Author: Ed Schröder
Date: 09:55:29 10/02/04
Go up one level in this thread
On October 02, 2004 at 11:10:51, David Dahlem wrote: >On October 02, 2004 at 04:59:04, Ed Schröder wrote: > >>On October 01, 2004 at 12:49:21, ThatsIt wrote: >> >>>Hi ! >>> >>>The last few weeks i have made some tests with ProDeo. >>>First with the defaults (Sep, 04) after them with my own creations. >>> >>>The results until now: >>>[played with 40moves/20min.+ 40moves/20min. + ... / 64MB HTs / >>> without any TBs / without pb / 20 startpositions = 40 games per match >>> on a celeron 2ghz] >>> >>>Pro Deo 1.0 (default) vs Shredder 8 09.0 - 31.0 (+ 4 =10 -26) >>>Pro Deo 1.0 (GS2) vs Shredder 8 16.5 - 23.5 (+ 9 =15 -16) >>>Pro Deo 1.0 (GS2a) vs Shredder 8 20.0 - 20.0 (+14 =12 -14) >>> >>>Pro Deo 1.0 (default) vs Junior 8 17.0 - 23.0 (+11 =12 -17) >>>Pro Deo 1.0 (GS2) vs Junior 8 17.0 - 23.0 (+11 =12 -17) >>>Pro Deo 1.0 (GS2a) vs Junior 8 still running >>> >>>Pro Deo 1.0 (default) vs Hiarcs 9 13.5 - 26.5 (+ 7 =13 -20) >>>Pro Deo 1.0 (GS2) vs Hiarcs 9 15.0 - 25.0 (+ 8 =14 -18) >>>Pro Deo 1.0 (GS2a) vs Hiarcs 9 follow soon >>> >>>Pro Deo 1.0 (default) vs Fritz 8.0.0.26 14.5 - 25.5 (+ 5 =19 -16) >>>Pro Deo 1.0 (GS2) vs Fritz 8.0.0.26 follow soon >>>Pro Deo 1.0 (GS2a) vs Fritz 8.0.0.26 follow soon >>> >>>Total: >>>Pro Deo 1.0 (default) 54.0 aus 160 (33.75%) >>>Pro Dep 1.0 (GS2) 48.5 aus 120 (40.41%) >>>Pro Deo 1.0 (GS2a) 20.0 aus 40 (50.00%) >>> >>>Changes in comparison to the default (Sep, 2004) >>>GS2: >>>[Bishop Value = 102] [Bishop Pair = 102] [Mobility = 120] >>> >>>GS2a: >>>this will be told after the games. >>> >>>The games, the startpositions, more details and other stuff can be found under: >>>http://www.pcschach.de >>> >>>Best >>>Thats It >>>(Gerhard Sonnabend / Stuttgart / Germany) >> >> >>Hi Gerhard, >> >>Nice work. Just one remark (question), I do hope you have turned off "position >>learning". In this type of testing (running matches from the same positions) >>when position learning is on the learning algorithm may have played a dominant >>role and you perhaps are wasting your valuable time because you are actually >>testing the learning mainly. This because I have noticed that PD's position >>learning algorithm is really good. >> >>If not then the results look indeed promising and encouraging. >> >>Keep up the good work. >> >>My best, >> >>Ed > >Hi Ed > >I realize this may not be possible, but a good idea for a future version of >ProDeo would be for each personality to have it's own learning files. :-) > >Regards >Dave Hi Dave, That would be perfect, but imagine the disk pollution. But now that you made me think about it what looks acceptable is something like this: [Position Learner = on] * off|on [Position Learner file = teach_me.lrn] * --> new <-- or [Personality = whoops.eng] * just kidding [Position Learner = on] * off|on [Position Learner file = on] * off|on In this case when [Position Learner file = on] is set learned data will be stored in the file whoops.lrn I will think about it. My best, 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.