our actual rating lists are online and can be found under the attached links.
40 / 20:
New games: 1597; 41 different engines
Total: 643.061
NEW Engines 676 Rodent 0.18 x64: 2473 - 250 games (starts with around +20 to Version 0.17) 763 Djinn 0.979 x64 : 2427 - 400 games (+77 to Version 0.969)
UPDATES 97 Hannibal 1.3 x64 4CPU: 2879 - 1604 games (-6) 405 Gaviota 0.86 x64 1CPU: 2648 - 820 games (+2) 533 DiscoCheck 4.00 x64 : 2563 - 496 games (-37 and now close to Version 3.62) 876 Nebula 1.5 x64 1CPU: 2356 - 653 games (+27 and now +78 to Version 1.0) 711 Rodent 0.17 x64: 2454 - 701 games (-15!)
Thanks Werner and CEGT team.
I am very happy with almost 80 ELO improvement from my evaluation changes.
My bullet tests indicated this, but I was afraid it will not translate into the longer time controls.
Current development version with LMR and singular extensions is already 100 ELO stronger than 1.5, but I would like to reach 2500 with it.
After that, ELO improvements will be much harder to achieve, because the LMR and SE were the last 'easy ELO' techniques that were missing in Nebula.
Dragan wrote:Thanks Werner and CEGT team.
I am very happy with almost 80 ELO improvement from my evaluation changes.
My bullet tests indicated this, but I was afraid it will not translate into the longer time controls.
Current development version with LMR and singular extensions is already 100 ELO stronger than 1.5, but I would like to reach 2500 with it.
After that, ELO improvements will be much harder to achieve, because the LMR and SE were the last 'easy ELO' techniques that were missing in Nebula.
Dragan,
I've been very impressed with your progress on Nebula in a short amount of time. It's becoming a very strong engine, well done.
Thanks Tom.
As I mentioned, improvements come easy at first, because you just implement standard techniques that every engine has.
The trick is in tuning and making it all work together and doing it without introducing major bugs.
After 2.0 my rate of improvement will most likely slow down. All the standard stuff will be implemented and I want to work on engine style a bit.
Don't like the way Nebula evaluates some positions. My positional eval scores seem to be too low.
Cheers, Dragan
Dragan wrote:Thanks Tom.
As I mentioned, improvements come easy at first, because you just implement standard techniques that every engine has.
The trick is in tuning and making it all work together and doing it without introducing major bugs.
After 2.0 my rate of improvement will most likely slow down. All the standard stuff will be implemented and I want to work on engine style a bit.
Don't like the way Nebula evaluates some positions. My positional eval scores seem to be too low.
Cheers, Dragan
Yeah, I've only started to make real progress again by using "cutechess-cli" to run thousands and thousands of games. It's slow and tedious but I don't know of any other way to make real progress. I'm also better now at resisting the urge to change 10 things at once, which makes a hash out of trying to really understand the results. It's an asymptotic hill, but at least it's still going up!
I spent $2000 on i7 3930 just to do the testing. My electricity bill more than doubled
Learned the hard way you can't test multiple changes at once. No matter how small or 'obvious' they are.
BTW, you made a tactical mistake replying to my post. Djinn just entered my test opponent list (hope it can handle bullet controls)
Dragan wrote:I spent $2000 on i7 3930 just to do the testing. My electricity bill more than doubled
Learned the hard way you can't test multiple changes at once. No matter how small or 'obvious' they are.
BTW, you made a tactical mistake replying to my post. Djinn just entered my test opponent list (hope it can handle bullet controls)
Dragan,
I'm glad it made the cut. Actually, I've spent a lot of time recently on the time management of Djinn, in fact I even added it to my webpage,
So it should handle just about any (semi-)reasonable time control (I think I tested it down to 1 sec. + 0.1). If it crashes or loses on time, please let me know as I really want it to be rock solid at the faster time controls.
