Firstly - if you have a 64-bit operating system, which is common these days, you would use Glass_tournament_x64.exe for standard play. In order to run personalities, you need to use Glass_Weak_x86.exe, which is just 32-bit. So straight away you lose 20-30 ELO or so by sacrificing the x64 advantage.
Secondly, Glass_Weak_x86.exe, run with default parameters, is perhaps another 20-30 ELO weaker than Glass_tournament_x86.exe.
So, to use personalities, straight away you start approx 50 ELO weaker So, if your aim is to try and improve the strength of Glass you are pretty much wasting your time. If on the other hand you want to create an engine with a particular style for human vs engine play, then of course that is worthwhile.
Anyway, below were my efforts before I decided that this was a waste of time:
- Time Control: 3+1 on a Phenom II at 3.7GHz
- 50 positions chosen randomly from HS-Grand_8-moves-31437.pgn, no ECO duplicates
- each Glass engine plays those positions with reversed colours against 10 opponents - so 10 x 100 = 1,000 games
- ponder off, because Glass seemed to have a ponder bug
Glass 1.7 Default on the list is 32-bit.
Code: Select all
Engine ELO + - Score AvOp Games
Gaviota 0.84 64-bit 2841 +28 -27 77.0% -207.6 500
GarboChess 3.0 64-bit 2724 +25 -25 62.9% -90.6 500
Cheng3 1.05 64-bit 2691 +24 -24 58.7% -57.6 500
Rotor 0.6 2691 +25 -24 58.5% -57.6 500
Philou 3.60 2665 +24 -24 54.7% -31.6 500
Glass 1.7 Default 2663 +18 -18 53.4% -21.2 1000
Glass 1.7 Weak Default 2639 +18 -18 49.8% +2.8 1000
Glass 1.7 Lousy 2627 +18 -18 47.9% +14.8 1000
Glass 1.7 Venom 2625 +18 -18 47.8% +16.8 1000
Glass 1.7 Lucena 2613 +18 -18 46.1% +28.8 1000
GreKo 8.2 2598 +24 -24 45.0% +35.4 500
Rodin 4.0 2596 +24 -24 44.4% +37.4 500
Delphil 2.8 2556 +24 -24 38.2% +77.4 500
Pawny 0.3 64-bit 2533 +25 -25 35.9% +100.4 500
Hussar 0.4 2523 +25 -26 34.7% +110.4 500