We need a lot of data to train the neural nets. For that we need a lot of of rollouts. To shorten the time of these rollouts it would be good to have a 2-sided bearoff database.
Goal: For all positions where both players have all their checkers in their home board or already off, we want to look up exact equities.
For each position that doesn't lead to winning immediately, calculate all possible positions for all possible moves that follow from there and look up the already calculated equities for those positions. Calculate the equity for that new position and store in the array.
External contributions are very welcome for this issue. Feel free to contact me for more technical details.
We need a lot of data to train the neural nets. For that we need a lot of of rollouts. To shorten the time of these rollouts it would be good to have a 2-sided bearoff database.
Goal: For all positions where both players have all their checkers in their home board or already off, we want to look up exact equities.
Howto:
External contributions are very welcome for this issue. Feel free to contact me for more technical details.