How to adjust neural network to work with 3 layers representation instead of 1 layers?

Question

I am trying to train an agent to play Connect4 game. I found an example of how it can be trained. Representation of board is 1x6x7 array:

[[[0 0 0 0 0 0 0]