Packing and unpacking

[donkirkby]

When I converted to one-hot game states, it seemed easy to save a lot of space with numpy's packbits(), but I'm worried that it will take a lot of time packing and unpacking. I think the real storage problem is for search trees and for some games' move histories.

A related problem is that I moved all of the extra settings out of the game state arrays to keep them purely ones and zeroes. That will cause a problem when I try to build neural networks, and they don't have access to all the features of a game state.

Suggested solution:

• Add create_snapshot() methods, similar to the one for Spargo, and store those in the search tree. Add a restore_snapshot() method.
• Add create_data() method that adds extra settings, maybe in another layer of the array, then use that for training data.