Author: Brian Richardson
Date: 17:45:52 04/08/00
Go up one level in this thread
On April 08, 2000 at 20:00:04, Tom Kerrigan wrote: >On April 08, 2000 at 18:28:51, Brian Richardson wrote: > >>On April 08, 2000 at 15:10:35, Severi Salminen wrote: >> >>>Hi! >>> >>>>Congratulations on getting your program playing. >>> >>>Thanx! >>> >>>> >>>>Try adding piece square tables to your evaluation function--see TSCP >>>>source for simple examples. This will make it much easier to direct >>>>opening play and piece development. >>> >>> >>>Well, I found the error (it's not so easy with assembler...) and now my program >>>plays 1...e6 and I'm satisfied so far. >>> >>>BTW how fast "should" the program be? Mine calculates now some 40000 positions >>>with Celeron 300Mhz. Is the NPS rating the actual number of positions evaluated >>>per second or is it the total number (even those cut off with A-B)? >>> >>>Severi >> >>NPS is calculated differently in many programs, but I think it is generally the >>total number of nodes visited during the search, but not including evaluations, >>(which would be already counted as leaf nodes). NPS is not very useful as a >>speed measure. However, it is useful to the programmer to compare various >>versions of the same program. A more useful measure of speed would be the time >>to reach certain depths for several test positions (starting, various midgame, >>endgame). You could then compare your program's speed against several others >>that you could download. TSCP would generally be the slowest, and you should >>shoot for about 10x its speed. Crafty would be among the faster ones (but not > >Just to clarify: A good program will find the solution to a given problem ~10x >faster than TSCP. (I don't think any program searches 10x as many NPS as TSCP.) > >-Tom This is why NPS is not a very good measure of speed. Indeed, Crafty does about 277K NPS to search to a depth of 8 plys from the initial position vs about 117K for TSCP (on my PIII 733MHz system). However, Crafty's time to depth is 1.8 sec vs 59 sec for TSCP. So one program can be roughly 30x "faster" measured one way, but less than 3x faster measured another. Of course, TSCP is intended as an easy to understand example (thanks again Tom), and is at least 10x easier to understand than Crafty. Data provided below (slightly modified TSCP bench command to not replace position with bench test position): Tom Kerrigan's Simple Chess Program (TSCP) version 1.5, 2/13/00 Copyright 1997 Tom Kerrigan "help" displays a list of commands. tscp> sd 8 tscp> bench ply nodes score pv 1 21 48 d2d4 2 84 0 d2d4 d7d5 3 800 35 d2d4 d7d5 b1c3 4 4219 5 e2e4 d7d5 f1b5 c8d7 b5d3 5 22461 35 e2e4 e7e5 d2d4 d7d5 g1f3 6 139848 13 e2e4 e7e5 d2d4 e5d4 d1d4 g8f6 7 1159551 30 e2e4 d7d5 e4d5 d8d5 d2d4 d5e4 g1e2 e7e5 8 6974698 18 e2e4 e7e5 g1f3 g8f6 b1c3 b8c6 d2d4 d7d6 Time: 59281 ms Nodes: 6974698 Best time: 59281 ms Nodes per second: 117654 tscp> EPD Kit revision date: 1996.04.21 found computer opening book file [./bookc.bin]. Crafty v17.6 White(1): easy pondering disabled. White(1): book off book file disabled. White(1): noise 1 noise level set to 1. White(1): sd 8 search depth set to 8. White(1): go clearing hash tables time surplus 0.00 time limit 30.00 (3:00) nss depth time score variation (1) 1-> 0.00 0.15 1. e4 2 0.01 -0.19 1. e4 d5 2-> 0.01 -0.19 1. e4 d5 3 0.01 -0.02 1. e4 d5 2. Nc3 3 0.01 -0.01 1. d4 d5 2. Bf4 3-> 0.01 -0.01 1. d4 d5 2. Bf4 4 0.01 -0.19 1. d4 d5 2. Bf4 Bf5 4 0.03 -0.12 1. e4 e5 2. d4 d5 3. dxe5 dxe4 4-> 0.03 -0.12 1. e4 e5 2. d4 d5 3. dxe5 dxe4 5 0.05 -0.06 1. e4 d5 2. exd5 Qxd5 3. Nc3 Qd4 5 0.06 -0.03 1. e3 d5 2. d4 Bf5 3. Nf3 5 0.06 -0.02 1. d4 e6 2. e4 Bb4+ 3. c3 Be7 5-> 0.08 -0.02 1. d4 e6 2. e4 Bb4+ 3. c3 Be7 6 0.09 -0.19 1. d4 d5 2. Bf4 Bf5 3. e3 e6 6 0.15 -0.10 1. e4 e6 2. d4 Nf6 3. Nc3 d5 <HT> 6-> 0.17 -0.10 1. e4 e6 2. d4 Nf6 3. Nc3 d5 <HT> 7 0.31 -0.06 1. e4 e6 2. Nf3 d5 3. exd5 Qxd5 4. Nc3 Qd8 7 0.44 -0.04 1. e3 d5 2. Nf3 Nc6 3. d4 Bf5 4. Bb5 7-> 0.47 -0.04 1. e3 d5 2. Nf3 Nc6 3. d4 Bf5 4. Bb5 8 0.84 -0.19 1. e3 d5 2. Bb5+ Bd7 3. Bxd7+ Nxd7 4. Nc3 e6 5. d3 Bb4 6. Ne2 8 1.44 -0.15 1. e4 e6 2. Nf3 Nf6 3. e5 Ne4 4. d3 Bb4+ 5. Nbd2 8-> 1.80 -0.15 1. e4 e6 2. Nf3 Nf6 3. e5 Ne4 4. d3 Bb4+ 5. Nbd2 time=1.81 cpu=99% mat=0 n=501550 fh=86% nps=277099 ext-> checks=7541 recaps=4080 pawns=48 1rep=326 thrt:0 predicted=0 nodes=501550 evals=226056 endgame tablebase-> probes done=0 successful=0 hashing-> trans/ref=32% pawn=90% used=47% White(1): e4 time used: 1.81 Black(1):
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.