New Subsystem With Forward and Backward TPMs

[ajbailey4]

Subsystem now contains forward and backward TPMs.

Three main new methods were added to subsystem.py:

1. sia - calculates sia for the current subsystem, previously done with backwards.sia
2. distinction - creates a distinction (Concept object) from the current subsystem and a given mechanism, previously done manually
3. all_distinctions - creates a CauseEffectStructure containing all candidate distinctions for a subsystem, previously done with backwards.compute_combined_ces

Other changes:

• backwards.py is no longer used, so it was deleted
• methods that previously required both backward and forward subsystems were updated to use just one
• node class also was updated to have both backward and forward TPMs
• some refactoring was done of repertoire.py, so that the methods in subsystem.intrinsic_information call self.forward_repertoire and self.unconstrained_forward_repertoire, which then call their respective repertoire methods. This allows for the caching to work there, and also eliminated the need for the directional dispatch methods in repertoire.py.
• validate.py was updated for subsystem to only check the forward TPM (should it check both?)

[ajbailey4]

There was also circular import situation with compute, subsystem and new_big_phi that I had to resolve by moving an import inside my new sia method which doesn’t seem ideal, but I couldn't figure out how to resolve it otherwise.

[isacdaavid]

@ajbailey4 I have one more request. Since you were already testing this against the jupyter notebooks, and given that this removes backwards, could you also update the notebooks so as to use your new subsystem API?

• docs/examples/IIT_4.0_demo.ipynb
• The unfolding one from the wiki (you should probably send it to Matteo or Jeremiah)