Author: Thorsten Czub
Date: 04:29:04 09/07/03
Go up one level in this thread
On September 07, 2003 at 06:19:56, Eelco de Groot wrote: >That's good news Ed! > >I saw that the Beta-testers have a few new options to test with Rebel...and >Thorsten having switched off about all the proven improvements again in his >settings. the question is for me: what is PROVEN. if you call something proven when you beat your older style in eng-eng matches, i would say: NOTHING is proven. i prefer let rebel play versus fritz, shredder, tiger and see how the scores of the matches are. if the NEW and "proven" styles fail to get good scores, i call the new switches "rubbish" and do not use them. when i have a switch that increases scores, i put it in. it is just a question of trying switches in games. this method takes more time than eng-eng games. also you need 2 pc's. but IMO in the end you get a better proof because you tried the ideas and switches out against your opponents. it is unimportant how rebel12 (new style) played against rebel12 (old style) IMO. the people will and want to know how Rebel (new style) plays against fritz and co. thats the way it works. > No pardon, in spite of all your admonishments not to change anything >still there... It was very comical to see...and where will he end up? :-))) the switches that "WORK" are used. the switches that do not produce better results are thrown out. thats sometimes difficult to stand, when you thought you have a new idea and you got good results against your older engines, but you must IMO be very carefully with incest-statistics. only if the opponents are a good mix, you can conclude... >Still struggling with the Singular Extensions. Do you think Thorsten is right to >put [PRUNING = SEX_HISTORY_DEPTH_9] in? > >Groeten, >Eelco i found that the SEX helps rebel to kill the opponent tactically. so i put it in. what made rebel play lousy were the many tricks ed implemented to beat old-rebel versions, all the reductions and selective things. those stuff worked against older rebel versions but not against fritz and co. the evaluations of rebel are ok. also the normal search is IMO ok and strong. it only took some fine -tuning. this has happened. now we collect all the new experience and can make a stronger rebel. ed has done something that is remarkable. he was open for changes ! normally a programmer fights like hell to defend his latest algorithms. ed was very pragmatical concerning this. this reminds me on Emanuel Lasker.
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.