This is inspired from my previous question, comments to which made me realize that a CNN was unsuitable for the problem The CNN required over 700k training datasets while a MLP did it in in less than 50k.
Now, I'm trying to solve the next problem and need to figure out if a CNN makes sense.
CNN detail -
Input - The board as an array of 9 elements that represents the board (0=empty, 1='X', 2='O')
Output - Recommended move as a one-hot encoded array of 9 elements. The index of 1 is the recommended move (for example, in [0,0,1,0,0,0,0,0,0] the recommended move is 2)
So, basically the CNN will be trained with a dataset that consists of boards and the move that the winner of the game has made for each of the boards. Then during evaluation, it'll try to predict the best move for a given board.
Does a convolution neural network make sense for this problem?
Note: The convnet that I was going to use for this problem is the same as in my previous question