Author: jonathan Baxter
Date: 14:50:09 03/22/99
Go up one level in this thread
On March 22, 1999 at 11:38:37, Will Singleton wrote: >Under 2600 Club >Mar 22 1999 > >Program ICC Blitz Change Games Platform Author > >TDChess 2679 +102 216 P/400 Jon Baxter >ZChess 2576 -1 5 K6/266 Franck Zibi >PostModernist 2560 +20 44 K6/300 Andrew Williams >Rajah 2531 +50 76 P/166 V. Manohararajah >Rookie 2514 P/360 Marcel van Kervinck >LambChop 2453 -21 28 P/133 Pete McKenzie >Amateur 2450 +58 67 Mac/300 Will Singleton >EXchess 2446 +114 21 Cel/400 Dan Homan >BugChess 2405 P/450 Erik van het Hof >Grok 2353 -55 21 K6/300 Peter Kappler >Hossa 2351 -205 287 P/350 Steffen Jakob >InmiChess 2292 +14 8 Cyrix/233 Werner Inmann >Pedestrian 2195 P/450 Peter Kasinski > > > ICC account names that differ from the program name: > ---------------------------------------------------------------------------- > Blik (Rookie), zchess1 (Zchess). > ---------------------------------------------------------------------------- > >Changes > >Stobor and XXXX-2 to inactive, EXchess to active. > > >Notable > >New highs were scored by TDChess (2711) and PostModernist (2603). > >Top spot goes to TDChess, which currently sits at 2679 (as of this morning), >and may very well be our first graduate. Excellent job, Jon! Its worth pointing out that I have not touched the code except in trivial ways for a long while (last major change was a few weeks ago where a "very nifty idea" for detecting trapped pieces cheaply was tried and discarded because it didn't work very well). So nearly all of the improvement in TDChess can be attributed to the book learning (it's not learning its evaluation function at present: that's constant). TDchess has no hand-coded openings---they are all learnt. Some may have already noticed that TDChess is wedded to some rather dubious looking variations of the center-counter defense as black (there are those who consider "dubious" too mild an adjective :). This is probably because it plays a lot against various versions of Crafty and Crafty can't seem to find really strong moves for white in these positions, so TDChess actually wins a lot of these games. However, over the last one or two days, TDChess has started to change its mind and currently prefers 1... e6 in response to 1. e4. Only took nearly 30,000 learning positions to be added to the brain for this to happen! I suspect a serious of standard games where it searched a lot deeper swayed the balance. However, the nice thing is that if TDChess does finally switch away for good from 1... d5 then it will know exactly why it is doing it. If you are interested, TDChess now lists the longest continuous brain variation in its finger notes, the total number of brain positions and the brain position it is currently "mulling" (researching). Cheers, Jon >Most active was Hossa, with 287 blitz games played. > >I notice that several members (active and inactive) will be competing at >Paderborn. They are LambChop, Stobor, XXXX-2, and Gromit. One would >hope that McKenzie will bring something faster than a P133. :) Good luck >to all! > > >u2600 Position of the Week > >This comes from a standard game played on ICC this week between TDChess and >crafty450. TDChess found a good move, which I don't believe would have >been found at blitz time controls. A good human player might be tempted >to play it right off the bat, but a program needs proof. > >See how long it takes yours to find the winning move. Also, see if you can >guess the move before your program finds it. > >2r4k/3q1pp1/1rn4p/p1R5/P2p1B2/5QP1/P4P1P/2R3K1 w - -
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.