Author: Brian Richardson
Date: 08:01:22 10/01/00
Go up one level in this thread
On October 01, 2000 at 06:38:48, Uri Blass wrote: >On October 01, 2000 at 05:14:42, Mike Adams wrote: > >>On October 01, 2000 at 04:48:41, Uri Blass wrote: >> >>>On October 01, 2000 at 04:20:46, Mike Adams wrote: >>> >>>> I've noticed that pulsar thinks the doubled pawns are worth more than the >>>>danger of exposing its castle. When its white and castles kingside this is >>>>particular dangerous when it opens the H file especially if the rook is sitting >>>>right there on H8. I'm not as concerned about the F and G files because I think >>>>it can better defend these files but it can be very difficult to moblize the >>>>pieces to defend the H file. Should i put a penalty in at 1.5 times the doubled >>>>pawn bonus? >>> >>>Chess is not a simple game. >>>There are cases when open h file is not dangerous inspite of castling in the >>>king side and there are cases when open f file is dangerous. >>> >>>I think that it is impossible to answer your question because the right panelty >>>for open file should be dependent in the rest of the evaluation. >>> >>>Uri >> >>yes but take a look at this position. pulsar is black and it plays Nxg3. >> >>[D] r1bq1rk1/ppp2p2/2n4p/3pP1pn/3P4/2P2NB1/P1P1B1PP/R2QK2R b KQ - 0 14 >> >> >>I think there are some general rules that you dont open the H file with the >>opponents rook on H8. Pulsar does have a king saftey function that tells it not >>to advance its castle side pawns but in some positions it seems to like doing >>that. Maybe i'm not doing it right i gave it penalties similar to what the >>simple chess program does for advancing castled pawns but sometimes when it >>advances one pawn I think it gets easier for it to advance another if it thinks >>it messes with the opponents postion. That is another area i'll have to look >>at. With a check extension and deeper search pulsar, now over 2250 icc blitz, >>can get itself out of a lot of diffictulties but some things are just stupid and >>in this case i need to build into the evaluate that NxG3 followed by G4 are >>really bad. The question is how to do that in a narrow sense that does not give >>it rules that are bad half the time. >>The move sequence of the game was: .. NxG3 HxG3 G4 Qg2 Gxf3 Qxh6 and of course >>its over. I assume you meant Qd2 (not g2) Getting it to see Gxf3 is bad is not easy to do. the material value >>of the Knight distorts the search and particulary with nullmove its hard to see >>at shallower depths like 5 or 6 even with check extensions. I'm trying the >>positon at Gxf3 on different versions of pulsar to see how much depth it takes >>to see that. So far with NUll move off it does the best and sees it at depth 6. I use the following penalties in Tinker (which may be somewhat extreme): KING_AH_NOPAWN_PENALTY 240 // Potential Trojan attack when king // castled and no A/H pawns. // asymmetric penalty must be enough // to postpone P x (N or B) KING_AH_ATTACK_PENALTY 120 // worse when opponent has 2xR + Q KING_AH_OPEN_PENALTY 480 // worse when file completely open It is asymetric in that these are the penalties for Tinker, but I divide by 4 when Tinker is the attacker. The "attack" and "open" penalties are cummulative with the nopawn penalty. They are only added when the opponent has a rook or queen on the open A/H file. The completely open is when the opponent has no pawns on the A/H file either. Tinker finds gxf3 to be bad (-420) at 6 ply (when forced, it is mate in 5 afterwards). >>A freind said he tried it with crafty and crafty had a hard time seeing that >>Gxf3 is bad but I did not get to see that for myself. > >Crafty17.11 on a slow pentium200 has no problem to see in 3 seconds that gxf3 is >losing. > >Crafty17.11 has also no problem to see after 68 seconds with the same hardware >that the position after g4 Qd2 is bad for black(the score is 4.40 pawns for >white at depth 9 for the move h5. > >The tactical mistake is g4 and there is no reason to push the pawns forward. > >I do not think that Nxg3 is a good idea but after Nxg3 hxg3 Kg7 I do not see a >clear win for white. > >Uri
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.