Author: Amir Ban
Date: 14:09:46 07/15/98
Go up one level in this thread
On July 15, 1998 at 08:21:50, frank phillips wrote: >Thanks to those who responded. There are some interesting pointers and most of >all encouragement. Tom asked about node-rate. I will do the test he suggested >of how many nodes it takes to a reach a ply. The current node rate is around >40k nps in the starting position, rising to up to 70k nps or so in the endgame: >where a node is a call to the quiescent search function or a call to Negamax >that is not also a call to Quiescent (ie nodes are counted in Negamax only after >the if(depth>=maxdepth) Quiescent(....); bit). > >Don indicated that there are no big leaps just lots of small improvements that >add up. This is useful to know. I obviously have an awful lot of small steps >to take. That's not quite true, for a start. You MUST have your move ordering covered reasonably well, and to do that you should first do killer-moves. Hashing is also a way to handle move-ordering, but if you have killer-moves setup in the right way, it is not so urgent. I would also take a look if your quiescence searches are not diverging. Killer moves are a great help in controlling that, but if you are still diverging, take a close look at your quiescence. > At the moment I bring a local array of structures into existence at >each incarnation of the search functions. It is of fixed size - int MoveT >move[MAX] - which is gross and may be slowing things down. I will move to a >global stack array and pointers next. I would skip this. Local declarations don't make programs slower, and making declarations local is a good programming practice, usually slightly faster than global ones. > You guys seem to have been born >genetically imprinted with the null move, killer heuristic, hash table, >bitboards etc concepts :-) I would skip bitboards in handling your current crisis. Bitboards are a complex project of debatable merit that can be postponed a year or two. Most of us have postponed it for good. > As a complete beginner I have found it difficult >to get hold of material describing these techniques and was looking for >references.- is there a relevant FAQ somewhere? There are few books on computer >chess and those mentioned here seem to be out of print. Perhaps the answer is >back issues of the ICCA journal - but hopefully not them all. (Are they >suitable for only the experienced expert?) . Without the net and the help of >members of CCC I would not have even taken my first step. On the other hand, if you listen too much to what people here have to say, you'll never write anything original. Amir
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.