Nebula 2.0 v Djinn 0.979- How Far Has Nebula Actually Come?

[geots]

The only downside to this match is a low number of games from CCRL for Djinn and Nebula 1.5. But maybe I can shed a bit more light in trying to see how far Nebula has progressed.

Nebula 1.5 x64=2472
Djinn 0.979 x64-2573

So let's call it a 100 elo difference.

Now we run Nebula 2.0 pcnt x64 v Djinn 0.979 x64 at 40/20 repeating



Inspiron 620 Intel i5-4 True Cores
Fritz 11 gui
1CPU/64bit
128MB hash
Bases=NONE
Ponder_Learning=OFF
Perfect 2012b.ctg w/12-move limit
40/11 Repeating- Benched to adapt to CCRL 40/20*
Match=100 games


*Please note that CCRL does NOT run 40/20 repeating. However, if they did, I use the same method they would for benchmarking controls.

I have stated before I use their benchmark because, simply- I trust it! Not much I trust out there- but I trust the work of CCRL 120%, any day.




Let the games begin-

george

[geots]

And of course updates will be given. Number of updates depends on the length of the match.


gts

[geots]

Call this a "pre me going to sleep update":


Game 1= drawn
Game 2= drawn
Game 3= Nebula has not given up on the fact that a rook isn't going to mate a bishop.



Best,

[Tom Likens]

The only downside to this match is a low number of games from CCRL for Djinn and Nebula 1.5. But maybe I can shed a bit more light in trying to see how far Nebula has progressed.

Nebula 1.5 x64=2472
Djinn 0.979 x64-2573

So let's call it a 100 elo difference.

Now we run Nebula 2.0 pcnt x64 v Djinn 0.979 x64 at 40/20 repeating



Inspiron 620 Intel i5-4 True Cores
Fritz 11 gui
1CPU/64bit
128MB hash
Bases=NONE
Ponder_Learning=OFF
Perfect 2012b.ctg w/12-move limit
40/11 Repeating- Benched to adapt to CCRL 40/20*
Match=100 games


*Please note that CCRL does NOT run 40/20 repeating. However, if they did, I use the same method they would for benchmarking controls.

I have stated before I use their benchmark because, simply- I trust it! Not much I trust out there- but I trust the work of CCRL 120%, any day.




Let the games begin-

george

Hey George,

I think Nebula 2.0 is going to stomp all over Djinn 0.979 I think 0.979 had an edge over Nebula 1.5, but Dragan in my testing, has more than made up for that. You're test will be interesting, regardless of how it turns out. More interesting yet, will be the match up between what I'm working on now and Dragan's 2.0 beast. I hope to make that more competitive, but of course time will tell.

One thing I will mention, is I like the longer time control test. I have to admit I run most of the matches at a very fast time control, so I don't get back as much data on longer matches.

regards,
--tom

[Dragan]

The reason is that I still didn't add any material evaluation terms other then 2 knights vs king and minor vs minor. I only adjusted the piece values.
This is on my to-do list for the next version. I tried an algorithm based on statistics collected from many games, but this approach made Nebula slightly weaker.

[geots]

The only downside to this match is a low number of games from CCRL for Djinn and Nebula 1.5. But maybe I can shed a bit more light in trying to see how far Nebula has progressed.

Nebula 1.5 x64=2472
Djinn 0.979 x64-2573

So let's call it a 100 elo difference.

Now we run Nebula 2.0 pcnt x64 v Djinn 0.979 x64 at 40/20 repeating



Inspiron 620 Intel i5-4 True Cores
Fritz 11 gui
1CPU/64bit
128MB hash
Bases=NONE
Ponder_Learning=OFF
Perfect 2012b.ctg w/12-move limit
40/11 Repeating- Benched to adapt to CCRL 40/20*
Match=100 games


*Please note that CCRL does NOT run 40/20 repeating. However, if they did, I use the same method they would for benchmarking controls.

I have stated before I use their benchmark because, simply- I trust it! Not much I trust out there- but I trust the work of CCRL 120%, any day.




Let the games begin-

george

Hey George,

I think Nebula 2.0 is going to stomp all over Djinn 0.979 I think 0.979 had an edge over Nebula 1.5, but Dragan in my testing, has more than made up for that. You're test will be interesting, regardless of how it turns out. More interesting yet, will be the match up between what I'm working on now and Dragan's 2.0 beast. I hope to make that more competitive, but of course time will tell.

One thing I will mention, is I like the longer time control test. I have to admit I run most of the matches at a very fast time control, so I don't get back as much data on longer matches.

regards,
--tom

I agree with you about the controls. As for which is stronger of these 2, it is hard to tell "going-in", because CCRL has you 100 points higher than version 1.5. I know he has made improvements and version 2.0 is much stronger than 1.5. But it's awfully hard to tell when the 100 point elo difference that I refer to is based on only 40 to 60 games total for each of you. That is just not enough games to be sure of the actual difference between those versions. It's not a perfect world, but this will be one hell of a good match. 2 very nice engines.


Best,

[geots]

22 games are in the books, and so the first real update!



Inspiron 620 Intel i5-4 True Cores
Fritz 11 gui
1CPU/64-bit
128MB hash
Bases=NONE
Ponder_Learning=OFF
Perfect 2012b.ctg w/12-move limit
40/11 Repeating (Benched to adapt to 40/20)
Match=100 games

Code: Select all

Nebula 2.0 pcnt x64     +6/-4/=12
Djinn 0.979 x64         +4/-6/=12

So at this point, Nebula has a 2 game lead. Very interesting.



Best,

george

[geots]

The reason is that I still didn't add any material evaluation terms other then 2 knights vs king and minor vs minor. I only adjusted the piece values.
This is on my to-do list for the next version. I tried an algorithm based on statistics collected from many games, but this approach made Nebula slightly weaker.

Dragan, another position Nebula has trouble with and doesn't understand: he is on the + side of KBP v K. But he doesn't realize that the corner square his pawn has to reach to promote is the opposite color from his bishop, and you know what that means.


Best,

[geots]

And at the 31 game mark, Nebula still clings to the 2 game lead he has had over the last 10 or so games. And still a long way to go, with some very nice chess being played!




Inspiron 620 Intel i5-4 True Cores
Fritz 11 gui
1CPU/64-bit
128MB hash
Bases=NONE
Ponder_Learning=OFF
Perfect 2012b.ctg w/12-move limit
40/11 Repeating (Benched to adapt to 40/20)
Match=100 games

Code: Select all

Nebula 2.0 pcnt x64    +9/-7/=15
Djinn 0.979 x64        +7/-9/=15

I am still watching for that mini-run I believe one of them will get on. But then again, we shall see............



Best,

[Dragan]

I already have the evaluation code for this, but it's currently disabled.
Never got the free CPU time to test it and it's covered by EGTBs so I wasn't in a big hurry to add it.
I will probably have it in by the next version.
Thanks, Dragan