TODO.txt

- Forget the OOP abstractions. Make the codebase simpler. It is trying to be too generic trying to comply to future refactors. 
ANY ADDITION OTHER THAN CORE FUNCTIONALITY CAN BE ANOTHER REPO! KEEP THIS ONE CLEAN!

- Implement explored belief region as a reachability graph. It should have an "expand" funcion which uses an ExplorationStrategy to expand the number of beliefs. This will allow you to use different exploration strategies on the same belief region.

x Add float quantization where DDs are being cached / or used in hashmaps.

- Add softmax exploration instead of SSGA / SSEA

- Write a simpler domain parser with some limited LISP-like functionality.

- Serialize policies and models in ways that are human-readable / lisp-like and can be easily loaded and parsed into usable objects

- Maybe limit the number of variables to say 1000. That way, primed vars can be defined 1001 onwards. This will greatly simplify priming / unpriming operations.

Also, an added advantage will be to dynamically add variables after defining agents. Eg: level 0 POMDP vars can be defined first. Then after solving level 0, you can define level 1 interactive state space using the number of opponent models computed from level 1. This can be done by adding the Mj state variable after level 0 is defined.