ProDeo 3.2 elo 3411 :lol!:

Just having fun.

I am redoing the Evaluation Rating List. Pity none was interested.

Code:

Evaluation Rating List
Running      : Evaluation Tournament
Time Control : Time control : depth=1
Games        : 2800

Results from file root.pgn:

No. Name           Win Draw Loss Unf.  Score Games       %
----------------------------------------------------------
  1 SlowChess 2.5 +404 =134 -162   *0  471.0   700   67.3%
  2 Komodo Dragon +395 =127 -178   *0  458.5   700   65.5%
  3 Nemorino 6.00 +333 =112 -255   *0  389.0   700   55.6%
  4 ProDeo 3.2    +320 =131 -249   *0  385.5   700   55.1%
  5 Stockfish 12  +261 =111 -328   *0  316.5   700   45.2%
  6 rofChade 2.3  +210 =153 -337   *0  286.5   700   40.9%
  7 Igel 3.0.5    +210  =89 -401   *0  254.5   700   36.4%
  8 Pedone 3.1    +191  =95 -414   *0  238.5   700   34.1%

Total Games:    2800
White Wins:     1182 (42.2%)
Black Wins:     1142 (40.8%)
Draws:           476 (17.0%)
Unfinished:        0 (0.0%)

Estimated ratings for this elo 3379 pool

   # PLAYER           :  RATING  POINTS  PLAYED   (%)
   1 SlowChess 2.5    :  3493.6   471.0     700    67
   2 Komodo Dragon    :  3481.1   458.5     700    66
   3 Nemorino 6.00    :  3414.8   389.0     700    56
   4 ProDeo 3.2       :  3411.6   385.5     700    55
   5 Stockfish 12     :  3347.9   316.5     700    45
   6 rofChade 2.3     :  3319.9   286.5     700    41
   7 Igel 3.0.5       :  3289.3   254.5     700    36
   8 Pedone 3.1       :  3273.6   238.5     700    34

Will play some more games.

28.000 games now.

Code:

Gambit Rating List
Running      : Evaluation Tournament
Time Control : Time control : depth=1
Games        : 28000

Results from file root.pgn:
 
No. Name           Win Draw Loss Unf.  Score Games       %
----------------------------------------------------------
  1 Komodo Dragon +4059 =1123 -1818   *0 4620.5  7000   66.0%
  2 SlowChess 2.5 +3948 =1315 -1737   *0 4605.5  7000   65.8%
  3 Nemorino 6.00 +3603 =980 -2417   *0 4093.0  7000   58.5%
  4 ProDeo 3.2    +3032 =1342 -2626   *0 3703.0  7000   52.9%
  5 Stockfish 12  +2585 =1051 -3364   *0 3110.5  7000   44.4%
  6 rofChade 2.3  +2252 =1336 -3412   *0 2920.0  7000   41.7%
  7 Igel 3.0.5    +2164 =909 -3927   *0 2618.5  7000   37.4%
  8 Pedone 3.1    +1879 =900 -4221   *0 2329.0  7000   33.3%

Total Games:   28000
White Wins:    11778 (42.1%)
Black Wins:    11744 (41.9%)
Draws:          4478 (16.0%)
Unfinished:        0 (0.0%)

Estimated ratings for this elo 3379 pool

   # PLAYER           :  RATING  POINTS  PLAYED   (%)
   1 Komodo Dragon    :  3484.6  4620.5    7000    66
   2 SlowChess 2.5    :  3483.1  4605.5    7000    66
   3 Nemorino 6.00    :  3433.8  4093.0    7000    58
   4 ProDeo 3.2       :  3397.6  3703.0    7000    53
   5 Stockfish 12     :  3343.0  3110.5    7000    44
   6 rofChade 2.3     :  3325.2  2920.0    7000    42
   7 Igel 3.0.5       :  3296.6  2618.5    7000    37
   8 Pedone 3.1       :  3268.2  2329.0    7000    33

Next run, added 4 oldies: Fruit 2.1 | Gaviota 1.0 | ice 3.0 | Ruffian 2.1 - 66,000 games.

ice together with Gaviota are unique engines, we will see...

Code:

Gambit Rating List
Running      : Evaluation Tournament
Time Control : Time control : depth=1
Games        : 28000

Results from file root.pgn:
 
No. Name           Win Draw Loss Unf.  Score Games       %
----------------------------------------------------------
  1 SlowChess 2.5 +3493 =885 -1138   *0 3935.5  5516   71.3%
  2 Komodo Dragon +3493 =826 -1198   *0 3906.0  5517   70.8%
  3 Nemorino 6.00 +3191 =745 -1581   *0 3563.5  5517   64.6%
  4 ProDeo 3.2    +2818 =980 -1719   *0 3308.0  5517   60.0%
  5 Stockfish 12  +2475 =821 -2220   *0 2885.5  5516   52.3%
  6 rofChade 2.3  +2219 =1097 -2201   *0 2767.5  5517   50.2%
  7 Gaviota 1.0   +2233 =980 -2303   *0 2723.0  5516   49.4%
  8 Igel 3.0.5    +2146 =708 -2662   *0 2500.0  5516   45.3%
  9 Fruit 2.1     +1964 =913 -2639   *0 2420.5  5516   43.9%
 10 Pedone 3.1    +1941 =705 -2870   *0 2293.5  5516   41.6%
 11 ice 3.0       +1310 =825 -3381   *0 1722.5  5516   31.2%
 12 Ruffian 2     +696 =753 -4067   *0 1072.5  5516   19.4%

Total Games:   33098
White Wins:    13944 (42.1%)
Black Wins:    14035 (42.4%)
Draws:          5119 (15.5%)
Unfinished:        0 (0.0%)

Estimated ratings for this elo 3379 pool

   # PLAYER           :  RATING  POINTS  PLAYED   (%)
   1 SlowChess 2.5    :  3537.6  3935.5    5516    71
   2 Komodo Dragon    :  3533.1  3906.0    5517    71
   3 Nemorino 6.00    :  3485.0  3563.5    5517    65
   4 ProDeo 3.2       :  3451.0  3308.0    5517    60
   5 Stockfish 12     :  3396.7  2885.5    5516    52
   6 rofChade 2.3     :  3381.6  2767.5    5517    50
   7 Gaviota 1.0      :  3376.0  2723.0    5516    49
   8 Igel 3.0.5       :  3347.5  2500.0    5516    45
   9 Fruit 2.1        :  3337.3  2420.5    5516    44
  10 Pedone 3.1       :  3320.7  2293.5    5516    42
  11 ice 3.0          :  3242.9  1722.5    5516    31
  12 Ruffian 2        :  3138.6  1072.5    5516    19

Tournament could not finish (crash), only 5500 games instead of 7000.

Any interesting playing style engines to test?

Admin wrote:

Any interesting playing style engines to test?

I always like your enignes, since the first. I am a fanboy.

But the only engine style I see is Elo.

Ed,

What is Pro Deo 3.2? Just an experiment or an engine you could (or want to) publish?

For reasons unknown to me I am trying to squeeze some elo out of my old dinosaur. Currently I have 15-20 elo which is way to less for a release. I am currently trying to improve king safety. While doing so I came up with the idea to compare my eval with NNUE engines. Depth=1 is controversial of course.

Admin wrote:

For reasons unknown to me I am trying to squeeze some elo out of my old dinosaur. Currently I have 15-20 elo which is way to less for a release. I am currently trying to improve king safety. While doing so I came up with the idea to compare my eval with NNUE engines. Depth=1 is controversial of course.

Ply 1 shouldn't be controversial: it's a very good thing! It answers the question "How much does it know about chess"!

TheSelfImprover wrote:

Admin wrote:

For reasons unknown to me I am trying to squeeze some elo out of my old dinosaur. Currently I have 15-20 elo which is way to less for a release. I am currently trying to improve king safety. While doing so I came up with the idea to compare my eval with NNUE engines. Depth=1 is controversial of course.

Ply 1 shouldn't be controversial: it's a very good thing! It answers the question "How much does it know about chess"!

In principle yes, except that:

1. It depends on the quality of QS (Quiescence Search) and especially if checks in QS are allowed or not.

2. It's known from Stockfish that it doesn't do a full root (depth=1) search.

I would be very interested to see a Pro Deo with a modern search and an eval adapted to some more depth. I am already using daily engines 300-500 weaker than SF without losing much, most of the time almost nothing. Since I need time to analyse a position myself, usually the engine comes with a good line.

When I use Pro Deo in checking opening lines, I noticed that the usually find the "right" move at depths 10-14, to change to another solution which is probably not optimal thereafter. E.g., if the theoretical move is a4, Pro Deo chooses it from depth 10, to switch to Bd3 at depth 15, while other more modern engines find a4 at depth 18. It is often the case, but I don't have enough data this to be more than an impression.

Ed, king safety is a good idea.
Another one is making the search multi cpu capable.

Then something that is not weakening opening or middlegame is Endgame knowledge.
For a long time your strongest version was the one with big opening book.
Then the 3.1 version overtook it.

All my PCs have nothing to do.
Because i am not interested in stockfish clones.
If you like i can give you at least 7 machines to run engines.

Mclane wrote:

Ed, king safety is a good idea.

In the ProDeo.eng file there is this paramter:

Code:

[King Safety Modus ? aggressive]

Change it to:

Code:

[King Safety Modus = aggressive]

It will go right to the throat of the opponent but in self-play it loses 25-30 elo.

matejst wrote:

I would be very interested to see a Pro Deo with a modern search and an eval adapted to some more depth. I am already using daily engines 300-500 weaker than SF without losing much, most of the time almost nothing. Since I need time to analyse a position myself, usually the engine comes with a good line.

When I use Pro Deo in checking opening lines, I noticed that the usually find the "right" move at depths 10-14, to change to another solution which is probably not optimal thereafter. E.g., if the theoretical move is a4, Pro Deo chooses it from depth 10, to switch to Bd3 at depth 15, while other more modern engines find a4 at depth 18. It is often the case, but I don't have enough data this to be more than an impression.

To make fast progress (like so many others) I need to start from scratch. Currently I am stuck with an old 32 bit compiler that since 2000 hasn't been updated. The engine is in assembler which was needed in the 80-90's but is an outdated concept. But whatever I do (old or new engine) to move to a modern search I will need to borrow a lot from open sources. I don't like the idea. It's best I focus on playing style and Benjamin 1.0 was a good start.

Ed, I understand these dilemma, and I would have had the same. I also think that Texel tunning kills engine personality, what made engines different -- although it clearly improves the evaluation in general.

And then, there is the question of "borrowing". There is a very fine line, indeed, but from what I see, not only that nobody cares any more, but I am sure that one would have found these same ideas soon or later. It is precisely what makes me frown my eyebrows about open source: open source is not a protection, an interdiction, but it should be an incentive. Incorporating creatively some existing ideas in our own code should be normal.

So if you can find a balance between including a few new concepts in Pro Deo, and keeping his overall personality, style of play, it would be fine with me. A bit like John Stanback with Wasp. The way he tuned his evaluation function with a procedure used with NNs gave great results, and Wasp, today, makes me think more and more of the last versions of Zarkov.

So, imho, Pro Deo can be improved and still remain Pro Deo. But it is on you to find the balance. I am absolutely persuaded that you can.

BTW, Ed, could you try Berserk and Seer 2.3.0 in ERL? I am very interested in Berserk's classic eval. I think it's good, but the engines is also very, very fast, and I see it still lacks a lot of features, so I am wondering. If it's not too much to ask, of course. Seer, otoh, has developed an original NN, using tablebases initially.

matejst wrote:

Ed, I understand these dilemma, and I would have had the same. I also think that Texel tunning kills engine personality, what made engines different -- although it clearly improves the evaluation in general.

And then, there is the question of "borrowing". There is a very fine line, indeed, but from what I see, not only that nobody cares any more, but I am sure that one would have found these same ideas soon or later. It is precisely what makes me frown my eyebrows about open source: open source is not a protection, an interdiction, but it should be an incentive. Incorporating creatively some existing ideas in our own code should be normal.

So if you can find a balance between including a few new concepts in Pro Deo, and keeping his overall personality, style of play, it would be fine with me. A bit like John Stanback with Wasp. The way he tuned his evaluation function with a procedure used with NNs gave great results, and Wasp, today, makes me think more and more of the last versions of Zarkov.

So, imho, Pro Deo can be improved and still remain Pro Deo. But it is on you to find the balance. I am absolutely persuaded that you can.

I already found the balance with the release of ProDeo 3.1 which besides an improvement also was a community test. It contained the PESTO tables from the author of Rofchade he published in the programmer forum. And as expected (and yet amazed) the criticism came. No longer an original engine. Open sources are a blessing and a curse at the same time.

Ed, I followed closely this case and I am well aware of it. I am also aware of the accusations against Chris when he refused to publish Coronachess, and also of the outraged reactions against K. Omar when he refused to reveal his code (despite immediately complying and stopping sharing his engine).

Open source is, of course, a blessing, but the abuse, the misuse is a curse -- and there is a lot of abuse lately, especially by newcomers not aware of the long history of computerchess.

I know I would be attacked and accused of trolling if I wrote on CCC something we all probably know: that almost all commercial engines were decompiled at one time or the other, and that their ideas, procedures, and probably (I am not sure) even snippets of code found their way into SF directly or indirectly. (We have proofs for Rybka, Rebel [if I remember well], Komodo, and you surely know better).

Computerchess did not start with Glaurung, Fruit, nor StockFish, but if my son, e.g., today started programming and reading CCC, he would have the impression that SF is everything that ever was and everybody adapted its code and used its ideas. It's rather the opposite that is true (it's a simplification, of course, SF is a marvel or engineering).

Practically, everything that is used in SF suddenly becomes a forbidden fruit for everybody else, and open source is a way to control the code/ideas of others developers. And while the SF developers themselves  for years did not care about it, a witch hunt has lately started, putting everybody in the same hat: CB, Eman, CorrChess, etc., and also disrespecting authors who did a lot of work to make SF more usable for the end-users, without thousands of PC at their disposition. It is so petty.

I strongly encourage, support your efforts to improve Pro Deo.  You gave a lot to the "community" (if community there is), explained a lot about chess programming, and part of your knowledge and personal contribution is everywhere. So, if you still enjoy programming, don't restrain yourself but do as you wish. Dogs bark, caravans pass.

Admin wrote:

I already found the balance with the release of ProDeo 3.1 which besides an improvement also was a community test. It contained the PESTO tables from the author of Rofchade he published in the programmer forum. And as expected (and yet amazed) the criticism came. No longer an original engine. Open sources are a blessing and a curse at the same time.

Let me help. The program people want has the following:

* fully original. No code, or even any ideas, taken from other engines

* tiny executable

* runs on even the slowest/cheapest hardware

* 3700 ELO

* selectable ELO level

* plays almost instantaneously

* fully featured and easy to use UI

* fabulous fun to play against - constantly making romantic swashbuckling sacrifices

...and that's it really. Not much to ask for.

Mclane wrote:

Ed, king safety is a good idea.

Code:

[King Safety Modus = aggressive]

I tried this setting with GRL gambit positions.

Code:

   # PLAYER            :  RATING  POINTS  PLAYED   (%)
   1 Nemo 1.01         :  2902.2   131.5     200    66
   2 Cheese 2.2        :  2853.5   118.5     200    59
   3 Olithink 5.9.9    :  2830.1   112.0     200    56
   4 ProDeo exp        :  2787.9   590.0    1200    49
   5 Nalwald 1.11      :  2768.6    94.5     200    47
   6 Orion 07          :  2731.3    84.0     200    42
   7 Velvet 1.2.0      :  2677.5    69.5     200    35

Elo pool : 2793
ProDeo 3.1 : 2784
ProDeo exp : 2787

So while in self play the option is a regressions of ~20-25 elo against other opponents (and gambit positions) it's likely a winner, with that I don't mean the +3 elo, but the playing style!

Replay of the match now with normal positions.

Code:

  # PLAYER            :  RATING  POINTS  PLAYED   (%)
   1 Olithink 5.9.9    :  2884.7   134.5     200    67
   2 Nemo 1.01         :  2851.8   126.0     200    63
   3 Cheese 2.2        :  2824.2   118.5     200    59
   4 Nalwald 1.11      :  2760.3   100.5     200    50
   5 ProDeo exp        :  2758.6   533.5    1200    44
   6 Orion 07          :  2748.1    97.0     200    49
   7 Velvet 1.2.0      :  2723.4    90.0     200    45

Elo pool : 2793
ProDeo 3.1 : 2784
ProDeo exp : 2758

Confirms my own self play tests, -26 elo while with the gambit positions it's +3 elo.

And so it means I can make Benjamin stronger than ProDeo 3.1 but only on the Gambit Rating List

matejst wrote:

Ed, I followed closely this case and I am well aware of it. I am also aware of the accusations against Chris when he refused to publish Coronachess, and also of the outraged reactions against K. Omar when he refused to reveal his code (despite immediately complying and stopping sharing his engine).

Chris initial intention was to publish Coronachess including source code. Then he had a lot of problems to subscribe to the tournament of Charles Robinson followed by the accusations of being a derivative. You can't do that with an old timer and certainly not when his name is Chris. So now nobody gets anything. If you accuse provide evidence, not assumptions based on nothing. I can sympathize with his decision.

Quote :

Open source is, of course, a blessing, but the abuse, the misuse is a curse -- and there is a lot of abuse lately, especially by newcomers not aware of the long history of computerchess.

I know I would be attacked and accused of trolling if I wrote on CCC something we all probably know: that almost all commercial engines were decompiled at one time or the other, and that their ideas, procedures, and probably (I am not sure) even snippets of code found their way into SF directly or indirectly. (We have proofs for Rybka, Rebel [if I remember well], Komodo, and you surely know better).

Computerchess did not start with Glaurung, Fruit, nor StockFish, but if my son, e.g., today started programming and reading CCC, he would have the impression that SF is everything that ever was and everybody adapted its code and used its ideas. It's rather the opposite that is true (it's a simplification, of course, SF is a marvel or engineering).

While it's true SF authors take everything from others (they plundered Rybka by own admission, but others also) and hardly give proper credit, it's also true the SF community contributed a lot of new stuff. It's the main reason why they are still on top.

Quote :

Practically, everything that is used in SF suddenly becomes a forbidden fruit for everybody else, and open source is a way to control the code/ideas of others developers.

The SF source code is not forbidden fruit, I can use it, add 100 elo to it and publish it. As long as I include the source code with the 100 elo they can do nothing about it. That's the spirit of the GPL, in the name of progress you share in the hope to receive back. And in this hypothetical case they (or everybody else) can take the 100 elo. That's what GPL is all about.

Quote :

And while the SF developers themselves  for years did not care about it, a witch hunt has lately started, putting everybody in the same hat: CB, Eman, CorrChess, etc., and also disrespecting authors who did a lot of work to make SF more usable for the end-users, without thousands of PC at their disposition. It is so petty.

These people deliberately broke the GPL the moment they published their first version refusing to include the source code, thus not willing to share, take but give nothing back in return. I have no sympathy. I am not sure if I am right but I have the feeling this whole affair would not have happened under the lead of Marco.

Quote :

I strongly encourage, support your efforts to improve Pro Deo.  You gave a lot to the "community" (if community there is), explained a lot about chess programming, and part of your knowledge and personal contribution is everywhere. So, if you still enjoy programming, don't restrain yourself but do as you wish. Dogs bark, caravans pass.

Thanks for all the compliments but I will receive the same treatment as with ProDeo 3.1 [Is Rebel a derivative?] when I suddenly starts gaining 80-100 elo every new release because it can't be done without taking well tested ideas from others. Also I am not feeling comfortable with the idea the engine is no longer fully mine.

This must be something I completely oversaw or did not follow. Maybe it was not in the Main forum but in the tournament forum or under a headline I did not follow
Last time I saw Chris successfully participate in some tournaments, then he published game examples and then no further postings.

It would have been enough IMO to present
data /games results and test it against other engines not to make an elo race with others but to show nice saccing games as we would expect from an engine of Chris w.

Elo race and bean counting we let to the stockfish team.

Admin wrote:

Thanks for all the compliments but I will receive the same treatment as with ProDeo 3.1 [Is Rebel a derivative?] when I suddenly starts gaining 80-100 elo every new release because it can't be done without taking well tested ideas from others.

and

Admin wrote:

Chris initial intention was to publish Coronachess including source code. Then he had a lot of problems to subscribe to the tournament of Charles Robinson followed by the accusations of being a derivative. You can't do that with an old timer and certainly not when his name is Chris. So now nobody gets anything. If you accuse provide evidence, not assumptions based on nothing.

When I wrote that

"Practically, everything that is used in SF suddenly becomes a forbidden fruit for everybody else, and open source is a way to control the code/ideas of others developers."

I thought precisely about this "treatment", and I knew what happened to Chris, although I was not the clearest in my post. Similar things happen to everybody who writes a strong engine. GPL code is free to use, but if there is a campaign against reuse of not only code, but ideas, if authors are consistently attacked, [and even the ones who do not reuse this code], it is against the spirit of free software and GPL.

Then, you know very well that, even if an idea is well tested, integrating it in a mature engine is very difficult, and sometimes in does not work. The sum is what counts, not the parts. You have to adapt, modify, and, eventually, to come with something new. I am absolutely sure that Pro Deo would remain an original engine with an interesting style even with a modern search. And, as an end user, I could use it not only in testing openings or playing [sequences] against personalities, but in comparative analysis, without Pro Deo searching 10 plies less than Wasp or Seer. And I do not give non-deserved compliments. Testing sharp Sicilian lines with Benjamin or Q3 is very useful, and surviving 10, 12 plies without being cleaned off the board [in an already analysed position!] is top-training for an amateur. Computerchess is not only about Elo [and 2800 Elo is not something to underestimate].

Finally, there was much written about GPL violations, but, in reality, there was only one, and most of the authors mentioned in the thread did not really breach the license. But it was just another step in an ugly campaign, where the accusations are much louder than the excuses. Even the programmer who indeed broke the GPL published most of his code in another engine, so, it was not completely about not giving back.

I am very happy to see ed and chris and even others (martin bryant) still doing computerchess.

While others do music now …