[Computer-go] Typical ELO for Light Playout Bot

[Petr Baudis]

  Hi!

On Sun, Jan 30, 2011 at 08:14:18PM -0500, Kenny Lam wrote:
> 
> Hi, I'm a newbie so sorry for the silly questions.  I've recently implemented a RAVE bot and ran it on the 9x9 CGOS for about 120 games. The bot seems to be quite weak with an ELO ~1420.  I'm wondering if this score is typical given my methods or if that's abnormally weak and I should be looking for any major bugs. 
> 
> My bot uses RAVE with light playouts fixed at 35k sims per turn.  Leaf nodes must be traversed 5 times before they are expanded. The bot is currently deterministic for debugging purposes if that matters. I appreciate any help. 

  See the list at

	http://senseis.xmp.net/?CGOSBasicUCTBots

for expected performance of various programs with the basic UCB1 formula
and various numbers of playouts. I suggest you check your program
attains the expected rating without RAVE first, then try experimenting
with RAVE. You probably have some bugs to catch or constants to tune.

  There are many variations in RAVE formulas, and e.g. in my experience
it starts being effective at all only with appropriately heavy policy.

				Petr "Pasky" Baudis