View Single Post
  #7   Report Post  
Old April 7th 04, 10:57 AM
Mikko Nummelin
Posts: n/a
Default Using stonewall against computers.

On Wed, 7 Apr 2004, David Richerby wrote:

Not thirteen seconds per move: thirteen seconds for the rest of the
game. This is a five-minute blitz game. The winning strategy is to get
into a solid, closed, drawn position with all the pawns on the board
and then move one's king from side to side for 200 moves until the
computer loses on time. Because Juha's moves in that period took no
thought, pondering wasn't really an issue.

I must still say that if beating Shredder 8 is that easy, the engine
cannot be considered as top-level. One important aspect of a good chess
engine is that it can handle time-controls appropriately. Losing on time
to humans is not good performance from a good chess engine.

If a computer program manages to do 500knps and does good job on move
ordering, it should achieve at least _full width_ ply levels following
this table:

1 ply in 70 microseconds
2 plies in 490 microseconds
3 plies in 3.4 milliseconds
4 plies in 24 milliseconds
5 plies in 170 milliseconds
6 plies in 1.2 seconds
7 plies in 8.2 seconds
8 plies in 57 seconds or say, 1 minute

In this table there are _no search extensions_ nor _null move pruning_
assumed! Those lead sometimes 3-5 plies deeper in principal variation.
This certainly means that top-level chess engines can both play good chess
and avoid time trouble in blitz when using their strongest settings.

Conclusion: Juha's Shredder 8 must be badly configured, corrupt or some
other heavy processes interfere too much with it causing failures in
respecting time controls.

Mikko Nummelin