Author: Ed Schröder
Date: 10:27:29 10/20/04
Go up one level in this thread
On October 20, 2004 at 09:46:26, Albert Silver wrote: >On October 20, 2004 at 07:15:44, Ed Schröder wrote: > >>On October 20, 2004 at 04:51:42, Dom Leste wrote: >> >>>Thank you for replying to my last question, i forgot theses settings also. >> >>>What differences in number does Selective search have? >> >>If you lower the default value (100) it will ensure a more safe search with less >>holes, its disadvantage is that the search becomes slower. The opposite will >>happen with values above 100, the search becomes faster, the risk of holes >>(missing essential moves) becomes bigger. >> >>Reasonable values are in the range of 60-200. Best setting is 100 (default). >> >> >>>Is their a command for good/bad knights & bishops? >> >>Actually there are two tuning possibilities. >> >>[Bad Bishop = 100] * default is 100 >>[Pruning = MISC_59] * double eval when in the endgame. >> >>Both are totally untested, hence its undocumented status. > >Would it be possible to know the other undocumented parameters and possibly a >minor description and evaluation such as has been done with some in the Pro Deo >parameters pages? I ask because I now have a *very* good idea as to the >difficulty of testing so many things, such as parameters that work well alone or >with another, but combined with yet a 3rd and/or 4th, suddenly are worse, etc. >Not to mention the colossal difficulty in going through them, testing them, and >the huge number of combinations. At least this way, we could break our heads >experimenting with a few. I understand the problem. I will see what I can do. I will work on a new public version together with an updated list of the parameters. >For example, I have had some luck with some of them: > >[Lazyeval MIDG = moderate] * has genuinely helped as far as I can tell Could be yes. >[Pruning = MISC_07] Untested yet, it's an anti-zugzwang algorithm for the endgame. It does a good job in zugzwang positions but slows down the search a lot. I don't expect much in practical endgame play. >[Pruning = MISC_17] >[Pawn Pressure MIDG etc.] > >and especially > >[Pruning = MISC_26] * Endgame extensions >[Pruning = MISC_44] * New endgame extension >big changes one way or the other with some others such as > >[Pruning = MISC_22] * Slightly safer Selectivity It's function has changed in the meantime. Drop it. >and to a lesser degree > >[Pruning = MISC_21] * Lazy Eval (tuning) > >which combined with others seems to make a difference. > >I've also rejected [MISC_58] despite good results as strange as that may seem. I >just don't like the evaluation oddities it causes, and it makes me very >uncomfortable, even though it seems to really improve the tactics in many >things, and probably other moves as well. > >And I've tested and rejected > >[MISC_09] >[MISC_10] >[MISC_46] Good. >[MISC_39] I use [MISC_03] but can't get this one to work for me with my >other parameters. You cab safely use both. >and more especially the new style [King_Safety]. Oddly enough, it seems to make >the King Safety concept weaker. For example, take one of the positions from the >WM Test: Don't use [King_Safety] indeed, stick to the old one. >[D]r1b1rk2/2qp1pp1/2p2n1p/p3pP2/1bB1P3/2N3R1/PPPBQ1PP/R6K w - - 0 1 > >With the old style King Safety, we see Bxh6 as one of the candidates, which is >quite understandable. The correct move is Qe3 of course. If I put the new style, >it completely ignores the kingside, per the mainlines it displays. Even if I >jack up the value to over 170, so I don't use it. You shouldn't put this position in the positional king_attack category because it needs white to sac 2 pawns for a rook. No problem for search, but a problem for eval. >Finally, I appreciate the problem with the [Attacking] parameter and why you >stated it weakens the engine at higher values. The [Attacking] algorithm looks for complications on the board. The effect is that it will ignore small positional issues and the higher you set its value the greater the risk the engine will neglect the positional aspects of a position. Ed
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.