Interview with Jay Honnold author of the Berserk chess engine.

[Dariusz]

I invite you to read.
https://www.chessengeria.com/post/inter ... ay-honnold

[Frank Quisinsky]

Hi,

I like interviews a lot.



Perhaps the start point for a new season of interviews?

To the interview:
I like it, very realistic and interesting what the Berserk author gave on information.

I wish me that programmers pay more attention to "move average of draw games".
Very important in times draw quotes goes higher and higher!!

Contempt today, in NN times, isn't very interesting.
To try to end drawn positions quickly is important.
All the long draw games are very boring.
All the problems a lot of engines have with mating are very boring.

I like Berserk, not the move average of draws.

Have a look here:
80min games without resign-mode on 4.4Ghz / 1 Core / 768Mb hash

Code: Select all

 8. Berserk 8.5.1 NN               (791+,   1096=,   78-)  68.1%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         97     1965
           white won         87      522
           black won         88      269
          white draw        118      453
          black draw         96      643
          white loss         88       10
          black loss         86       68
            all wins         87      791     1965      40.3%
           all draws        105     1096     1965      55.8%
           all loses         86       78     1965       4.0%

97 moves???
The average from around 100.000 80min games is 90 moves!

If we compare the move average with strong engines in tactcial chess ... really to high!!

Same for a lot of other very strong engines, like Koivisto for an example.
Maybe an order with tricky settings to try a lesser move-average?

Other things I don't like ...

In school times many readers here have a "best-friends" book.
More the girls as the guys!
End of readme on github sites you can see the old idea from the 60er ...

Speaking from question 9 by Darius and the answer!

Most of programmers forgot the computer chess history "!!!"
If a programmer start a best-friends list ... in such a list must be inside around 1.000 names.
Because, without all the work by others ... never we have such strong engines today!
Forgotten, and forgotten, and forgotten ... again and again!

So in my opinion ...
A list of "best-friends" is very out and for me a bit children's-like!
Hard but my own opinion!

But it seems that the programmer of Berserk is a very cool guy!!!
95% of answers I like a lot!
Berserk will have a great future, engine is very strong!

Thank you Dariusz and Jay!

Best
Frank

[Frank Quisinsky]

Code: Select all

52. Koivisto 8.6 NN                (432+,   623=,   45-)  67.6%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games        100     1100
           white won         91      284
           black won         87      148
          white draw        116      256
          black draw        103      367
          white loss         83       10
          black loss         82       35
            all wins         89      432     1100      39.3%
           all draws        109      623     1100      56.6%
           all loses         82       45     1100       4.1%

Speaking from Koivisto and forgot the example!

An other example:
Here the programmer is thinking on the important ressources (waste time and electricity is the topic).

Really fantastic stats for games without resign-mode!!

Code: Select all

70. Revenge 2.0 NN                 (897+,   1289=,   109-)  67.2%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         83     2295
           white won         79      581
           black won         83      316
          white draw         86      549
          black draw         84      740
          white loss         94       20
          black loss         91       89
            all wins         81      897     2295      39.1%
           all draws         85     1289     2295      56.2%
           all loses         92      109     2295       4.7%

[mar]

!goJay

[Graham Banks]

Very good.

Succinct and to the point - no waffle.

[Dariusz]

Hi,

I like interviews a lot.



Perhaps the start point for a new season of interviews?

To the interview:
I like it, very realistic and interesting what the Berserk author gave on information.

I wish me that programmers pay more attention to "move average of draw games".
Very important in times draw quotes goes higher and higher!!

Contempt today, in NN times, isn't very interesting.
To try to end drawn positions quickly is important.
All the long draw games are very boring.
All the problems a lot of engines have with mating are very boring.

I like Berserk, not the move average of draws.

Have a look here:
80min games without resign-mode on 4.4Ghz / 1 Core / 768Mb hash

Code: Select all

 8. Berserk 8.5.1 NN               (791+,   1096=,   78-)  68.1%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         97     1965
           white won         87      522
           black won         88      269
          white draw        118      453
          black draw         96      643
          white loss         88       10
          black loss         86       68
            all wins         87      791     1965      40.3%
           all draws        105     1096     1965      55.8%
           all loses         86       78     1965       4.0%

97 moves???
The average from around 100.000 80min games is 90 moves!

If we compare the move average with strong engines in tactcial chess ... really to high!!

Same for a lot of other very strong engines, like Koivisto for an example.
Maybe an order with tricky settings to try a lesser move-average?

Other things I don't like ...

In school times many readers here have a "best-friends" book.
More the girls as the guys!
End of readme on github sites you can see the old idea from the 60er ...

Speaking from question 9 by Darius and the answer!

Most of programmers forgot the computer chess history "!!!"
If a programmer start a best-friends list ... in such a list must be inside around 1.000 names.
Because, without all the work by others ... never we have such strong engines today!
Forgotten, and forgotten, and forgotten ... again and again!

So in my opinion ...
A list of "best-friends" is very out and for me a bit children's-like!
Hard but my own opinion!

But it seems that the programmer of Berserk is a very cool guy!!!
95% of answers I like a lot!
Berserk will have a great future, engine is very strong!

Thank you Dariusz and Jay!

Best
Frank

Frank, thanks for the cool and constructive comment.

Perhaps the start point for a new season of interviews?

Let us hope so


You pointed out some very interesting data regarding the large number of moves in draw games. This is indeed boring and deserves to be changed by the programmers.

If a programmer start a best-friends list ... in such a list must be inside around 1.000 names.
Because, without all the work by others ... never we have such strong engines today!
Forgotten, and forgotten, and forgotten ... again and again!

100% correct!
We live in very interesting times when engines are playing at skyrocketing levels. Even 20 years ago, I wouldn't have dared to think that computers would play chess so well and with such power. This is not a coincidence but a result of passion, determination and hard work of so many talented people.

I'm very glad you enjoyed the interview, thx

[Dariusz]

Very good.

Succinct and to the point - no waffle.

Thank you Graham, I really appreciate that.

Jay is a very nice person and an engaged conversationalist. I look forward to the new Berserk

[Frank Quisinsky]

Hi Darius,

in a lot of cases the very high number of moves in draw games is a result from "contempt".
Engines avoid the 50-moves-rule. In cases of Berserk or Koivisto, others no UCI contempt parameter.
A big problem for games without resign-mode.

It makes more sense to try to force draws in the end of the mid-games if a longer time the evaluation is equal.
The final result is more games in the same time.

Example:
Move average per game is for Koivisto or Berserk in comparsion to Revenge (the three programs have around the same playing strength) 15-17 moves higher. For 1.000 games = 15.000 - 17.000 moves.

15.000 : 85 moves = 176 games!

In the time Revenge play 1.176 games ... Berserk or Koivisto can play 1.000 games.

- ressources are more important as 2-3 Elo performance.
- a more nice playing style is more important as 2-3 Elo performance.

Best
Frank

PS: I added yesterday your link on my link-selection and will follow what you do.
Thanks for the work you do for us.

[Dokterchen]

Great! Thank you

[Dariusz]

Hi Darius,

in a lot of cases the very high number of moves in draw games is a result from "contempt".
Engines avoid the 50-moves-rule. In cases of Berserk or Koivisto, others no UCI contempt parameter.
A big problem for games without resign-mode.

It makes more sense to try to force draws in the end of the mid-games if a longer time the evaluation is equal.
The final result is more games in the same time.

Example:
Move average per game is for Koivisto or Berserk in comparsion to Revenge (the three programs have around the same playing strength) 15-17 moves higher. For 1.000 games = 15.000 - 17.000 moves.

15.000 : 85 moves = 176 games!

In the time Revenge play 1.176 games ... Berserk or Koivisto can play 1.000 games.

- ressources are more important as 2-3 Elo performance.
- a more nice playing style is more important as 2-3 Elo performance.

Best
Frank

PS: I added yesterday your link on my link-selection and will follow what you do.
Thanks for the work you do for us.

Frank, Indeed, contempt is very important in the context of draws.
(I wrote about contempt and other parameters here https://www.chessengeria.com/post/engine-parameters, maybe this information will be useful for someone).

I like it when engine authors provide parameters that can be changed. We can then change a lot. It can have a big effect on the engine behavior... but sometimes it is worth it....

- ressources are more important as 2-3 Elo performance.
- a more nice playing style is more important as 2-3 Elo performance.

Exactly right!

PS: I added yesterday your link on my link-selection and will follow what you do.
Thanks for the work you do for us.

Thank you for highlighting my work in this way. It means a lot to me and is motivating for further activities.

I published the blog just a few weeks ago. This is the first such undertaking for me. Therefore, using this opportunity I would like to ask everyone for "green light" for the beginning and also apologize for any slip-ups and errors in my work - I will always correct them and do not allow for another.

Thank you