Change state and reward design

[Edouard360]

Different shapes for looking around the square might be possible.

• Maybe use a mask ? In tensorflow or in python
• For now the game state is 27 number, and it is hard-coded, which is bad for modularity... We might have states of various sizes

[leonardhussenot]

• We might want to implement a reward depending on production/(distance to border). Typically, if we are far from the border, loosing strength is not important (as we want it to go to the borders) and we shouldn't penalize square giving their strength. Same thing if the square is not really productive, it is not really important.

• On the other side, we should reward a productive square for being productive.

[Edouard360]

Solved the hard coding of the state size, creating a state object, and a strategy.json file