Author: Peter Kappler
Date: 09:25:29 10/10/99
Go up one level in this thread
On October 10, 1999 at 05:23:01, Ed Schröder wrote: >>Posted by Harald Faber on October 10, 1999 at 03:22:13: >> >>In Reply to: PB-ON vs PB-OFF (results experiment-1) posted by Ed Schröder >>on October 10, 1999 at 03:14:10: >> >>On October 10, 1999 at 03:14:10, Ed Schröder wrote: >> >>>PB-ON vs PB-OFF (results experiment-1) >>> >>>To enrich the discussion about the value of the "Permanent Brain" (PB) I >>>have started 2 experiments with Rebel Century (RC) which will give some >>>data for a better judgement. >>> >>>Experiment-1: >>>RC (PB=ON) vs RC (PB=OFF) >>>100 auto232 games >>>Time control: 60 secs average. >>>Hardware: 4xPII-266 + 2xPII-450 >>>Result: 61-39 >>> >>>Experiment-2: >>>RC (PB=ON) vs RC (PB=OFF) >>>100 auto232 games >>>Time control: RC (PB=ON) 30 sec average >>>Time control: RC (PB=OFF) 60 sec average >>>Hardware: 4xPII-266 + 2xPII-450 >>>Status: in progress >>> >>>Ed Schroder >> >>Sorry Ed, but where is the sense in it? Is there any difference than playing >>Rebel10-Rebel9 which is also meaningless? >>Would you expect a result of 50-50 when both playing PB=on or both PB=off? I >>wouldn't. >>I think the most interesting idea is to take Rebel+PB=on against another >>program >>with PB=on and as comparison play the same match with both PB=off. Of >>course you >>can extend this to Rebel+PB-vs-Opp X PB=off and Rebel PB=off vs Opp X PB=on. >> >>But Rebel vs Rebel, sorry, there is really no sense in it. > >I think you missed the discussion of the value of the PB. There are many who >believe the PB is only worth 20-40 elo points. I believe this is not true and so >I started these 2 experiments. I believe the value of the PB is worth AT LEAST >50 elo points but most probably 100 elo and I wouldn't be surprised if it will >be higher. > >Now you can argue about playing Rebel-Rebel as a base. The big advantage >playing Rebel-Rebel is 2 ways: a) you have a 100% equal opponent which is >of course very important b) using Rebel you can turn off all learning to avoid >any unwanted side effects for the experiment. > >The only disadvantage I see is that most probably the percentage of the >predicted moves will be a bit higher than the average you get playing ALL >other programs. I would expect it to be quite a bit higher. Note that Rebel (or any other program) will have a different >percentage of predicted moves against another opponent. > >Rebel may predict 40% against Crafty, 75% against Hiarcs and 20% against >Fritz or vice versa. Who knows? Thus picking another program as a base is >flawed also. > >Again: this is only about the ELO of the PB valid for Rebel. The difference >in ELO might fluctuate for each other program. > Right, and this fluctuation is exactly what sparked the original debate: Do single-computer (ponder=off) Winboard matches favor some engines more than others. I think there may be some small variations between engines, but not more than 10-15 ELO points... I'm afraid your experiment doesn't address this question. --Peter >Ed
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.