http://erikbern.com/2014/11/29/deep-learning-for-chess/
"I downloaded 100M games from FICS Games Database and began training a machine learning model. My function is learned from data by using two principles
Players will choose an optimal or near-optimal move. This means that for two position in succession observed in the game, we will have .
For the same reason above, going from , not to , but to a random position , we must have because the random position is better for the next player and worse for the player that made the move.
The model
We construct as a 3 layer deep 2048 units wide artificial neural network, with rectified linear units in each layer. The input is a 8 * 8 * 12 = 768 wide layer which indicates whether each piece (there are 12 types) is present in each square (there are 8 * 8 squares). After three matrix multiplications (each followed by a nonlinearity), there’s a final dot product with a 2048-wide vector to condense it down to a single value."
"I’m pretty curious to see if this could fare well for Go or other games where AI’s still don’t play well. Either way, the conclusions above come with a million caveats. The biggest one is obviously that I haven’t challenged a “real” chess engine. I’m not sure if I have the time to start hacking on chess engines, but if anyone is interested, I’ve put all the source code up on Github. "
https://github.com/erikbern/deep-pink
help....w32 and w64 compiles!! Thx......AR
