Author: GeoffW
Date: 08:36:37 07/28/04
Hello
Continuing the chat about IID from the Winboard forum
Thanks to Dan Corbit and Fabien for pointing out the errors in my attempt at
IID. It's tricky to get things right first time now without having the CCC
Archive to read through :-(
Anyway I have made a slight tweak to my implementation to make it match the
Beowulf logic you posted. This now gives me an improvemnt of varying degress in
almost all test positions. My first attempt was better on some but worse on at
least as many positions
As you were tinkering with IID as well here are a few test results for my
program before and after.
I am guessing these figures are pretty much what would be expected with IID. The
positions where it gave a big improvement are probably due to IID hiding some
nasty weakness or bug in my code !
Regards Geoff
All Tests with 64Meg Hash Table
2k1r2r/ppn3pq/b7/5p2/5Qn1/2N1B1P1/1P3PB1/2RR2K1 w - -
without IID D11 47.48 Seconds
with IID D11 45.50 Seconds = 96% of original time
3r2k1/1pp2ppp/3r4/1P1npq2/3n4/3PPP2/1BQ1BP1P/2RR3K w - - 1 1
without IID D12 94.45 Seconds
with IID D12 92.23 Seconds = 98% of original time
r2qkb2/5p1p/p1npb1r1/1p1Np2Q/4Pp2/N2B4/PPP2PPP/R4R1K w q - 0 15
without IID D11 96.15 Seconds
with IID D11 92.45 Seconds = 98% of original time
6kr/1b2bp1p/p6Q/Pp2p3/1q3p2/4B1N1/2PR2PP/6K1 w - - 0 33
without IID D12 88.14 Seconds
with IID D12 44.48 Seconds = 50% of original time !! Wow !!
This one seems to improve the screwy fail high to fail low scores I was
originally getting on consecutive iterations
1r1q2k1/pp3p2/2p3p1/n2p4/3P1NQp/2P5/P1P2PPP/4R1K1 b - -
without IID D12 54.56 Seconds
with IID D12 51.96 Seconds = 95% of original time
1rb2rk1/p4ppp/1p1qp1n1/3n2N1/2pP4/2P3P1/PPQ2PBP/R1B1R1K1 w - - 0 1
without IID D12 230.06 Seconds
with IID D12 230.78 Seconds = 100% of original time
3r2k1/pb3qbp/1pnrp3/1B3p2/Q2P4/4B3/P3NPPP/1R1R2K1 b - - 3 23
without IID D11 249.14 Seconds
with IID D11 168.12 Seconds = 67% of original time
r3k2r/p1ppqpb1/bn2pnp1/3PN3/1p2P3/2N2Q1p/PPPBBPPP/R3K2R w KQkq - 0 1
without IID D10 159.34 Seconds
with IID D10 152.95 Seconds = 96% of original time
kr6/2K5/p7/P7/6PQ/8/8/8 w - - 0 1
This was am interesting one
without IID
13 1313 812 10222353 h4g3
14+ 1358 964 11715903 h4g3
14+ 1658 1025 12232965 h4g3
14 9981 7114 133063205 h4h1 a8a7 h1h2 b8b7 c7c8 b7b5 h2c7 a7a8 c7c6 a8a7
c8c7 b5b8 c6b6 b8b6 a5b6 a7a8 b6b7 a8a7 b7b8q
with IID
14 2037 1323 18828582 h4f6
15+ 2082 1376 19255645 h4f6
15+ 2382 1404 19497011 h4f6
15 9983 3253 42150409 h4h1 a8a7 h1h2 b8b7 c7c8 b7b3 h2c7 a7a8 c7c6 a8a7
c6c5 a7a8 c5d5 a8a7 d5b3 a7a8 b3b8
IID made it see a mate in 9 instead of mate in 10 but took 1 ply more to see it.
The main point being though it found a mate in 46% of the time
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.