Ah! This is so beautiful. It plays good backgammon!
However, the most important feature must be to add probabilities for gammon and backgammon. In other backgammon software we usually represent the wins, gammon and backgammons in a vector of five elements. (There is six possible outcomes of a game, but each of their probabilities must add up to 1. A vector of five should therefor do fine.)
0: Probability of winning (any flavor)
1: Probability of winning gammon or backgammon
2: probability of willing gammon
3: probability of losing gammon or backgammon
4: probability of losing backgammon
A neural network with 5 outputs (sigmoid activation) should therefore be able to estimate the gammon and backgammon probabilities.
Glad you got it working! Unfortunately I don't have any plans to start actively working on this project again. So for now it will have to be a first-to-1-game bot.
Ah! This is so beautiful. It plays good backgammon!
However, the most important feature must be to add probabilities for gammon and backgammon. In other backgammon software we usually represent the wins, gammon and backgammons in a vector of five elements. (There is six possible outcomes of a game, but each of their probabilities must add up to 1. A vector of five should therefor do fine.)
0: Probability of winning (any flavor) 1: Probability of winning gammon or backgammon 2: probability of willing gammon 3: probability of losing gammon or backgammon 4: probability of losing backgammon
A neural network with 5 outputs (sigmoid activation) should therefore be able to estimate the gammon and backgammon probabilities.