For the null move pruning, Stockfish prunes a lot more, which is why it gets deeper. Leptir scrutinizes more bad looking moves, which is why it finds tactical stunners. Stockfish: Depth R = std::min(int(eval - beta) / 165, 6) + depth / 3 + 4 - (complexity > 800);
For the null move pruning, Stockfish prunes a lot more, which is why it gets deeper. Leptir scrutinizes more bad looking moves, which is why it finds tactical stunners. Stockfish: Depth R = std::min(int(eval - beta) / 165, 6) + depth / 3 + 4 - (complexity > 800);
For the null move pruning, Stockfish prunes a lot more, which is why it gets deeper. Leptir scrutinizes more bad looking moves, which is why it finds tactical stunners. Stockfish: Depth R = std::min(int(eval - beta) / 165, 6) + depth / 3 + 4 - (complexity > 800);
Another fun and pretty amazing game. Interesting moment is the dubious 25. b5, somehow burying its own bishop on a6. Of course this is still a draw (it's a complicated and tactical game), but Crafty never understood this as a potential source of problems. Fun is that this resulted in a position where Crafty had 3 pawns for the exchange, quite a lot. Rebel understood the position way better and saw the win at a time where Crafty was still feeling all comfortable for several more moves.
For the null move pruning, Stockfish prunes a lot more, which is why it gets deeper. Leptir scrutinizes more bad looking moves, which is why it finds tactical stunners. Stockfish: Depth R = std::min(int(eval - beta) / 165, 6) + depth / 3 + 4 - (complexity > 800);
The code is clipped out of SF and Leptir without changes.
You probably did not notice this: int complexity = 900; int eval = 2000; int beta = 750; So 7 plies from eval minus beta. It's not a generic prune, but I just stuck some constants into it to show a possible pruning scenario.
This one might look like a book victory at first sight, but it wasn’t . First move out of book for both was 9.ab and both were perfectly happy with their own position for some more moves. Crafty only realized there was a slight problem when Rebel played 16. Nxg6. I really like this game, as I used to think that this line was perfectly fine for black (having looked at it like 20 years ago) – but this doesn’t seem to be the case. Everything looked pretty forced. A light game, that you can just have a look at here. Score is 10.0-0.0 now, and I seriously wonder how Crafty will ever even get a draw in this match. [pgn] [Event "Lang 120min+10sek"] [Site "Berlin"] [Date "2023.04.06"] [Round "?"] [White "Rebel-16.2"] [Black "Crafty 25.6"] [Result "1-0"] [ECO "E04"] [PlyCount "93"] [TimeControl "7200+10"]
This one might look like a book victory at first sight, but it wasn’t . First move out of book for both was 9.ab and both were perfectly happy with their own position for some more moves. Crafty only realized there was a slight problem when Rebel played 16. Nxg6. I really like this game, as I used to think that this line was perfectly fine for black (having looked at it like 20 years ago) – but this doesn’t seem to be the case. Everything looked pretty forced. A light game, that you can just have a look at here. Score is 10.0-0.0 now, and I seriously wonder how Crafty will ever even get a draw in this match.
I wouldn't expect crafty to win a single game given its almost 450 Elo weaker than rebel, and it gets almost no draws against Fritz 16 which is 3200
I wouldn't expect crafty to win a single game given its almost 450 Elo weaker than rebel, and it gets almost no draws against Fritz 16 which is 3200
Interesting - I don't know much about this as I am not up to date with rating lists. But if this is true, there is a clear problem with current Stockfish, as Crafty can definitely draw some games against it with same conditions.
I wouldn't expect crafty to win a single game given its almost 450 Elo weaker than rebel, and it gets almost no draws against Fritz 16 which is 3200
I am pretty sure by now that with enough patience wins for Crafty would happen eventually. Look at today's game. I won't pretend to understand too much here (it is one of these games to look at with the help of powerful engines) , other than: this seems to be a pretty miserable opening line in general (that used to be considered as completely fine). Rebel definitely felt pretty much lost at some point in the game (that featured heavily oscilliating evals). Still - 11.0-0.0 now.
I'm confused because in this game, rebel played perfectly and crafty made a mistake the first move out of book. something important to note and why I don't think crafty could ever win against rebel is, crafty 25.2 on 4,CPU, scored +2-30=8 against fritz 16 on 4CPU, which is rated 3148. almost no wins against an engine 150 elo higher. so when rebel us almost 5-600 elo higher, it seems highly unlikely for a win.
I'm confused because in this game, rebel played perfectly and crafty made a mistake the first move out of book. something important to note and why I don't think crafty could ever win against rebel is, crafty 25.2 on 4,CPU, scored +2-30=8 against fritz 16 on 4CPU, which is rated 3148. almost no wins against an engine 150 elo higher. so when rebel us almost 5-600 elo higher, it seems highly unlikely for a win.
You certainly have a good general point here, maybe the ELO difference is just too high. I disagree about the game though - first move out of book from Crafty was 12. c5 which certainly is no mistake but just logical. And at at least two points in the game Rebel itself thought it was more than 1.0 down.
Damir Desevac
Posts : 319 Join date : 2020-11-27 Age : 43 Location : Denmark
Why are you all degrading Ed's brainchild to Crafty's level..Rebel is much stronger than Crafty, so any of such tests are completely irrelevant. You are wasting both yours and Ed's time with such tests. If you want to test Rebel, you have to find a quality opponent...
Why are you all degrading Ed's brainchild to Crafty's level..Rebel is much stronger than Crafty, so any of such tests are completely irrelevant. You are wasting both yours and Ed's time with such tests. If you want to test Rebel, you have to find a quality opponent...
I feel that you mean well, so I will try to explain. Playing against a weaker opponent is nothing, that weakens or degrades you in general. My personal reason for this setup: 1. I have witnessed about 1000 games under these same exact conditions over several years. I know Crafty better than any other engine. And though I am of course way weaker than it by far I can usually predict most of its moves, or when it will move e.g. 2. So I am changing only one thing, the opponent – and I can watch the games without having to think about Crafty. And it is still a pretty decent player. 3. Rebel is not in active development anymore, so for what other reason than fun should you write about your experiences with it? When it is about rating lists – there are several people who are doing them, and they are doing perfectly fine for all i know. There are also practical reasons – on Crafty’s notebook I don’t have any chess GUI or other engines installed. I love it that the two notebooks are nearly identical performancewise, but I don’t really want to install chessprograms on it. When it is about recent bugs in Rebel I feel that I could still produce reasonable input – be it about time management or be it about book. And I am having fun watching these games live myself. Crafty is an ideal opponent to see what engines are up to and planning to do, as Crafty will just allow them to go along with their plans. If they sac something Crafty will just take it etc. Anyway, I am well aware of the fact that Crafty is no opponent for Rebel. And I also realize that I probably have to change my general testing setup soon for it to be of any use to anyone else than me. It’s just that I will be useless for such a long time, before I find an opponent that I understand nearly a quarter that much AND that can provide a better challenge for modern engines.
Admin Admin
Posts : 2536 Join date : 2020-11-17 Location : Netherlands
To make Crafty a competitive engine again it needs NNUE evaluation and that's not so simple. You need someone with at least a 32-thread PC playing 50-60 million self-play games to create a decent net. Before you start you need to find out at which time control these 50-60 million games must be played. As a rule of thumb we must find out at which depth Crafty is about equal in strength against SF14 at depth=8 to ensure a reasonable strong net.
For ProDeo 3.1 the ideal depth is depth=9, see a 1000 game match result.
Code:
No. Name Win Draw Loss Unf. Score Games % ----------------------------------------------------- 1 sf14 +430 =180 -390 *0 520.0 1000 52.0% [depth=8] 2 ProDeo31 +390 =180 -430 *0 480.0 1000 48.0% [depth=9]
From experience I know 50-60 million self-play games at depth=9 takes ~3 months.
Now it's Crafty turn to figure out the depth to be about equal to SF14. Depth 8-11 were unsuccessful, depth=12 is closing in.
Code:
No. Name Win Draw Loss Unf. Score Games % ------------------------------------------------------ 1 sf14 +535 =155 -310 *0 612.5 1000 61.2% [depth=8] 2 crafty251 +310 =155 -535 *0 387.5 1000 38.8% [depth=12]
We are trying depth=13.
Code:
No. Name Win Draw Loss Unf. Score Games % ------------------------------------------------------ 1 sf14 +487 =187 -326 *0 580.5 1000 58.0% [depth=8] 2 crafty251 +326 =187 -487 *0 419.5 1000 42.0% [depth=13]
We are still not there, depth=14 then.
Code:
No. Name Win Draw Loss Unf. Score Games % ------------------------------------------------------ 1 sf14 +482 =167 -351 *0 565.5 1000 56.5% [depth=8] 2 crafty251 +351 =167 -482 *0 434.5 1000 43.5%[depth=14]
Time to stop, a reasonable net can be made crafty self-play at depth=14, but someone with a 128 core machine needs to volunteer.
Peter Berger
Posts : 126 Join date : 2020-11-20
Subject: Re: Rebel 16.2 release Sat Apr 22, 2023 12:13 am
I am no programmer, so please read the following with a little (or a lot of) patience. I feel that there may be another problem. Recent versions of Crafty introduced some heavy LMR pruning scheme that seems to behave somewhat differently than that of other engines.It allows Crafty to reach impressive depths at times, but it comes with a price. It usually goes along with a 0.00 evaluation, that virtually never changes again whatever time you let it think (even hours). The opponent replies with some pretty obvious move and Crafty AT ONCE realizes that it is toast, as if it had just completely missed the move before. This seems to work ratingwise according to rating lists but I can well imagine that this might give any kind of learner a hard time (given my non-existant understanding of things). Anyway – I wished there were someone who provided this 128 CPU machine to see this new generation kind of engine, that still somehow plays like Crafty .
Admin Admin
Posts : 2536 Join date : 2020-11-17 Location : Netherlands
Subject: Re: Rebel 16.2 release Sat Apr 22, 2023 1:23 am
I see, the idea of playing so many games is to make some sort of blueprint of the evaluation, if Crafty in the early plies (by search) only is looking for material (the 0.00 scores) and later moves in the evaluation that makes a nnue net almost impossible. And yes, searching for material only goes extremely fast.
Throw out duplicates:?? -> There are No duplicates..maybe something wrong with your browser?
Make a screen..as no one else ever told me there are duplicates in your lists
Ipman.
By duplicates, I mean different versions of the same engine. We could probably find 100 versions of stockfish that will beat just about any engine.
Ipmanchess
Posts : 42 Join date : 2022-06-08
Subject: Re: Rebel 16.2 release Wed May 03, 2023 8:20 am
My lists has two version from each engine ,as authors can check there progress each time they have a new version! And also most people wants to know how much stronger the new version is..