Computer Chess Club Archives


Search

Terms

Messages

Subject: Junior infinite analysis

Author: Gordon Rattray

Date: 13:35:39 09/10/01


Many people are aware that Junior often does its iterative deepening in steps of
greater than one.  For example, once completing depth 12, it may move directly
to a depth of 15, as opposed to going through 13 and 14.

I had noticed that while playing a game, Junior would tend to use smaller steps
if it had less time to think.  This makes sense as too big a jump may involve a
search that would take more time than it could afford.  However, I hadn't
realised that while using "Infinite Analysis" mode, the time control also has an
effect.

For example, I set Junior to play at tournament conditions (40 in 2 hours,
etc.); put it into analysis mode; and then played the following moves:

1) d4  Nf6
2) Bg5  Ne4
3) Bf4  c5
4) dc  Nc6
5) Qd5

Now the key move is 5) ..  f5

This required about 3 min 12 secs to find. (Athlon 1.1 GHz, 128 Mb hash)

However, with the time control set to be "all moves in 1 min" (blitz), the move
f5 is found after 1 min 52 secs.  The main reason is that the first setup jumps
from depth 15 to 17, whereas the second one finds it at depth 16.  Also, both go
on to search at depth 18 and the second one is still quicker to complete.  These
results are specifically for Junior 7, but I believe earlier versions will be
similar.

My main point is: do people take this into account when posting analysis times
for Junior?  Are test suite results affected?  Am I wrong in thinking that it is
worth setting the level to "all in 1 min", even when only being used for
infinite analysis?

When I tried to repeat my experiment, I initially found it hard to get the "1
min 52 s" time again, even with "all in 1 min".  So I switched to "game mode",
played a quick game at this level, then done a "new game" and switched back to
"analysis mode".  This succeeded in getting it to reproduce the "1 min 52" time.

Finally, I haven't had time to try more positions, etc.  I admit that this
leaves my testing a bit narrow to say the least.

Gordon




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.