Future of computer chess: 2 top engines?

[Frank Quisinsky]

Wow, a good answere and a nice explanation!

Shortly:
A strong positional play is very heavy to find out.
I have many problems with it to say ... that's a good positional move.
Last year I am thinking I have to read Karpov books for learn more about positional moves. But after reading I am thinking Karpow have in best years the optimal understandings for openings and best fields for knights and bishops. Not easy for me to learn after reading Karpow material.

I read some chess book to the topic (at the moment I read the books from Gelfand, Quality Chess).

Two German chess books are fantastic:
- Eliskasis, Stellungsspiele
- Bronzik, Techniken des Positionsspiel im Schach

But after all ...
With long time analyzes we can see good tactical moves and many chess books are available to that topic. 50 times more as to the topic positional play. To positional play a handfull good chess books in German language are available.

What I like to say:
The same problems our best chess player have to explain the positional play, the same problems we have to see that in computer chess. It's an order Madleine to finding strong chess engines for positional play. Often I am thinking for strong engines it's make not a different ... positional play or tactical play.

Engines:
Programmers should be careful with NN. If end of the day the programs playing the same style of chess many users lost interest. The own style of chess is the secret for engines. Komodo have a very nice and strong endgame (Komodo 14 is after stats stronger as Stockfish 11 in pure endgames). For opening understandings is NN very danger. Programs lost here own face. Komodo in earlier versions are more passive as Stockfish in earlier versions for openings. But Komodo in the earlier versions able to play chess, Karpow like to play. Not the attacker but able to find out the right fields for the important pieces. After all I understand and many analyzes after reading material about Karpow.

Thank you for your nice text after my provocative message!
Good luck to find out the engines for positional play (not easy).

Have a nice day!

Best
Frank

[yurikvelo]

For Vincent's recent tough problems list, LC0 solved one in a few seconds that bleeding edge Stockfish failed to solve in an hour using 60 threads and 65 GB RAM with full 6 man TB files.

But it was SF (or other strong A/B engine) who:
1) selected this position among millions
2) proved solution


Can you verift list of problems and claim some move as a solution (or "not a solution") with Lc0?

[towforce]

I have many problems with it to say ... that's a good positional move.
Last year I am thinking I have to read Karpov books for learn more about positional moves. But after reading I am thinking Karpow have in best years the optimal understandings for openings and best fields for knights and bishops. Not easy for me to learn after reading Karpow material.

It's probably wrong to think that people can explain, or even understand, why they think the way they do: the human brain is a deep neural network, and it's difficult enough to even understand a shallow neural network. Karpov (and the rest of us) will have deep NNs that tell him what's "correct", and other NNs that give him a surface explanation for these "feelings".

Sometimes, people can give clear and accurate explanations, but a lot of the time, trying to understand why people think the way they do is a bad idea that will lead you in the wrong direction.

[Milos]

Engines:
Programmers should be careful with NN. If end of the day the programs playing the same style of chess many users lost interest. The own style of chess is the secret for engines. Komodo have a very nice and strong endgame (Komodo 14 is after stats stronger as Stockfish 11 in pure endgames). For opening understandings is NN very danger. Programs lost here own face. Komodo in earlier versions are more passive as Stockfish in earlier versions for openings. But Komodo in the earlier versions able to play chess, Karpow like to play. Not the attacker but able to find out the right fields for the important pieces. After all I understand and many analyzes after reading material about Karpow.

Interesting. So your "style of chess" in reality means which type of blunders the engine makes. So you actually prefer that engines make different type of blunders.

Btw. Komodo 14 is nowhere near Stockfish 11 in "pure endgames" whatever that one actually meant. The only "advantage" of Komodo to SF in endgames is that it reports inflated nps numbers which in reality doesn't translate to Elo. One can easily prove that on typical 7-8-men endgames when EGTBs are turned off.

[Frank Quisinsky]

Hi Milos,

to look in the endgame in detail is easy but need time!!
Fish out the won games is the secret here!

Different steps:

Step 1
A strong databases with games without resign mode and no 1+1 games or such things!
As next you are searching the games over 100 moves.
Games ended with 100-300 moves ... for this we added for the moment an easy table in the Excel for FCP Tourney-2020 statistics.

Better is to do the same in Steps 2-3 with won games 79-100 moves. Need clearly longer!!

Code: Select all

   # Player                          :      Elo  Games  Score%  won  draw  lost  Points  Draw%  Error   OppAvg   OppE   OppD
   1 Komodo 14.0 BMI2 x64            :  3185.13    315    67.6  119   188     8   213.0   59.7  23.59  3049.99  16.55   35.5
   2 Stockfish 11 BMI2 x64           :  3167.37    363    65.6  115   246     2   238.0   67.8  21.70  3046.33  16.70   35.8
   3 Houdini 6.03 Pro x64            :  3163.86    376    65.8  130   235    11   247.5   62.5  21.62  3044.71  16.67   37.1
   4 Ethereal 12.25 PEXT x64         :  3130.66    659    61.7  172   469    18   406.5   71.2  15.69  3043.21  16.94   38.2
   5 Fire 7.1 POP x64                :  3108.67    498    59.4  126   340    32   296.0   68.3  17.03  3039.78  16.58   37.2
   6 rofChade 2.3 BMI x64            :  3102.57    593    59.3  148   407    38   351.5   68.6  15.17  3034.65  16.66   38.9
   7 Shredder 13 x64                 :  3101.60    491    59.1  139   302    50   290.0   61.5  17.18  3034.40  16.42   37.7
   8 SlowChess BC 2.2 x64            :  3092.21    448    56.3   94   316    38   252.0   70.5  17.59  3044.82  16.59   37.0
   9 Xiphos 0.6 BMI2 x64             :  3088.82    429    56.4   96   292    41   242.0   68.1  18.16  3039.94  16.58   37.7
  10 Laser 1.7 BMI2 x64              :  3085.08    507    56.2  118   334    55   285.0   65.9  16.25  3038.90  16.56   38.8
  11 RubiChess 1.7.3 x64             :  3082.45    531    57.1  118   370    43   303.0   69.7  16.19  3031.89  16.43   38.3
  12 Booot 6.4 POP x64               :  3068.18    303    52.0   56   203    44   157.5   67.0  21.55  3050.15  16.53   34.3
  13 Defenchess 2.2 POP x64          :  3056.06    646    53.2  107   473    66   343.5   73.2  14.64  3033.16  16.49   38.6
  14 Fritz 17 (Ginkgo) x64           :  3053.78    496    52.0   84   348    64   258.0   70.2  16.09  3038.62  16.54   38.6
  15 Fizbo 2.0 BMI2 x64              :  3053.53    540    53.1   90   394    56   287.0   73.0  15.98  3031.15  16.54   38.7
  16 Arasan 22.0 BMI2 x64            :  3052.45    508    53.2   99   343    66   270.5   67.5  16.60  3029.85  16.39   38.0
  17 Andscacs 0.95 BMI2 x64          :  3043.65    465    51.2   82   312    71   238.0   67.1  17.53  3036.52  16.32   37.5
  18 Winter 0.8 x64                  :  3042.57    492    52.1   91   331    70   256.5   67.3  15.75  3028.28  16.48   38.1
  19 GullChess 3.0 BMI2 x64          :  3030.85    562    50.4   79   408    75   283.0   72.6  15.44  3029.78  16.42   38.5
  20 Schooner 2.2 SSE x64            :  3022.07    496    47.3   57   355    84   234.5   71.6  15.99  3039.63  16.56   37.8
  21 Nirvanachess 2.4 POP x64        :  3021.72    481    49.7   76   326    79   239.0   67.8  16.34  3024.87  16.19   37.2
  22 Fritz 16 (Rybka) x64            :  3020.95    561    48.8   77   394    90   274.0   70.2  15.13  3030.03  16.47   38.5
  23 Vajolet2 2.8 BMI2 x64           :  3020.50    523    48.8   74   362    87   255.0   69.2  15.91  3029.68  16.32   37.9
  24 Minic 2.33 x64                  :  3000.87    444    47.0   53   311    80   208.5   70.0  17.56  3024.03  16.28   37.2
  25 SmarThink 1.98 AVX2 x64         :  3000.07    510    47.5   69   346    95   242.0   67.8  16.61  3020.66  16.27   37.0
  26 Nemorino 5.00 BMI2 x64          :  2999.12    559    46.4   77   365   117   259.5   65.3  15.70  3025.52  16.40   37.8
  27 Equinox 3.30 x64                :  2994.79    589    45.4   64   407   118   267.5   69.1  14.60  3028.44  16.46   38.7
  28 Critter 1.6a x64                :  2990.26    550    45.0   63   369   118   247.5   67.1  15.50  3027.56  16.34   38.0
  29 Pedone 2.0 BMI2 x64             :  2986.42    512    44.0   58   335   119   225.5   65.4  16.34  3029.80  16.35   37.7
  30 Chiron 4 x64                    :  2983.63    565    43.4   58   374   133   245.0   66.2  15.90  3031.12  16.49   38.8
  31 iCE 4.0 v853 Modern x64         :  2982.88    470    44.3   50   316   104   208.0   67.2  16.37  3026.02  16.37   37.8
  32 Igel 2.5.0 BMI2 x64             :  2982.07    458    43.8   49   303   106   200.5   66.2  17.01  3027.67  16.36   37.5
  33 Protector 1.9.0 x64             :  2981.90    520    43.7   52   350   118   227.0   67.3  15.82  3029.58  16.43   38.8
  34 Wasp 4.00 Modern x64            :  2981.48    546    43.5   52   371   123   237.5   67.9  15.43  3030.07  16.39   38.0
  35 Combusken 1.2.0 x64             :  2981.02    500    45.3   47   359    94   226.5   71.8  16.00  3017.43  16.25   36.4
  36 Senpai 2.0 BMI2 x64             :  2978.28    537    44.2   50   375   112   237.5   69.8  16.14  3021.02  16.25   37.2
  37 Hannibal 1.7 x64                :  2977.44    623    43.6   67   409   147   271.5   65.7  14.09  3025.45  16.38   37.6
  38 Demolito 2020-05-14 PEXT x64    :  2972.72    671    42.5   60   451   160   285.5   67.2  14.06  3028.44  16.49   38.8
  39 Texel 1.07 BMI2 x64             :  2969.05    569    42.5   61   362   146   242.0   63.6  15.22  3022.64  16.21   37.0
  40 Monolith 2 PEXT x64             :  2952.28    415    41.0   26   288   101   170.0   69.4  18.32  3019.93  16.16   36.3
  41 Rodent IV 0.22 POP x64          :  2925.97    487    37.3   32   299   156   181.5   61.4  18.21  3021.93  16.17   37.2

White advantage = 18.86 +/- 1.95
Draw rate (equal opponents) = 73.43 % +/- 0.49

Step 2:
Interesting are the won games!
In Step 4 later the draw and lost games.
Is the postion, start of 12 pieces on board (looking often in 14 too), equal?!
So you have to look in the won games (why the game won).

I have an Excel table with 78 types of different won-ways!

Step 3:
Now you can build a muster for won games with the different types of pieces on board.

I think we will offer the way for such stats in excel later ...
For the moment I have to many work with the next FCP Tounrey seria.

Step 4:
Now the draw and lost games are to set in a "relationschip".

After all ...
You can see the advantage between Komodo and Stockfish ...

I do that 3 times now:
Stockfish 8 to Komodo 10, differents +31 Elo ... Komodo 10 is 31 points stronger
Stockfish 11 to Komodo 14, differents +19 Elo ... komodo 14 is 19 points stronger (after 1.200 of the 2.000 FCP Tourney-2020 games).

At the moment 1.840 of 2.000 games played.
But after such a work it's around the same different as in the table I added before.
For looking in 10 minutes the table is good enough.

In detail ...
6 types of important Krp endgames, often on the board, Komodo seems to have a big advantage to Stockfish. Stockfish draw in positions Komodo can win. In 2x bishop and pawns Komodo have an advantage. Stockfish have a very small advantage to Komodo with knight / pawns endgames.

But I am not the beta tester of Stockfish, I thing the Stockfish group have enough people for make the program stronger. I like to look in detail a lot, that is much more interesting as to look of Elo alone.

So, most important for looking in style of engines are a good database.
A good database is a playground and I hope of stronger software here because all what we have on professional software in combination with strong statistics is more or less for beginners.

Best
Frank


For the different Stockfish 8 to Komodo 10 I am looking also in games 80-99 moves and pickup the won games. Not for differents Stockfish 11 and Komodo 14. Need a lot of time and other questions to playing styles from other engines are more interesting for me.

[Frank Quisinsky]

Hi Twoforce,

that's right what you told here to 80%.

I am quit sure that an explanation for positional good moves is possible, speaking from 2 chess German chess books before. With books you can indeed learn and understand ideas. In many chess books the commented games, the players explain his own ideas / positional ideas, are interesting. But here I am thinking in most of cases the same you have written before.

Karpow plays often strongest positional moves from his "Subconscious" I think (without start thinking in detail).
His brain saw a complicated position and know directly (without thinking) bishop x or knight y need an other field.
If Karpow have all pieces on the right positions he start thinking for a plan. Often I have the feeling that this one is the secret of Karpow's opening play in combination with a lot of knowledge to opening systems he must have!!

Karpow is a very interesting chess player for analyzes.

But after all ...
I gave up to analyze him because never I will understand his "Subconscious".
To many surprises if you are thinking you are to 100% sure what Karpow would like to play now and you can see .... the move is an other one.

Chess must be a strategic war in his brain ... often I am thinking!
Karpow plays moves the opponent never have in the brain and won many games psychological (the final results of my analzyes).

But exactly this one made Karpow interesting ... much more interesting as Kasparow!!

Best
Frank

[Peter Berger]

I read some chess book to the topic (at the moment I read the books from Gelfand, Quality Chess).

Two German chess books are fantastic:
- Eliskasis, Stellungsspiele
- Bronzik, Techniken des Positionsspiel im Schach

How many ELO did you gain by reading these books?

Let me make an educated guess: none.

So this means, you didn't understand them, but they gave you some fuzzy feeling of doing something somewhat useful and worthwhile. You +felt+ like you were getting the point.

By the way - there are other chessbooks that earned people a significant amount of ELO points - to give an infamous example: Simon Webb - Chess for Tigers - some of the Silman books also come to mind.

We can't really understand current computer games, at least I can't. I have a hard time to imagine you can.

[Frank Quisinsky]

Hi Peter,

Oh yes, Silmans books are great and I like the books a lot.
Many good German chess books are available!

I try to hold the books I like most, give away the others.
In my collection are around 550.

Can be see in the about.me page on my site www.amateurschach.de.

I think most important for chess books is:
Willingness to learn!

So, players can learn with Euwe, others with Watson, again others with Jussupow and so one.
I have not the opinion that a chess book is the guarantor for x Elo more.
Not Silman and not the others.

Different authors like to say this!
A good example is Willy Hendricks (Erst ziehen dann denken).
Very controversial book but I like it ... in German ... der Knabe ist sehr überzeugt von dem was er erzählt ... aber viele Meinungen bereichern!

That's the point (understandings).
With statistic I try to understand, I haven't other possibilities ... the chess books I have and my own understanding.
Chess books can give us a lot of energy for our hobby because it's all the time interesting what for opinions others have to all the different topics around chess.

Best
Frank

The two German books I spoken before have the main topic (positional play) and both are really great!!

[Frank Quisinsky]

The book from Simon Webb (very old chess book) is more or less "Psychologic".
Not to compare with things Silman do ... complete other main-topic.
Simon try to make the readers stronger with more "self-confidence" and give nice examples.
I think it can be help, yes!!

Simon Webb wrote here one of the most interesting chess books in my opinion.
Silman goes a very own and strong way. Most I know like Silman a lot, I too.

But I have other favorits!!
Aagard, Dvoretsky / Jussupow, Richter and different others.
Herman Grooten I like a lot and what I read at the moment from Gefand is very interesting.
Erstürmung der Barrikaden (Larry Christiansen) or Brett in Flammen (Alexey Schirow), Richtig Opfern (Spielmann) and so one ... great material!!

Best
Frank

[jp]

For Vincent's recent tough problems list, LC0 solved one in a few seconds that bleeding edge Stockfish failed to solve in an hour using 60 threads and 65 GB RAM with full 6 man TB files.

But it was SF (or other strong A/B engine) who:
1) selected this position among millions
2) proved solution

Can you verift list of problems and claim some move as a solution (or "not a solution") with Lc0?

Which position was Dann talking about?