Hey everybody,

I'm making a connect 4 game and I want to make it stronger by using an opening book. I've looked all over, and the only one I can find is by John Tromp.

http://homepages.cwi.nl/~tromp/c4/c4.html

I thought the book would be a list of moves that I could read into a tree. However, it consists of positions at 8-ply and whether they eventually lead to a win, loss, or draw for the first player. My question is how do I incorporate this into my program?

Any help is appreciated. Thanks.

You should be able to make a Connect Four AI that is sufficiently effective without any use of opening books. Unlike say, chess, there are very few openings and no considerations to be made for openings that might lead to bad positions in a dozen moves or more, other than your program should definitely play a central stone on its first turn (if it goes first). That is assuming you're doing a standard 7 x 6 board.

You should be able to make a Connect Four AI that is sufficiently effective without any use of opening books. Unlike say, chess, there are very few openings and no considerations to be made for openings that might lead to bad positions in a dozen moves or more, other than your program should definitely play a central stone on its first turn (if it goes first). That is assuming you're doing a standard 7 x 6 board.

Thanks for your reply Alvaro.

So are you saying that for the first 8 moves, if n moves have been played, then i do minimax with a depth of (8-n) and use the database values instead of my evaluation function? And then after 8 moves have been played, then I switch back to my evaluation function and increase depth back to normal?

Also I noticed that the database contains 67557 positions at depth 8, while my perft function returns around 180,000 positions at depth 8. How should I evaluate positions that are not in the database? Should I give them a value so that they won't be considered over the other nodes?

Thanks.

I'm not saying that John is wrong. On his website he says that he only includes positions where neither player has won yet, and in which the next move isn't forced. I believe this is why there are less positions.

I did take into account transposition and made sure not to count the same position twice. Also my numbers agree with this numberphile video:

If I run into one of these nodes not in the database, should I set its value to + infinity so that the minimizer at depth 7 will cut them off and not look at them?

Thanks for your reply. I ran the Fhourstones benchmark with command prompt and I got a printout similar to what is listed on this site:

http://homepages.cwi.nl/~tromp/c4/fhour.html

How can I use the benchmark to generate a database with values for all of the 8-ply positions? Is there a command that I can use to output a text file of all 8-ply positions? I tried looking at the java source code but I don't understand it.