RE: [Bug-gnubg] report of bug

[Ian Shaw]

Jim,

Your test uses the hint function, with gnubg as human. All you have
shown is that the hint function works. The problem seems to occur when
player 0 is set to be gnubg with a high number of plies. 

I found that when player 0 was set to 4-ply play, it moved very quickly,
much quicker than I'd expect 2-ply to play.

I found the same issue whether using the gui or cli, which is slightly
different than Wojciech reported.

When looking back, through the analysis in the gui, plays seemed to be
evaluated at 1 ply.

Possible causes:
1) The chequer plays were simple enough to only require 1-ply.
2) Gnubg only supports chequer play with up to 3 plies, even though the
interface allows higher plies. (In comparison, you can also set the
player to use a rollout for each play, but I know this isn't implemented
yet.)
3) The default move filters for 4-ply are set such that only a 1-ply
evaluation passes through.
4) Other

-- Ian

> -----Original Message-----
> From: address@hidden 
> [mailto:address@hidden 
> On Behalf Of Jim Segrave
> Sent: 01 December 2007 12:46
> To: Wojciech Zaremba
> Cc: address@hidden
> Subject: Re: [Bug-gnubg] report of bug
> 
> On Sat 01 Dec 2007 (09:01 +0100), Wojciech Zaremba wrote:
> > Yes, I know, but I think that on the gnubg-cli options with ply 
> > doesn't work.
> 
> See below - it does work.
> 
> > 2007/11/30, Jim Segrave <address@hidden>:
> > >
> > > On Fri 30 Nov 2007 (15:32 +0100), Wojciech Zaremba wrote:
> > > > In gnubg-cli doesn't work command "set player gnubg chequerplay
> > > evaluation
> > > > plies 4" and/or "set player gnubg cubedecision evaluation plies 
> > > > 4". When
> > > I
> > > > use that option in gnubg I must wait because computer compute. 
> > > > Using
> > > gnubg
> > > > from command line it's so fast like without that option 
> (I'm also 
> > > > try to
> > > use
> > > > option "set player gnubg chequerplay evaluation plies 5" and I 
> > > > have this same effect in gnubg-cli). I played a lot 
> with computer 
> > > > using command
> > > line.
> > > > It's the reason why I thought that computer isn't so good :).
> > >
> > > If you make the assumption that each ply will make the computer 
> > > takes perhaps 100 times as many evaluations as the 
> previous ply, you 
> > > can see what's happening here (I think the multiplier may 
> be nearer 
> > > 200 to 300.
> 
> I just ran a pair of tests using command files and gnubg in cli mode.
> 
> For the first test, I used an opening roll of 6-4, with a command
> file:
> ================
> set evaluation chequerplay evaluation plies 0 set evaluation 
> cubedecision evaluation plies 0 set player 0 human new match 
> 7 set dice 6 4 clear hint hint quit y =========== The set 
> player human prevents gnubg from trying to play an initial 
> move. There were copies of this file with plies set between 0 
> and 4 inclusive. The 'clear hint' is essential, otherwise it 
> uses the evaluation settings for the player on roll.
> 
> Using a command:
> 
> for i in 4 3 2 1 0
> do
> time -p gnubg -t -c /tmp/pos$i > /tmp/hint$i done
> 
> I got the time for each of the plies and a check that the 
> hints were done at the requested number of plies. The real 
> time is wall clock time and is heavily influenced by my 
> laptop pausing when the CPU temperature became too high (dual 
> core CPU running with 2 gnu threads)
> 
> plies 4:
> real 155.82
> user 101.78
> sys 0.07
> 
> plies 3:
> real 10.46
> user 8.49
> sys 0.02
> 
> plies 2:
> real 1.34
> user 1.31
> sys 0.02
> 
> plies 1:
> real 0.17
> user 0.15
> sys 0.02
> 
> plies 0:
> real 0.09
> user 0.06
> sys 0.02
> 
> Here 8 moves survive 2 ply evaluation and 2 make it to 3 and 4 ply.
> 
> 
> I then tried a midgame arbitrary position:
> 
> set evaluation chequerplay evaluation plies 0 set evaluation 
> cubedecision evaluation plies 0 set matchid QYn1AEAAIAAA
> set board   2M7BAyDYzsEDIA
> set player 0 human
> set dice 43
> clear hint
> hint
> quit
> y
> 
> and got timings of:
> 
> plies: 4
> real 89.09
> user 88.94
> sys 0.09
> 
> plies: 3
> real 7.07
> user 7.04
> sys 0.02
> 
> plies: 2
> real 52.17
> user 1.10
> sys 0.02
> 
> plies: 1
> real 0.14
> user 0.12
> sys 0.02
> 
> plies: 0
> real 0.08
> user 0.06
> sys 0.02
> 
> In this case 6 moves go to 2 ply evaluation, 2 continue on to 
> 3 or 4 ply evaluation
> 
> -- 
> Jim Segrave           address@hidden
> 
> 
> 
> 
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg
>