v0.14

• Caching of partially computed states in the neural network during the interpretation search.
• Refactoring of the interpretation search. The search is now iterative to prevent stack overflows. The debugging output is much more detailed now.
• Ongoing work on the training algorithms.