Author: Heiner Marxen
Date: 19:44:08 08/16/01
Go up one level in this thread
On August 16, 2001 at 21:12:41, leonid wrote: >On August 16, 2001 at 20:03:13, Heiner Marxen wrote: > >>On August 16, 2001 at 19:24:33, leonid wrote: >> >>>On August 16, 2001 at 16:47:03, Heiner Marxen wrote: >>> >>>>On August 16, 2001 at 12:03:51, leonid wrote: >>>> >>>>>Hi! >>>>> >>>>>This position is intended to be solvable by every program, after only minor >>>>>resistance. >>>>> >>>>>[D]1rrbbq2/3k4/1qqpnq2/3B4/2NPR3/1QB1KQ2/QN3PQQ/Q5QR w - - >>>>> >>>>>Please indicate your result. >>>>> >>>>>Thanks, >>>>>Leonid. >>>> >>>>Hello Leonid! >>>> >>>>You should try to find titles for your beautiful creations! >>>>For this one I suggest "The King", for obvious reasons :-) >>> >>>Hi, Heiner! >>> >>>Is my pleasure to see you once again. >>> >>>This last position is in reality Chinese character that signify "Day". I thought >> >>Hey, that is even better than I imagined! Lovely! >> >>>that it is not correct that I use only Latin letters as position form. Actually >>>today I put "Day" instead of "Middle" that is too much deeper in its solutiion. >>>"Middle" will come next. >> >>I'll have to delay my work on it until I'm done with "US" :-( >>Hopefully I will catch up, one day :-) >>I will not miss any of your creations! :-) >> >>>>I've not yet worked on this new creation of yours, since my K7/600 after >>>>60 hours is still busy crunching on your corrected "US". >>>>http://www.icdchess.com/forums/1/message.shtml?183489 >>>>I know, you warned us :-) but I thought that would not apply to Chest. >>>>That there is no mate in 8 needed only 40 minutes with an apparent EBF >>>>below 10, so I expected less than 10 hours. But this one is a hard nut >>>>for Chest. I have done some investigations into the variants Chest does >>>>compute... and I'm not pleased. There are lots of partial solutions >>>>within the tree, and there are quiet white moves involved. The trees >>>>get really huge :-( >>>> >>>>So... that "US" appears to be a good example for a problem where Chest >>>>does not do very well. I will try to use it to improve Chest, but I'm >>>>not yet sure, how exactly that improvement has to look like. >>> >>>Absolutely the same impression I have when I see some solutions that your mate >>>solver find. Tried even to imagin some possible ways to push my solver farther >>>but until now not went to my code. Still the most attractive for me is my dream >>>to start Linux learning and eventually rewrite everything for 64 bits chip. >>>Fascinating!!! >>> >>>Heiner, I tried "US" on few strong program and branching factor went there >>>beyond 100. Maybe this will make you feel better. >> >>Umm, Uhh, 100+, well, yes! Now I feel better indeed! >> >>>>Anyway, many thanks for your creations! They _are_ helpful at least for me. >>> >>>They are helpful for both of us. Even if I was not that ready to restart my >>>writing, because of your presence, I speeded my mate solver twice since you are >>>here. Before I had no reason even to think that speeding mate solver have some >>>sense. >> >>So, a bit of competition is quite refreshing, isn't it? >>E.g. your "selective search" is already on my TODO list. I want to be as >>fast as you are :-) :-) >>Also, I currently work on access to Nalimov EGTBs, in order to not be >>outsmarted by those "normal" programs :-) >>That competition makes me move, also. Fine! > >Heiner, I am asking me on what program all those Nalimove's positions were >calculated? Because if it use some program like we can see around, they miss a >lot in speed when dressing all those mate tables. It could be that dressing >table for 6 and 7 men is very realistic. At Bob's ftp, where you can download those EGTBs, you can also get "tbgen.zip", containing the C++ source for the generator program. (The current release misses the source for the compression program) ftp://ftp.cis.uab.edu/pub/hyatt/TB/ Some 6-piece tables are already calculated. They are quite huge, already, and you need a really large machine to compute them. In general, 7-piece tables are not a good idea with todays technology. Look at a typical 6-piece table, say KQBKRR: its size is 2GB. Multiply by 64, and you have the size of a typical 7-piece table. There are smaller ones, like KNNKNN (2MB), but those are not very interesting. Also, if you want to have _all_ 7-piece tables... that are a lot, already. Do some math yourself, and you see what I mean. And regarding speed: accessing EGTBs from even a fast disk can become a problem speed-wise. If I recall correctly, crafty has some code to reduce limit the access frequency. In the time for 1 disk read you can do a lot of CPU instructions, which you miss, if you wait on that read access. So, EGTBs do not solve all our problems :-) And for your creations they are really useless :-))) Cheers, Heiner
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.