Könnte ja noch einen draufsetzen ...

Spark 130 Elo bei dieser Auswertung bis Matt in 59 stärker.
150 Elo bei Matt in 40 stärker
190 Elo bei Matt in 35 stärker

Stärker als die tatsächliche Spielstärke.

Interessanter wären Diagramme mit Kurven, hinter einer Elo.
Wenn das möglich wäre, wäre das die perfekte Darstellung von Spielstärke. Wir sehen ... aha, Vancouver hat 2000 Elo und sehen dahinter wie genau es zu dieser Stärke kommt.

So genug ...
Sitze zu lange am PC heute!

Viele Grüße
Frank

Ende vom Lied war dann ... die Fernschachspieler kam um die Ecke.
Frank ich brauche einen Cocktail für offene Stellungen, viele Figuren auf dem Brett.
Die meisten lassen 3-4 Engines gleichzeitig laufen und wollten also wissen welche, nach Stellungstyp.



Schrieb mir wie heute die Finger wund.

Und mein Aufhänger heute ist:
Mit Schachcomputer lässt sich alles einfach feststellen.
Die Stärken und Schwächen gehen weiter auseinander. Die Spielstärke ist besten Geräte liegt auf dem eigenen Level. Es lässt sich einfach mehr feststellen, was dann vielleicht auf anderes drübergestülpt werden könnte!? Lässt sich vieles einfacher entwickeln. Durch die enormen Spielstärken heute gehen gute Entwicklungsideen baden, weil deutlich weniger erkannt wird. Je stärker die Engines desto blinder schaut der Mensch zu.

Hallo Nick,

ist Deine Auswertung ähnlich zu der Methode von Yazgac aus den 80ern? Dieser hat doch immer GM Partien mit Schachcomputern nachgespielt und dann Punkte für einzelne Züge vergeben. Kann mich an eine CSS erinnern, wo der Super-Conny besser als alle anderen 32/16 Bitter abgeschnitten hatte.

https://www.schach-computer.info/wik...ch_K%C3%B6nnen

Gruß, Peter

__________________
http://chess-computer.blogspot.com/

Sorry bin auf der Arbei daher kann ich nicht schnell beantworten. Heute Abend wenn ich nach hause komme setze ich mich dran..

Gruss an alle

Nick

Überhaupt kein Käse, so ist Schach. Da kann nunmal ein einziger Zug eine an sich gute Partie wegwerfen. Spielstärker ist der, dem das auf lange Sicht seltener passiert.

Ich werde gleich versuchen meine Antwort zu Schreiben auf Englisch weil wenn ich das auf Deutsch versuche dann werde ich es bestimmt nicht gut erklaeren koennen.

Gruss
Nick

Absolut richtig! Jeder Mensch und jeder Computer macht Fehler. Entscheidend ist aber die Anzahl!

__________________
Schach ist das Spiel, das die Verrückten gesund hält. (Albert Einstein)

Hallo Rasmus,

diese Spielchen haben was mit "Objektiv" und "Subjektiv" zu tun.

Warum findet Person A Tal gut?
Warum findet Person B Novag geil?
Person C ist TheKing Fan!
Person D mag den SuperConny!

Frage den Super Conny Fan:
Warum findest Du den Super Conny gut?!

Kommt nicht als Antwort: 1800 Elo, eine dumme Zahl aus 1:0, 0:1 und Remis.
Da kommt eine ganz andere Antwort. Da muss eine ganze andere Antwort kommen sonst schaut das irgendwie komisch aus.

Frage den SG Wattenscheid 09 Fan.
Da kommt nicht Liga IV 25 Punkte nach 21 Spielen ... shit, will gar nicht daran denken, verschweige das ja lieber.
Würde das also subjektiver verpacken ... WENN DIE CHANCE BESTEHT (menschlich).

Egal was dann als Antwort kommt: Subjektive Komponente!
Die Antwort könnte sich aber auch objektiver gestalten.

Fakten, Fakten, Fakten ... Focus!
Fakten bedeutet nicht nur in Zahlen zu reden, die ein Dritter nicht versteht oder die in sich nicht logisch sind weil unterschiedlich ausgelegt.

Oft denke ich mir ...
Wenn nicht "Mission Control Center" hoffentlich nicht die Damen Boutique um die Ecke. Wenn die Damen Boutique um die Ecke als Antwort gibt:

1800 Elo ... Mittelspiel = 76%, Übergang Endspiel = 28%, Endspiel = 16% (jetzt mal der Super Constellation in Realität)
Was soll diese dumme Frage, bist Du nicht im Bilde?

Und mir dann erzählt das Super Conny das Mittelspiel sehr aufregend spielt und dort extrem stark ist.

Würde jetzt noch ein bissel was dahinter stehen:
Opferbereitschaft mit einer Prozentzahl
und in einer Liste plötzlich Schachcomputer auftauchen die das gar besser spielen ...

Oh mein Gott ...
Wäre das ganze Volk wieder unterwegs.

Wenn die Dame mir das erzählen würde dann glaube mir ...
Ich würde da stehen wie bekloppt und nicht abgeholt.
Ich würde sofort Herrenmode verkaufen.

Also:
Computerschach = Mission Control Center beim Schach allgemein.
Irgendwie muss das lösbar sein und zwar schlüssig und am besten von der Community selbst.
Bevor überspitzt die Damen Boutique um die Ecke ...

Das was Nick hier gepostet hat, absolut schlüssig.
Muss ja nur auf das gebildete Gesamtergebnis schauen!

Gruß
Frank

There is a book that I played a thousand times when I was young and still enjoy it today it was written in the late 1960's it is called Solitaire Chess.

https://www.amazon.com/Solitaire-Che.../dp/4871878236

This one was written by Al Horowitz. But there are also others with the same concepts that are fun for humans to play when they are at home alone. It has a points rating system but it is very basic.

Peter mentioned Yagaz, his ideas are similar and probably have the same roots but also have the same problems today as the book example I just gave.

The problem with these books today is that if you run the games through a modern computer, you see that the people who wrote them had something important missing.

1) They did not allow for the hundreds of alternative reasonably good moves. As they gave scoring points for one or two moves and that is it.

2) In many cases the points they gave were wrong when you study them today as they missed better moves which they evaluated as zero points.

So although these books are great, they had gaps in knowledge and research. It did not matter if Grandmaster helped them to create it. Modern analysis will show the mistakes made.

But I loved these books and the concept behind them. So in 2014 when I created these tests I decided to use modern technology to help me evaluate everything. Every possible move, move by move for any game.

But to start I had to sit back and think through the exact same questions which are being asked today, because I knew that regardless of what is done people will find a way to critize its accuracy, choice of game etc etc.

So before I even started working on the first test, I made a small list of problems:

1) What games do I pick
2) How do I know that the evaluations are correct
3) How do I evaluate them
4) etc...

And when I started making these notes, just like comments I am reading here in the posts. A light bulb went on and it said this is crazy!

1) Why am I worried about which game to pick?
2) Why am I worried about the absolute exactness of the evaluation?

Well for 1) and 2) I realized that my questions and worrying about them is absolutely crazy and totally irellevant if I plan to evaluate every single move possible.

So the question is why is it crazy?

Well lets start with the obvious. In 2014 Stockfish 8 or 9 was the strongest program, cant remember which. It sits at 3400 ELO and is about 600 ELO better than the best human player in existence. So why would I be worried about it evaluating a move here and there perhaps differently to Houdini or Komodo. Who cares and what difference does it make.

Let me explain the "Who cares" comment so I will put it again in bold and perhaps the Light will shine with more people. Here is the Light:

I AM EVALUATING EVERY POSSIBLE MOVE WITHIN EVERY POSSIBLE MOVE IN A GAME AND SCORING IT!

That is the Light that has to first shine before the game rating system can be understood. So what does that mean?

Lets say the best move within a given tolerance gets 30 points. And the next move which is almost as good gets 29.4 points and so on down to zero. What difference does it make to the final score whether I used Stockfish, Komodo, Houdini or LC0? The answer is the almost "ZERO DIFFERENCE"

Why? Well because if Komodo evaluated the one move here of there better than Stockfish and vice versa over the the 238 moves that are being evaluated they probably cancel each other out. So lets say over these 5 games there is a difference of 30 points between the best programs out of a Total of 6780 points what is that as a percentage 30/6780? Well it is 0.44% What if the difference is 100 points? Well it is 1.4% Well in a rating system 1.4% amounts to a +/- 14 ELO on a 2000 ELO computer.

So once you sit back and realize this you start to laugh at how crazy you are to even ask a question such as what if the evaluation is wrong from a Stockfish that has a rating of 3400 ELO and where you are planning to evaluate each move by about an hour at a time to get to at least 35 ply or higher (this was a few years ago when i did this test). Try running Critter 1.6 rated CCRL at 3159 ELO. Try running it at 3 minutes per move or 30 seconds per move and you will see that it will not come close to 35-50 ply. So yes once you know this, you also know that you can pretty much evaluate everything and rate it.

The second light that needs to shine is this: IT MAKES ABSOLUTELY NO DIFFERENCE WHICH CHESS GAME YOU USE!

That is the second misconception. I can rate any game I want. Because the game itself is cosmetic. It is the evaluation from Stockfish for example that matters.!!!

Once you get past these two fundamental misconceptions which are bred into us because of the Tests we grew up with like BT2450 or BT2630 then you will realize that there are other ways to do this nowadays!

Anyway this what made me realize that I could do something different, be very accurate in comparisons, not only in the test scores between computers, but also at the same time have a very accurate tool for Clone tests and related programs as you can see deviations or no deviations in each of the 5 games.

I converted the test scores to ELO only for cosmetic reasons because I just like everyone else here likes to get a feel were the rating stands in ELO. But it is cosmetic only.

Frank you asked how many games are needed?

Well when I realized that it made no difference which game to evaluate. I then became nostalgic. I wanted to also rate famous names in history right from the beginning to today. That is why I started with the 1st test which came from a Poem written in the year 1475 that follows a game between Francesco di Castellvi playing Narcisco Vinyoles. That is the earliest ever chess game recorded in history. The second game is the Turk (Automaton) from 1770. The 3rd game is a beautiful series of sacrifices made by Thomas Bowdler in the year 1788. Games 4 and 5 are Philidor. All these games are pre 1800.

I am planning to someday do famous games and players between 1800-1900 and then 1900 to 2000 and then 2000 to now.

As I said it makes no difference to me what the game is. I just want them to be famous players and beautiful games that lets me also compare the rating of the human who played them. Not just the computer that is being tested.

Just imagine for example if I picked 20 Steinitz games or picked his World Championship games. I would be able to rate Steinitz as a player and compare him to Lasker for example. The possibilities are endless.

Anyway to answer your question my 5 games is too small a representation for obtaining accurate data across many different styles of play. But if for example I had another set of games for each of the following timelines then I think it would be extremely accurate and fun to test and play:

5 for before 1800
5 for 1800-1850
5 for 1850-1900
5 for 1900-1950
5 for 1950-2000
5 for 2000-2019

Now these should be classic Games..ie Anderssen's Immortal Game !!

BTW each game because of the depth of analysis needed takes about a week to prepare so you can see why so far I had only done 5 games.

Best regards

Hi Nick,

now I need time for an answer.
Very late and tomorrow I have a long day but you can be sure, my answer will come as soon as possible.

Thank you for your time!

Later ...

Best
Frank

For the moment, I read it ... interesting points you give us for thinking.

Hi Kurt,

From your question, I imagine that your thoughts are still around manual human tests made in mid 1990's were a computer if you were lucky reaches 10-12 ply and the human controled the destiny of the test through manual manipulation to suit the end result?

In my test I manipulate nothing. Every evaluated score is a formula that gives me the rating score and a second formula converts this to ELO so that it can be easier for someone to compare it to other lists.

As a result every test game follows 100% exactly the same formula's and I determine or manipulate nothing manually in any test game as each test game is done 100% exactly the same. No human opinions.

So unless the formula is wrong which I doubt as I am pretty good with this kind of stuff like spreadsheets and formulas... there is nothing here that could be questioned.

PS... I think if you were to just look at the tests and see scores like 29.4, or 3.6 or 0.6... etc for individual moves that a computer played you will quickly realize that these are formula drive scores.


Best regards