Author: Georg v. Zimmermann
Date: 11:55:12 10/10/99
Go up one level in this thread
On October 10, 1999 at 13:02:27, Ratko V Tomic wrote: >> My god, you can turn of all learning if you play against the program >> and do want to get different openings. What is your problem ? > >Time. Why every time it gets out of the book I have to wait minutes to >have it go through identical calculation with the identical result >it did earlier? Some people have other work to do, it takes enough >time to play chess as it is, to have to _needlessly_ wait (I don't >mind waiting if it is the first time calculating a position). You play that much games ?? I'm impressed. Maybe make a bigger tree/book not to get the same position again. Or _you_ can not play the same again ?! > >The problem is that the program manufacturers have gotten caught up so much in >getting an edge in the mindless machine-machine autoplay, they forgot the >customer, and thus have designed the "learning" feature which is well suited The customer, us, looks at the SSDF. If we'd stop looking at it, they would design program differently. >quickly locking in into a "killer line" against another program and (due to the >lack of common sense in the testing procedure) winning identical game ten times >against a program which doesn't have this kind of "learning." Well, thats the other programs problem. > >Since human player doesn't fall for (and doesn't appreciate someone even trying >out) such idiotic tricks, Its not a trick. If you play against someone and have sucess with one line, will you play it again or not ? This is just normal. > the feature is not only useless but outright contrary >to the customer's convenience. He can either play the same line ad infinitum >(with "learning" on) or waste time waiting for mindless repetition of the >identical calculation. Either win against it and it will drop the line or improve your understanding of the position, which is certainly necessairy if you keep loosing. > >It would be enough to enable this kind of "learning" in machine-machine autoplay >only, and make it do a common-sensical user oriented learning otherwise, i.e. >all moves, evaluations & thinking times are remembered so it never calculates >the same thing over (user may wish to choose whether, when playing at same level >in the same position next time, s/he wants the earlier computed move played >instantly or deepened, for the duration appropriate to a given level). The game >result should not skew the odds of an opening, so it can keep the variety for >users convenience. The only way the learning should affect future choices is >that if in a previous game the program has obtained negative evaluation at some >point (regardless of the final game result, which may be due to completely >unrelated causes far away from this evaluation), it should back off one >(preferably) or two moves (at most) and pick something else. I agree here, also a all programs should have a smal "random evaluation". > But it certainly >should not drop the whole Sicilian or French, or even a particular line, just >because it lost few games in that (perfectly good) opening/line for completely >unrelated reasons well beyond the opening. maybe true. > >So it's the publicizing of mindless machine-machine autoplay results in >combination with the short term sales-folk cheap gimmick mentality prevailing >among the program manufacters that has resulted in shunting the customer's time >and convenience out of the loop. It will be enough for one leading program to >have a usable customer oriented learning for others to snap out of the loop >caught themselves in. That cycle occurs with other software and other products >all the time, it will happen here. The discussion on the subject will only help >speed up the inevitable. I think you are a bit impolite here. Regards, --Tec.
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.