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

[Tom Likens]

Tom, a unique way to address the bullet game issue- trash it. Just a thought.................

I think ignoring it would be perfectly valid. Especially, if you were trying to optimize things for longer time controls. Ultimately, though I created the engine not only for my own enjoyment but hopefully for others as well. And as long as people use the fast time controls, I'd like to support them. Ironically, in this age of faster and faster machines, we're all running shorter and shorter matches.

This match has been eye-opening for me though. Especially, since most people have reported that the performance of their engines are usually consistent between fast and slow time controls (to within some +/- epsilon). This could show a deficiency in Djinn that might lay the groundwork for a nice improvement. At the very least I'm going to dig into it to see what I can find. I can't tell you how many times in my career as an engineer I've found major improvements in designs that started with the phrase, "...hmm, that doesn't seem quite right?!". Things that seem slightly off can lead to incredible optimizations or uncover massive bugs.

regards,
--tom

[geots]

A very good match from start to finish, and it was very enjoyable to watch what I could of it. In the end, Nebula came on late and put it out of reach. But the outcome was in doubt most of the way. The fact this Djinn version was rated 100 elo stronger than Nebula 1.5 unfortunately does not tell us the exact progress of Nebula. As the games those figures were based on were for each engine less than the 100 each they played here. We know for sure Nebula has made a vast improvement- but it will take more games to get a better idea exactly how much. At any rate- you have to give it to Dragan- he has done some damn fine work. And let's not forget Tom and Djinn, as they were in it till the bitter end. And Tom has a few cards up his sleeve still left to play.

I will at this point, instead of waiting and putting it at the bottom of this thread, go ahead and give Dragan and Tom a link to the complete 100-game PGN set, as well as a very nice chessbase crosstable along with it:

http://dl.dropbox.com/u/115682714/Nebul ... 00.979.rar

(And of course for anyone else who may be interested)





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

SP8-i5, 11'/40+11'/40+11'/40  3/15/2013  

                         
Nebula 2.0 pcnt x64   +28    +34/-26/=40   54.00%   54.0/100
Djinn 0.979 x64       -28    +26/-34/=40   46.00%   46.0/100

You know, these testing groups do a very good service for the chess community. But so many engines now- extremely hard if not impossible for any group to keep up with all of them. And these 2 engines just have not had enough of a chance to show what they can do. I like to feel that maybe I could step in and help a bit- getting them some more exposure they deserve. And we are of course not thru yet.

But let's give Dragan his due and a chance to savor this very fine victory. I am just glad I could in some way play a small part in it.



Best and we shall be back-

george

[Graham Banks]

Code: Select all

Nebula 2.0 pcnt x64   +28    +34/-26/=40   54.00%   54.0/100
Djinn 0.979 x64       -28    +26/-34/=40   46.00%   46.0/100

Thanks George. This was very useful to know.

[Dragan]

Thanks for the test and games George.
And for reminding me about the kbp vs kp draw detection. It's been tested and added to the engine.
Cheers, Dragan

[geots]

Code: Select all

Nebula 2.0 pcnt x64   +28    +34/-26/=40   54.00%   54.0/100
Djinn 0.979 x64       -28    +26/-34/=40   46.00%   46.0/100

Thanks George. This was very useful to know.

Than you for the compliment, Graham. I appreciate that.


Best,

[geots]

Thanks for the test and games George.
And for reminding me about the kbp vs kp draw detection. It's been tested and added to the engine.
Cheers, Dragan

Dragan- don't forget that Nebula doesn't realize KB v KR is a dead-draw. But let me take this time to tell you what wonderful work you have done. You have extremely wonderful programming ability- and you deserve to be proud. And I am proud of you as well.


Best,

[geots]

And if I were you, I might not get too far into testing Djinn 0.979 x64. Along with trying to keep up with the beta testing of Komodo, Strelka 5.6, Equinox X beta- I am also beta testing Djinn for Tom. And I doubt it is going to be really long before he comes out with his next version- which is completely different from 0.979.

Just a thought...........


george

[Graham Banks]

...... I might not get too far into testing Djinn 0.979 x64. Along with trying to keep up with the beta testing of Komodo, Strelka 5.6, Equinox X beta- I am also beta testing Djinn for Tom. And I doubt it is going to be really long before he comes out with his next version- which is completely different from 0.979.

Just a thought...........

george

Yeah - as a tester, it's a pain in the butt when versions are released too regularly, but hey, that's just something we have to be prepared to accept.