Fritz 12meschle wrote:Good news for rybka - thanks ted
just in progress 5' match, 100 games in aquarium.
Out of interest which gui did you carry out your match?
Fire 1.3 Burns Rybka 4
Moderators: hgm, Rebel, chrisw
-
- Posts: 6340
- Joined: Mon Mar 13, 2006 2:34 pm
- Location: Acworth, GA
Re: Fire 1.3 Burns Rybka 4
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers
__________________________________________________________________
Ted Summers
-
- Posts: 1335
- Joined: Thu Mar 09, 2006 5:30 am
- Location: Cabo Frio, Brasil
Re: Fire 1.3 Burns Rybka 4
Thanks, Ted. I play a few games (3min) just for experience.
Seemed equal.
With your test things are more clear.
I am waiting for the match running in the strong Paul hardware.
Seemed equal.
With your test things are more clear.
I am waiting for the match running in the strong Paul hardware.
-
- Posts: 1154
- Joined: Fri Jun 23, 2006 5:18 am
Re: Fire 1.3 Burns Rybka 4
It is quite possible to over-train positional values to your positions...I have done it myself . In fact, a 50+ elo difference in quick games is not out of the question. So when I see testing results showing an unlikely result using relatively few public starting positions, I get suspicious.Albert Silver wrote:I don't actually think Vas uses my suite, since last I heard some of his tests ran in the thousands of games, far beyond the limits of my suite. Plus the test was designed to cover as much of the full gamut of typical positions as possible, meaning no special favoritism.BubbaTough wrote:
I would also recommend using positions that the developers were unlikely to have used for training purposes (i.e. don't use standard positions such as Nunn or Silver or such).
-Sam
The point being, if you prefer, that if you can *book* my suite via playing algorithms (not book moves), then you will have booked chess per se.
Its not meant as a disparagement of your suite (which I love and consider a valuable contribtuion) its just one of those common errors that folks have to worry about when the testers and the developers are in close communication.
-Sam
-
- Posts: 3019
- Joined: Wed Mar 08, 2006 9:57 pm
- Location: Rio de Janeiro, Brazil
Re: Fire 1.3 Burns Rybka 4
I quite understand, I just don`t believe one can *book* the entire suite. A section of openings perhaps, but not overall. The suite contains every single major opening as well as most, if not all, the major types of strategic positions. All balanced in their quantity according to their actual frequency in play according to database statistics. So if you can book the entire suite, that means you 'booked' the French, Sicilian, KID, English, QGD, KIA, Alekhine, Caro-Kann, not to mention isolani (isolated d-pawn), hedgehog, etc. You have effectively 'booked' chess. Add to that that I don't even see games repeated move per move, except perhaps the first few moves, and I just don't think booking the entire suite is possible. At least not without actually just improving the engine's performance overall.BubbaTough wrote:It is quite possible to over-train positional values to your positions...I have done it myself . In fact, a 50+ elo difference in quick games is not out of the question. So when I see testing results showing an unlikely result using relatively few public starting positions, I get suspicious.Albert Silver wrote:I don't actually think Vas uses my suite, since last I heard some of his tests ran in the thousands of games, far beyond the limits of my suite. Plus the test was designed to cover as much of the full gamut of typical positions as possible, meaning no special favoritism.BubbaTough wrote:
I would also recommend using positions that the developers were unlikely to have used for training purposes (i.e. don't use standard positions such as Nunn or Silver or such).
-Sam
The point being, if you prefer, that if you can *book* my suite via playing algorithms (not book moves), then you will have booked chess per se.
Its not meant as a disparagement of your suite (which I love and consider a valuable contribtuion) its just one of those common errors that folks have to worry about when the testers and the developers are in close communication.
-Sam
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."
Re: Fire 1.3 Burns Rybka 4
Again hardly looks like a clone to me ...or any of the others of this family. About time these engines can be talked about in the regular forum. I see even by the die hards around here the clone issue is finally subsiding but i doubt too many will be dishing out some "my bad's" around here. Funny how different the engines actually are huh
BT
BT
-
- Posts: 1154
- Joined: Fri Jun 23, 2006 5:18 am
Re: Fire 1.3 Burns Rybka 4
I am not talking about booking, I am talking about tuning weights (like the value of the pawn, the value of pieces on certain squares, the value of mobility, the value of a rook vs. 2 pieces, and most importantly the tradeoff between these things). I know for a fact that it is possible to tune to your suite such that an engine plays better in it than in a larger selection of positions, because I have done it. The think you did a good job of picking the positions, many more are needed to prevent over-tuning for most automated techniques.Albert Silver wrote:I quite understand, I just don`t believe one can *book* the entire suite. A section of openings perhaps, but not overall. The suite contains every single major opening as well as most, if not all, the major types of strategic positions. All balanced in their quantity according to their actual frequency in play according to database statistics. So if you can book the entire suite, that means you 'booked' the French, Sicilian, KID, English, QGD, KIA, Alekhine, Caro-Kann, not to mention isolani (isolated d-pawn), hedgehog, etc. You have effectively 'booked' chess. Add to that that I don't even see games repeated move per move, except perhaps the first few moves, and I just don't think booking the entire suite is possible. At least not without actually just improving the engine's performance overall.BubbaTough wrote:It is quite possible to over-train positional values to your positions...I have done it myself . In fact, a 50+ elo difference in quick games is not out of the question. So when I see testing results showing an unlikely result using relatively few public starting positions, I get suspicious.Albert Silver wrote:I don't actually think Vas uses my suite, since last I heard some of his tests ran in the thousands of games, far beyond the limits of my suite. Plus the test was designed to cover as much of the full gamut of typical positions as possible, meaning no special favoritism.BubbaTough wrote:
I would also recommend using positions that the developers were unlikely to have used for training purposes (i.e. don't use standard positions such as Nunn or Silver or such).
-Sam
The point being, if you prefer, that if you can *book* my suite via playing algorithms (not book moves), then you will have booked chess per se.
Its not meant as a disparagement of your suite (which I love and consider a valuable contribtuion) its just one of those common errors that folks have to worry about when the testers and the developers are in close communication.
-Sam
-Sam
-
- Posts: 3019
- Joined: Wed Mar 08, 2006 9:57 pm
- Location: Rio de Janeiro, Brazil
Re: Fire 1.3 Burns Rybka 4
Can you show me?BubbaTough wrote:I am not talking about booking, I am talking about tuning weights (like the value of the pawn, the value of pieces on certain squares, the value of mobility, the value of a rook vs. 2 pieces, and most importantly the tradeoff between these things). I know for a fact that it is possible to tune to your suite such that an engine plays better in it than in a larger selection of positions, because I have done it.
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."
-
- Posts: 1154
- Joined: Fri Jun 23, 2006 5:18 am
Re: Fire 1.3 Burns Rybka 4
I don't want to do any retraining (its a bit time consuming), but after my current round of testing (ending this weekend) I can see if I have an old version that acts like that to send you. In exchange for the private version, perhaps you can give me some feedback on the engine . I always value such from strong chess players such as yourself.Albert Silver wrote:Can you show me?BubbaTough wrote:I am not talking about booking, I am talking about tuning weights (like the value of the pawn, the value of pieces on certain squares, the value of mobility, the value of a rook vs. 2 pieces, and most importantly the tradeoff between these things). I know for a fact that it is possible to tune to your suite such that an engine plays better in it than in a larger selection of positions, because I have done it.
-Sam
-
- Posts: 3019
- Joined: Wed Mar 08, 2006 9:57 pm
- Location: Rio de Janeiro, Brazil
Re: Fire 1.3 Burns Rybka 4
It's a deal. Just so you know, I won't limit the test to one opponent. For a 50 Elo diff, I would test against at least 2, for both the before and after, since there is nothing uncommon about gaining a couple of % against one opponent, only to lose it against another.BubbaTough wrote:I don't want to do any retraining (its a bit time consuming), but after my current round of testing (ending this weekend) I can see if I have an old version that acts like that to send you. In exchange for the private version, perhaps you can give me some feedback on the engine . I always value such from strong chess players such as yourself.Albert Silver wrote:Can you show me?BubbaTough wrote:I am not talking about booking, I am talking about tuning weights (like the value of the pawn, the value of pieces on certain squares, the value of mobility, the value of a rook vs. 2 pieces, and most importantly the tradeoff between these things). I know for a fact that it is possible to tune to your suite such that an engine plays better in it than in a larger selection of positions, because I have done it.
-Sam
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."
-
- Posts: 92
- Joined: Tue Jan 26, 2010 5:03 am
Re: Fire 1.3 Burns Rybka 4
1) Probably the easiest OS to use huge (intel) or large (ms) pages with is XP64. All you have to do is start your gui and engine right after booting up. So if you used to use this, you might consider downgrading to it. I've been using huge pages with all my other engines for years now using a third party memory controller (7-max). This program has a memory defragmentation built into it, which should be added to R4 so that you don't have to reboot so often.
Any program including fire 1.3 can use large pages for 15 to 20 percent improvement with 7-max that should keep the edge away from rybka because in my tests rybka 4 is still behind and I used ivanhoe
Any program including fire 1.3 can use large pages for 15 to 20 percent improvement with 7-max that should keep the edge away from rybka because in my tests rybka 4 is still behind and I used ivanhoe