Add example that uses neural net instead of Table

[barrybecker4]

TicTacToe has only a few thousand states, but for most applications the number of states will be more than will fit in memory. In those cases, some sort of approximation like nerual nets must be used.
Using a neural net on the trivial TTT example will provide some good intuition as to how the nn works and can be applied more generally.

[barrybecker4]

See https://medium.com/@shiyan/get-a-taste-of-reinforcement-learning-implement-a-tic-tac-toe-agent-deda5617b2e4

http://blog.karmadust.com/training-a-tic-tac-toe-ai-with-reinforcement-learning-part-2/

https://davidsanwald.github.io/2016/12/11/Double-DQN-interfacing-OpenAi-Gym.html

[barrybecker4]

https://dratewka.wordpress.com/2013/03/15/ai-overkill-teaching-a-neural-network-to-play-tic-tac-toe/
https://github.com/rahular/TicTacToe
http://colinfahey.com/neural_network_with_back_propagation_learning/neural_network_with_back_propagation_learning.html

Back propagation explained:
https://mattmazur.com/2015/03/17/a-step-by-step-backpropagation-example/

General Alpha-zero
https://github.com/suragnair/alpha-zero-general