Investigate back propagation for training

[cohen990]

for back propagation, we will need a value function

the value function is able to evaluate the potential value of a current board state. There are two ways to do this:

1. Algorithmically - we write an algorithm that can assess a tetris board
2. Using a neural network - the neural network is able to assess a board position and the likelihood of it leading to point gains later.

This requires investigation

[cohen990]

here is a paper which describes how to build a neural network to assess the value of a chess game - it may come in useful https://arxiv.org/pdf/1509.01549.pdf