Author: Robert Hyatt
Date: 14:45:26 10/31/01
Go up one level in this thread
On October 31, 2001 at 16:01:40, Uri Blass wrote: >On October 31, 2001 at 11:13:09, Robert Hyatt wrote: > >>On October 31, 2001 at 09:56:02, Miguel A. Ballicora wrote: >> >>>On October 30, 2001 at 23:39:33, Robert Hyatt wrote: >>> >>>>On October 30, 2001 at 16:12:20, Miguel A. Ballicora wrote: >>>> >>>>>On October 30, 2001 at 15:54:06, Robert Hyatt wrote: >>>>> >>>>>>The point here is that KRBP vs KRPP might seem to be the same as a KRB vs KRP >>>>>>ending. The KRB has no chance of winning in a real game. There may be a >>>>>>contrived position where the KRB side wins, but I haven't seen any in real >>>>>>games yet myself. I don't want _my_ program to trade from a possiblly >>>>>>winnable KRBP vs KRPP, to a absolutely unwinnable KRB vs KRP ending, if I can >>>>>>help it. >>>>> >>>>>This is true against a computer with EGTB's, against humans it is not exactly >>>>>unwinnable. Particularly with "game in x" time control where it is closer to a >>>>>win than a draw. Practical chances are very high, it is a tough endgame to >>>>>defend unless you have plenty of time and know the theory. >>>>> >>>>>Regards, >>>>>miguel >>>> >>>> >>>>It isn't that hard for a computer. I ran a few such tests a few years ago. >>>> >>>>The first was KQ vs KR, with crafty playing KQ with no tablebases, against >>>>Crafty with KR using tablebases. Crafty with KQ won _every_ time within 50 >>>>moves, with no special heuristics at all other than "drive the king to the >>>>edge and then the corner." In fact, on a P6/200, it could win every time >>>>given only a couple of seconds. >>>> >>>>When Steven Edwards made me the KRP vs KR tablebase, along with the promotion >>>>cases, I tried this again after watching Crafty play a KRB vs KR ending and >>>>being unable to win it (I had never noticed that this is generally drawn). I >>>>tried tablebase KRB vs no-tablebase KR and the no-tablebase side had no >>>>problems in drawing every game. With very shallow searches. >>>> >>>>I _have_ seen programs lose drawn endings. Crafty and WchessX once played a >>>>KR vs KN where Crafty had the KR (no tablebases back then) and it still won the >>>>game. However, I would not expect to repeat that against _any_ human or >>>>computer I would really expect to have to play. Maybe against a 1500 it might >>>>win. But not against a strong player. And I am generally thinking of IM/GM >>>>players as the competition I play against, which simplifies things a bit. >>> >>>The last point is what you might want to reconsider. Maybe I was not clear >>>before, this endgame is tough for IM and GMs that do not know the theory (and >>>there are many). Tough even at slow time control. If they do not know the >>>theory, and they play blitz, Crafty will score >95%. If they know the theory and >>>it is blitz crafty will still score a lot of points since is very easy to make a >>>mistake on the crucial positions. That's all you need. >>> >>>I personally was able to hold this position against an IM many years ago with 3 >>>minutes on my clock only because I knew the theory perfectly and I had it >>>_fresh_ on my mind. He was very dissapointed not to bring the point home. >>>If you are a IM/GM and you knew the theory but haven't touched it in a long time >>>you might lose precious time before you remember. Particularly at speed chess! >>> >>>If you do not believe me, I will try to find a Nunn's quote about this. >>> >>>Miguel >> >> >>I will try to find a few games with this position in Crafty's huge PGN archive. >>I know it has played two GMs and reached this position and they drew it with >>no difficulty. I'm sure some can't, but the issue is that I do _not_ want to >>reach KRB vs KR or KRN vs KR against _anybody_ because it _should_ be >>drawn. If I can beat 'em with a KRB vs KR ending, then I can almost certainly >>beat them with a KRBP vs KRP ending. > >1)The problem is not this case. >The problem is that Crafty may trade a bishop for a pawn and get KR vs KR and >not KRB vs KRP > >2)if Crafty evaluate everything in KRB vs KP as a constant evaluation of 0.00 or >even 0.01 for the side with more material then it is not going to try to improve >it's position and it may help humans to draw KRP vs KRB with no difficulty > >3)Crafty can avoid trading pawns with evaluation that is not a constant >evaluation and this subject was discussed here in previous posts. > >Uri I understood the previous discussion... I just didn't agree. I don't want it to think that KRB vs KRP is better for the KRB side. It is a draw. Plain and simple. Of course, I also think KR vs KR is a draw, even though one side _can_ blunder away the rook and lose. But I try to avoid doing that. For lots of reasons. But the main reason is "politeness". If Crafty's search returns a draw score, then Crafty will, after a few of those draw-score searches, offer a draw. Which tends to make humans happy. I don't want to drag out a KRB vs KR hoping the human will make an error. I'd rather call it a draw, and move on to the next game. Trying, all the time, to not reach a KRB vs KR position in the first place... But for me, getting the evaluation to <draw> is an important consideration. For those that never offer draws, and for those that won't accept draws, it doesn't matter. But I do both. And I want to _try_ to do them both at the right times... That requires that my search keep me informed about the game-theoretic value, rather than some "hoped-for mythical advantage".
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.