About
Open source software, shogi and my life outside of workplace.
Yoshiki Hayashi
Subscribe
RSS
Flavours
| Sun | Mon | Tue | Wed | Thu | Fri | Sat |
|---|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | |||
| 5 | 6 | 7 | 8 | 9 | 10 | 11 |
| 12 | 13 | 14 | 15 | 16 | 17 | 18 |
| 19 | 20 | 21 | 22 | 23 | 24 | 25 |
| 26 | 27 | 28 | 29 | 30 |
Categories
Sun, 12 Jun 2005
Realization Probability cont'd
This weekend ended up researching period for me. While I was searching about
Shogi game tree search using realiziation probability,
I found a comment about it.
It points out the same problem we have seen in GPS Shogi, saying that certain important moves in search process like check does not necessarily occur frequently in master games.
Original paper mentions that the method described there is the best one they could come up with so they are aware of the shortcoming of the frequency based method. Move categorization and frequency based probability is only an approximation of real realization probability which is a probability a move will be played in certain board position. I hope we can find a better approximation than the current method.