what a pathetic lie, all of this from start to finish.
If it's so sufficient with your superior understanding, why do you merge all stockfish patches? This question was asked to you like dozens of times and yet no one ever heard an answer.
And answer is simple. Without actually merging actual stockfish gainers you wouldn't be able to keep up small gap that allows you to fluke out tests of even 200 games because you yourself are incapable of producing a single elo of gain in your entire life.
So you shit on stockfish methodology 24/7 while fully merging whatever it produces, this is the definition of being a pathetic scumbag.
This is always puzzling me, all cloners are like "60+0.6 tests are baaaad" and than proceed to merge every single freaking commit inherited from this 60+0.6 tests.
How much of a mental gymnastics you have to do to justify this?
Not to mention that when you are getting called out you go for racial slurs and death threats, which is simply amazing.
And then you cry about getting banned for "superiority of shashchess" while in reality both me and you know the reason, like wishing death to stockfish maintainers (or other people) is smth that should get you out of the community.
The fact that such a prick as you has some platform to talk is really shameful. You have no skills, no achievements, nothing. You are a complete fake from start to finish and a really dishonest one.
Ah btw stop merging my patches into shashchess, they are not good since they are tested at 60+0.6 UHO openings and not your methodology, I look at the code and always wonder what my patches that fail yellow at LTC do there.

[Rebel]

It is obvious that I am the author of Shashin's theory in Stockfish and the rest is from the community.
A long time sprt test on 15000 games is simply impractical.
Therefore, I adopted a different strategy I think more intelligent since it also makes use of the chess concept of positional traits.
200 LTC games out of 100 different positional tracts ended with that gap is largely sufficient evidence that ShashChess is superior at least at that time cadence.
Any person with rudiments of statistics can confirm this, but there is none so deaf as those who will not hear and blind as those who will not see.
Unfortunately, most programmers do not understand this because they are not rated chess players
Now, I also find myself having to restate obviousness.

https://github.com/egh-s/emulate

Simple command line tool to discover how many games are needed to proof a program change.

Example -
Games : 1000
Rounds: 20
Will produce the following result of 20 x 1000 game matches.

Code: Select all

488-511 (48.9%)
499-501 (49.9%)
513-487 (51.3%)
496-503 (49.6%)
485-514 (48.5%)
490-510 (49.0%)
502-498 (50.2%)
507-492 (50.8%)
500-500 (50.0%)
476-524 (47.6%)
512-487 (51.3%)
513-487 (51.3%)
506-494 (50.6%)
489-511 (48.9%)
477-522 (47.8%)
514-485 (51.5%)
487-512 (48.8%)
474-525 (47.5%)
515-484 (51.5%)
487-512 (48.8%)

One can see the enormous fluctuation between the matches, one may think (as the match with a 51.5% score suggests) to have found an improvement of 10 elo but rerunning the same match easily can produce a regression of 10 elo points, see for instance the result of the match with a 48.5% score, the one with a 47,8% score is even worse. Conclusion, playing 1000 games in engine-engine self play is meaningless.

Type (A) for Again to try next emulations, 2000, 5000, 10,000 etc.

And you will notice that you need at least 50,000 games before all the 20 match scores start to settle close to 50%.

[Uri Blass]

It is obvious that I am the author of Shashin's theory in Stockfish and the rest is from the community.
A long time sprt test on 15000 games is simply impractical.
Therefore, I adopted a different strategy I think more intelligent since it also makes use of the chess concept of positional traits.
200 LTC games out of 100 different positional tracts ended with that gap is largely sufficient evidence that ShashChess is superior at least at that time cadence.
Any person with rudiments of statistics can confirm this, but there is none so deaf as those who will not hear and blind as those who will not see.
Unfortunately, most programmers do not understand this because they are not rated chess players
Now, I also find myself having to restate obviousness.

I am a rated chess player and I disagree with you.

200 LTC games can be enough to believe that one engine is better in one of 2 cases:
1)There is a huge difference in playing strength at LTC(for example you get 50 wins for Shashin and 1 loss and 149 draws).
2)There is a different good reason to explain why one engine is better(games have moves and you may explain based on the moves why you think that one engine is better).

I saw no explanation from you about 1 or 2 and you even did not release the time control and the games that you believe that ShashChess is superior based on them.

I can imagine ways that are not games to show evidence that A is better than B when you need games only to check for no bugs(for example if there is a speed improvement) but I saw no explanation from you about it.

Engine A is significantly faster than B in some practical test positions is a good evidence only if there is a good reason to believe that A is not significantly slower than B in finding the right move in different important positions.

[BrendanJNorman]

...being a pathetic scumbag.

The fact that such a prick as you has some platform to talk is really shameful. You have no skills, no achievements, nothing. You are a complete fake from start to finish and a really dishonest one.

Come on Mods, are these types of personal attacks allowed here?

Dude makes a new account and comes here just to attack people.

Surely against the forum rules.

[BrendanJNorman]

I have no interest in Shashin theory until it's proven to work.

I have no idea (nor do I care) if Shashin theory adds Elo to Stockfish (I doubt it).

But what I do like about ShashChess is that using a Tal, Petrosian or Capablanca setting produces quite varied play, and not *that* much weaker than other top engines.

For example, I used the Tal setting and the Capablanca setting to discover 2 completely different approaches (piece setups/middlegame plans) to a certain variation of the Caro Kann, both plans which I have now tested with success in my own OTB games.

This makes ShashChess useful to me as a CHESS PLAYER.

Don't care if programmers think it isn't useful or adding anything.

My 2 cents.

[AndrewGrant]

But what I do like about ShashChess is that using a Tal, Petrosian or Capablanca setting produces quite varied play, and not *that* much weaker than other top engines.

Proof is needed for the claim that anything amchess has committed into Stockfish is anything but noise.

Slapping the words "Tal", "Petrosian", and "Capablanca" around is meaningless. You can introduce some meaningful style with a list of if statements that don't do anything. A smarter man would train neural networks and filter the data in some good way to induce "style"