Author: Robert Hyatt
Date: 16:46:01 08/17/98
Go up one level in this thread
On August 17, 1998 at 17:26:13, Don Dailey wrote: >On August 16, 1998 at 21:09:03, fca wrote: > >>On August 16, 1998 at 20:47:57, Don Dailey wrote: >> >>>On August 16, 1998 at 16:17:51, fca wrote: >>> >>>>On August 16, 1998 at 10:06:53, Don Dailey wrote: >>>> >>>>>On August 15, 1998 at 22:18:24, Jeff Anderson wrote: >>>>> >>>>>>Can someone perhaps give me a rundown of the piece values used by different >>>>>>chess programs? How do small changes piece values in programs affect their >>>>>>play? >>>>>>Thanks, >>>>>>Jeff >>>> >>>>>Here are some values Larry Kaufman recommended that he felt would make >>>>>most decisions reasonably correct. It is based on 1/3 pawn units, >>>>>which he felt was the lowest unit that can return good values. He >>>>>also considered finer resolution like 1/4 units but thinks the 1/3 >>>>>unit is the best if you want the unit size to be relatively grainy: >>>> >>>>>pawn 3 >>>>>knight 3 >>>>>bishop 10 >>>>>rook 15 >>>>>queen 29 >>>>>Bish Pair 1 >> >>>>Try 9 for knight instead, Jeff, else expect some heavy losses for your program >>>>:-)) >> >>>Try 10 instead, that is what I meant. Of course that makes the following >>>discussion moot, sorry about the typo! >> >>Ha! >> >>With all the conspiracy talk abounding (was here for a millisecond, forever in >>the other place), I am sure this was a conspiracy to waste the time of these >>Moreland, Hyatt and fca guy (oops that is me). We need a moderator to look into >>this troll (oops that is you). >> >>;-)) >> >>So, with N = 10 to start off with, I propose the need for R=15 to be increased >>has itself increased. >> >>Else, from a 2*B start, >> >>B+N vs R+P >> >>= 10 + 10 + 1 vs 15 + 3 >> >>i.e. to give up B & N for R & P is *materially* a whole pawn down, at any stage >>in the game. Clearly problematic! > >I disagree. I think its pretty bad to give up bishop and knight for >a pawn and rook. If you lose the bishop pair too, it's really bad, >giving you a game that is hard to hold, like being a pawn down! I >feel fairly comfortable with these numbers. On the other hand, I >would not feel too bad about giving up 2 knights for a rook and pawn, >although I still think this might be slightly bad. > > >>So in an even endgame (pawns semi-advanced, averagish) with: >> >>KRBBNPPP vs KRBBNPPP >> >>would you (first player) really give up the RPP for BN, leaving you with: >> >>KBBNP vs KRBPPP >> >>I think the effect might be a dramatic game-shortening, *on average* here. >>Throughout the game averaging, I think it would be unwise. >> >>Bob? Bruce? Vishy ( ;-) )? Anyone? >> >>Kind regards >> >>fca >> >>PS: As already realised by everyone, ccertain combi's work better. A vector >>(matrix if you prefer) is needed, not a scalar, to represent material count etc. >>etc. > > >So another master feels that the traditional rook value is already >way too high, 1, 3, 3, 5, 9 He thinks 5 should be 4.5 or 4.6 at >the most. This is not Larry, it's a different master. But when >you bump up the 3's to 3.3 or so you effectively lower the rooks >and make 5 seem more reasonable. But I still think 5 is too high. > >In experiments Larry has done, the bishop pair is so important it >seems to be worth about half a pawn. I don't think he advocates >this value but some game database studies he did showed a very large >value was appropriate for this feature. At any rate, most of >the percieved advantage of the bishop is tied up in the bishop >pair, but probably the bishop is still worth slightly more even >when alone. > >Here is another one for you, would you rather have 2 bishops or >2 knights and an extra pawn? One USCF master (over 2300 strength >for many years) believes this is an even trade. This is supported >by Larry's contention that the bishop pair should be high and that >there should be a knight pair penalty! > >Naturally, all of these situations are highly dependent on the >specific positions involved but it is food for thought. > >WE have experimented with pretty high bishop pair bonus (like 40% >of a pawn) and find some problems however. The main problem is >that your program will choose the bishop pair over almost any >other positional feature, in games we played the decision was >ALWAYS in favor of the pair so bishop takes knight never happens >if you have to give up the pair. So long Ruy Lopez exchange! > >BUT this does not mean this value is not correct. It might just >mean the other evaluation needs a lot of work! > you might need some GM analysis. IE a couple of years ago, Roman called me one morning and said "Bob, I have played hundreds of games against crafty over the past month or so, and I've found one major problem that is hurting it... you apparently have too large a bonus for keeping the bishop pair." I was surprised he even knew what this was all about, but listened, and he went over some games and explained what was going on. I lowered the bonus... and he played some more and said "nope, still too high" as it will allow itself to get into a really ugly position rather than trade a bishop for knight. I kept reducing it until finally "this is right now... it keeps the pair when possible, but won't accept gross positional penalties to do so." The number he liked was .2, or 1/5th of a pawn... and over the next month or two I would bump it up and he would complain. I've left it there ever since... >I also do not believe 2 bishops = 2 knights + 1 pawn but a couple >of strong players (much stronger than me) have told me this was >not unreasonable. > >- Don I don't believe this either. In some cases, yes... but with all pawns on the same side, the two knights can actually be better, since they can gang up on a single weak pawn while two bishops can't...
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.