New open-source engine, Bit-Genie

[Gabor Szots]

I found the bug. It was in the function that checks if a move was pseudo-legal or not. Have fixed that + the time management issues. I hope its alright now, thank you @lithander!

Does the master include the fixes now?

BTW, may I suggest you to reflect the changes in the name so that not all versions display 1.0.0? E.g. 1.0.x if the fix is minor.

EDIT: the newest master compiles to a 20 % slower exe. I see you have removed -march=native from the makefile and included -static, but those does not seem to account for the slowdown.

[aryan1508]

Sorry for the changes in the make file, they were accidental. The reason you saw a slow down is because of the statically linked compilation process. yes the master includes the fix. You should be able to compile it the same way now, I update the make file.

Thank you for reminding about the version, forgot that I have to update that too

[lithander]

Have fixed that + the time management issues. I hope its alright now, thank you @lithander!

No problem! I look forward to playing your engine again!

[mvanthoor]

...

I ran a tournament with my engine Rustic against Bit-Genie (my own compile from the master-branch, from yesterday). It seems Rustic Alpha 2 is MUCH stronger. Look at that _MASSIVE_ gap in the score:

Code: Select all

Score of Rustic Alpha 2 vs Bit Genie 1.0.0: 84 - 83 - 33 [0.502]
...      Rustic Alpha 2 playing White: 42 - 39 - 19  [0.515] 100
...      Rustic Alpha 2 playing Black: 42 - 44 - 14  [0.490] 100
...      White vs Black: 86 - 81 - 33  [0.512] 200
Elo difference: 1.7 +/- 44.2, LOS: 53.1 %, DrawRatio: 16.5 %
200 of 200 games finished.

Bit-Genie is as fast as Rustic, and exactly as strong over a 200 game match, but Bit-Genie has some more features at this point:

1. Killers + History (which Rustic doesn't have)
2. Rustic has a more efficient hash table with buckets
3. A more extensive evaluation (it seems it's tuned, and has some more terms)
4. Bit-Genie uses SEE, Rustic MVV-LVA.

It seems 1 and 2 cancel each other out: in the early middle-game, Bit-Genie sees a bit further ahead; in the late middle-game and end-game, Rustic sees deeper. This difference, combined with the difference in evaluation, makes for some interesting games. Rustic is stronger tactically, but Bit-Genie is better in strategy.

Observation:

- I had to add "-static-libstdc++ -static" to the make-file to be able to run the engine outside the MSYS2 folder; by default, it's dynamically linked.
- The engine doesn't react to the "stop" or "quit" commands when searching (but if I remember correctly, that has already been mentioned and maybe even already been fixed in a newer build).

[aryan1508]

...

I ran a tournament with my engine Rustic against Bit-Genie (my own compile from the master-branch, from yesterday). It seems Rustic Alpha 2 is MUCH stronger. Look at that _MASSIVE_ gap in the score:

Code: Select all

Score of Rustic Alpha 2 vs Bit Genie 1.0.0: 84 - 83 - 33 [0.502]
...      Rustic Alpha 2 playing White: 42 - 39 - 19  [0.515] 100
...      Rustic Alpha 2 playing Black: 42 - 44 - 14  [0.490] 100
...      White vs Black: 86 - 81 - 33  [0.512] 200
Elo difference: 1.7 +/- 44.2, LOS: 53.1 %, DrawRatio: 16.5 %
200 of 200 games finished.

Bit-Genie is as fast as Rustic, and exactly as strong over a 200 game match, but Bit-Genie has some more features at this point:

1. Killers + History (which Rustic doesn't have)
2. Rustic has a more efficient hash table with buckets
3. A more extensive evaluation (it seems it's tuned, and has some more terms)
4. Bit-Genie uses SEE, Rustic MVV-LVA.

It seems 1 and 2 cancel each other out: in the early middle-game, Bit-Genie sees a bit further ahead; in the late middle-game and end-game, Rustic sees deeper. This difference, combined with the difference in evaluation, makes for some interesting games. Rustic is stronger tactically, but Bit-Genie is better in strategy.

Observation:

- I had to add "-static-libstdc++ -static" to the make-file to be able to run the engine outside the MSYS2 folder; by default, it's dynamically linked.
- The engine doesn't react to the "stop" or "quit" commands when searching (but if I remember correctly, that has already been mentioned and maybe even already been fixed in a newer build).

Yes, I got similar results in my tests too. It's just that some features work better with other features already implemented. I think my feature list is a little messed up in that aspect xD

[aryan1508]

Released v7!
https://github.com/Aryan1508/Bit-Genie/releases/tag/v7

[tmokonen]

Nicely done, it searches 3-4 ply deeper on most moves at 40/1 than version 6, and seems to be quite a bit stronger. One small typo in the UCI options, OwnBoon instead of OwnBook, but that's just cosmetic.

[j.t.]

Impressive how strong it is considering that it doesn't implement that many technical features as far as I can see.

[aryan1508]

Nicely done, it searches 3-4 ply deeper on most moves at 40/1 than version 6, and seems to be quite a bit stronger. One small typo in the UCI options, OwnBoon instead of OwnBook, but that's just cosmetic.

Thanks for pointing it out, corrected it at https://github.com/Aryan1508/Bit-Genie/ ... dd0a32ba37

[Rebel]

Released v7!
https://github.com/Aryan1508/Bit-Genie/releases/tag/v7

You seem to be going strong, hence I have put your engine in a stronger pool, let's see how that goes

Code: Select all

C:\cc>elo-pool @gauntlet-bitgenie.bat

 1. Galjoen_0.41.2 2566.2 [2566.2]
 2. Jumbo_0.7.0 2579.5 [2572.8]
 3. Supernova_2.3 2681.0 [2608.9]
 4. Fruit_2.1 2700.0 [2631.7]
 5. Dumb_1.8 2701.3 [2645.6]
 6. k2_099 2720.6 [2658.1]

Elo pool = 2658.1

http://rebel13.nl/a/grl.htm

700 games.

After 45 games +252 elo