Rebel 14

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

User avatar
Rebel
Posts: 7208
Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder

Re: Rebel 14

Post by Rebel »

About the choice for the name REBEL.

I can understand the criticism. At development time the internal name "Fruity" was used, the compiler still produces "fruitig.exe" (dutch for fruity). The reason to call it REBEL (and not ProDeo or Fruit) is that it doesn't play like Fruit, nor like ProDeo, with NNUE it won back its original attractive REBEL playing style of the 90's, Rebel Century and Rebel 12 (both DOS) being the last ones.

As many of you know very well all that massive pruning delivers a lot elo, the side effect is perhaps less known, or it is known but not cared about because elo rules. Point of massive pruning is you also prune in the chess knowledge you worked so hard on for years, it affects the playing style of an engine.

While it's true ProDeo gained 200-250 elo over the last Rebel (DOS) 90% due to search changes I hardly recognize the engine any longer due to all the pruning. NNUE changed all that and much better. NNUE makes a blue print of your hard worked on HCE evaluation without pruning.

-----

About the choice for the FRUIT search and not using my own.

When I retired in 2002/3 I was using an up to date compiler (Digital Mars, formally Symantec) of that time but with my retirement the compiler guys retired also. No more updates. No 64 bit support, no SSE support, and also Rebel / ProDeo and Benjamin are all 32 bit assembler code. Moving to NNUE requires a modern compiler and C or C++ is a must. And I have one, Gideon 1992/93 is entirely in C and it compiles nicely with VS-2019. But to go back 30 years in time and redo all the changes... I checked --> mission (almost) impossible.

As I wanted to do NNUE anyway anyhow I looked for suited inactive GPL engines similar in strength with the latest ProDeo. And Growing Fruit looked excellent for that purpose. There are many other inactive engines out there and much stronger, some even without any licence. I have no intention to step back in the elo arena.
90% of coding is debugging, the other 10% is writing bugs.
User avatar
mclane
Posts: 18802
Joined: Thu Mar 09, 2006 6:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub

Re: Rebel 14

Post by mclane »

You could choose a multi cpu engine like senpai, fruits successor.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....
chrisw
Posts: 4472
Joined: Tue Apr 03, 2012 4:28 pm
Location: Midi-Pyrénées
Full name: Christopher Whittington

Re: Rebel 14

Post by chrisw »

mclane wrote: Thu Jan 13, 2022 3:26 pm You could choose a multi cpu engine like senpai, fruits successor.
He can use a blistering fast skin and bones search bitboard engine from me, with all subroutine Chessic querying attached, plus MP, plus uci. Then all that need be done is one NNUE (well already got that), and feed in the search trickery. Ed likes to use calculations and so on from eval, but I think he has to do that anyway with “Fruitig”. I guess mine could even support a parallel 0x88 structure or whatever dumb name that has nowadays (mailbox?). Although actually, Fruitig now has a “hidden” bitboard structure used to feed the net.
It’s not where I’ld start, but I’m not Ed.
User avatar
Rebel
Posts: 7208
Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder

Re: Rebel 14

Post by Rebel »

Ed is happy with the regained playing style. Search is an necessary evil :wink:
90% of coding is debugging, the other 10% is writing bugs.
Vernon Crawford
Posts: 73
Joined: Wed Sep 01, 2021 2:05 am
Location: London, England
Full name: Vernon Crawford

Re: Rebel 14

Post by Vernon Crawford »

Rebel wrote: Thu Jan 13, 2022 10:38 am About the choice for the name REBEL.

I can understand the criticism. At development time the internal name "Fruity" was used, the compiler still produces "fruitig.exe" (dutch for fruity). The reason to call it REBEL (and not ProDeo or Fruit) is that it doesn't play like Fruit, nor like ProDeo, with NNUE it won back its original attractive REBEL playing style of the 90's, Rebel Century and Rebel 12 (both DOS) being the last ones.

As many of you know very well all that massive pruning delivers a lot elo, the side effect is perhaps less known, or it is known but not cared about because elo rules. Point of massive pruning is you also prune in the chess knowledge you worked so hard on for years, it affects the playing style of an engine.

While it's true ProDeo gained 200-250 elo over the last Rebel (DOS) 90% due to search changes I hardly recognize the engine any longer due to all the pruning. NNUE changed all that and much better. NNUE makes a blue print of your hard worked on HCE evaluation without pruning.

-----

About the choice for the FRUIT search and not using my own.

When I retired in 2002/3 I was using an up to date compiler (Digital Mars, formally Symantec) of that time but with my retirement the compiler guys retired also. No more updates. No 64 bit support, no SSE support, and also Rebel / ProDeo and Benjamin are all 32 bit assembler code. Moving to NNUE requires a modern compiler and C or C++ is a must. And I have one, Gideon 1992/93 is entirely in C and it compiles nicely with VS-2019. But to go back 30 years in time and redo all the changes... I checked --> mission (almost) impossible.

As I wanted to do NNUE anyway anyhow I looked for suited inactive GPL engines similar in strength with the latest ProDeo. And Growing Fruit looked excellent for that purpose. There are many other inactive engines out there and much stronger, some even without any licence. I have no intention to step back in the elo arena.
Ed, Rebel 14 contains a new improved +100 Elo search with quite a bit of new pruning and reductions compared to Fruit 2.1:

// late move reduction
// static null move / beta pruning
// calculate reduction - simplified Stockfish formula
// Razoring
// late move pruning

On my PC, Rebel 14 searches to depth 17 in 5 secs while Fruit 2.1 only achieves depth 13.

If you're so against pruning and not interested in Elo, why did you implement PK's new heavily Stockfish influenced search?
connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Rebel 14

Post by connor_mcmonigle »

Vernon Crawford wrote: Thu Jan 13, 2022 6:03 pm ...
If you're so against pruning and not interested in Elo, why did you implement PK's new heavily Stockfish influenced search?
Outside of a log based reduction formula, what makes PK's changes "heavily Stockfish influenced". All these other heuristics predate Stockfish. Have you even read the source or are you just making baseless claims?
Eduard
Posts: 1439
Joined: Sat Oct 27, 2018 12:58 am
Location: Germany
Full name: N.N.

Re: Rebel 14

Post by Eduard »

I don't want to criticize, but the engine is very weak. In almost every test position, the wrong move is preferred.

For this position Rebel 14 needs 2 minutes and 30s:

[fen]8/p5p1/1p4k1/1P1PQp2/Pq4p1/6P1/4K2P/8 b - - 0 66[/fen]



Like in the stone age.
Last edited by Eduard on Thu Jan 13, 2022 6:48 pm, edited 1 time in total.
User avatar
RubiChess
Posts: 631
Joined: Fri Mar 30, 2018 7:20 am
Full name: Andreas Matthies

Re: Rebel 14

Post by RubiChess »

Maybe Rebel doesn't like the bishop on i1.

Edit: Come on, don't edit the board, now my comment reads really stupid :-)
Last edited by RubiChess on Thu Jan 13, 2022 6:52 pm, edited 1 time in total.
Vernon Crawford
Posts: 73
Joined: Wed Sep 01, 2021 2:05 am
Location: London, England
Full name: Vernon Crawford

Re: Rebel 14

Post by Vernon Crawford »

connor_mcmonigle wrote: Thu Jan 13, 2022 6:20 pm
Vernon Crawford wrote: Thu Jan 13, 2022 6:03 pm ...
If you're so against pruning and not interested in Elo, why did you implement PK's new heavily Stockfish influenced search?
Outside of a log based reduction formula, what makes PK's changes "heavily Stockfish influenced". All these other heuristics predate Stockfish. Have you even read the source or are you just making baseless claims?
// late move reduction
// static null move / beta pruning
// calculate reduction - simplified Stockfish formula
// Razoring
// late move pruning

Connor, the comments I posted above are copied from Rebel 14 source code, and these features are not present in Fruit 2.1.
Please refrain from knee-jerk reactions and accusations without due diligence.

My question was addressed to Ed, why so eager to defend him any question or comment?
Last edited by Vernon Crawford on Thu Jan 13, 2022 6:51 pm, edited 1 time in total.
User avatar
Rebel
Posts: 7208
Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder

Re: Rebel 14

Post by Rebel »

Vernon Crawford wrote: Thu Jan 13, 2022 6:03 pm
Rebel wrote: Thu Jan 13, 2022 10:38 am About the choice for the name REBEL.

I can understand the criticism. At development time the internal name "Fruity" was used, the compiler still produces "fruitig.exe" (dutch for fruity). The reason to call it REBEL (and not ProDeo or Fruit) is that it doesn't play like Fruit, nor like ProDeo, with NNUE it won back its original attractive REBEL playing style of the 90's, Rebel Century and Rebel 12 (both DOS) being the last ones.

As many of you know very well all that massive pruning delivers a lot elo, the side effect is perhaps less known, or it is known but not cared about because elo rules. Point of massive pruning is you also prune in the chess knowledge you worked so hard on for years, it affects the playing style of an engine.

While it's true ProDeo gained 200-250 elo over the last Rebel (DOS) 90% due to search changes I hardly recognize the engine any longer due to all the pruning. NNUE changed all that and much better. NNUE makes a blue print of your hard worked on HCE evaluation without pruning.

-----

About the choice for the FRUIT search and not using my own.

When I retired in 2002/3 I was using an up to date compiler (Digital Mars, formally Symantec) of that time but with my retirement the compiler guys retired also. No more updates. No 64 bit support, no SSE support, and also Rebel / ProDeo and Benjamin are all 32 bit assembler code. Moving to NNUE requires a modern compiler and C or C++ is a must. And I have one, Gideon 1992/93 is entirely in C and it compiles nicely with VS-2019. But to go back 30 years in time and redo all the changes... I checked --> mission (almost) impossible.

As I wanted to do NNUE anyway anyhow I looked for suited inactive GPL engines similar in strength with the latest ProDeo. And Growing Fruit looked excellent for that purpose. There are many other inactive engines out there and much stronger, some even without any licence. I have no intention to step back in the elo arena.
Ed, Rebel 14 contains a new improved +100 Elo search with quite a bit of new pruning and reductions compared to Fruit 2.1:

// late move reduction
// static null move / beta pruning
// calculate reduction - simplified Stockfish formula
// Razoring
// late move pruning

On my PC, Rebel 14 searches to depth 17 in 5 secs while Fruit 2.1 only achieves depth 13.

If you're so against pruning and not interested in Elo, why did you implement PK's new heavily Stockfish influenced search?
Read again, the red.
90% of coding is debugging, the other 10% is writing bugs.