The error for correct and incorrect could be different, maybe to preserve normalization?.Is it a good idea to set error directly instead of using goal predictions?.Maybe I should also update the options that weren't chosen (in the opposite direction)?.Bad move? -> chosen step gets negative error (so prediction goes down)īut I haven't found literature supporting this guess.

Good move? -> chosen action gets positive error (so prediction goes up).How should I train this? My current thought is like this: If the chosen move is bad, I don't know which of the other ones was better (if any). However, the training data I have is just the state, the move that was made, and whether that had good or bad results. The output vector is normalized (softmax layer). I want to train a neural network that chooses a move, so I have 4 neurons in output layer and I make the move with the highest prediction. Say I have a game like 2048 where the possible actions each step are fixed (and more than two). I'm having some trouble describing in one line what I want, which is probably why I haven't had much luck with Google. Mark Asks: Train a classifier for a game with feedback on chosen move instead of true labels

Install endnote on word 2016