Thursday, June 21, 2007

Game played by chess computer

Played 2007-06-18, against itself.




4 plies (half-moves) deep.

time avarage per move: 0,613s
position analyzed per second, avarage: 30520
branching factor avarage: 11,0546037560704

Branching factor is how many moves it tries out in a given position. The avarage number of available moves during middlegame is something like 35. An unsophisticated search-algorithm (usually "minimax") tries them all out, whereas the better ones (always some version of "alphabeta") cut off lots of branches. How many branches it after pruning will try out depends on some features in its implementation -- it ranges from about 6 to 36. So my avarage branching factor isn't that bad compared to the worst case scenario. However, there is still a huge difference between 11 and 6. If you go 8 half-moves deep with a branching factor of 6, that's 1.7M moves to analyze, whereas 8 moves deep with a branching factor of 11 is 214M...

(It is possible to let it play 5 plies deep within reasonable time... about 7 seconds in avarage per move. Next game to post will be a game against Fritz, set to play at a lower rating, 5 plies deep.)

2 comments:

Blue Devil Knight said...

Good piece coordination required for that checkmate. Very cool.

XY said...

It did OK, especially since it didn't see the mate until it actually played the mating move (because of how it's implemented -- it has to literally take the king in order to see the mate).

But there are also some really funny moves, like moving the king needlessly (as black did). And not castling is bad too, but there is a explanation for that (which will be given in a later post).

 
Locations of visitors to this page