Implement the other archetypes

[drvinceknight]

Currently in axelrod_dojo.archetypes I've implemented the FSMParams (and written unit and integration tests) which simplifies FSM_evolve.py. It's not tough work to do the same with the others (but they should be fully tested).

[gaffney2010]

I'd like to take a stab at HMMParams...

[drvinceknight]

👍 Go for it :)

[marcharper]

Note for HMM we can use either swarm algorithm or the evolutionary algorithm, so if it's possible to test both that would be ideal.

[drvinceknight]

The current implementation that is to be modified is here: https://github.com/Axelrod-Python/axelrod-dojo/blob/master/bin/hmm_evolve.py

[gaffney2010]

I'm looking at this now. Here's what I think the steps are:

1. Move HMMParams to hmm.py, similar to fsm.py.
2. Test HMMParams similar to archetypes/test_fsm.py.
3. Implement PSO algorithm, by implementing receive_vector, create_vector_bounds, and vector_to_instance in HMMParams. [Basically serializing the transition matrices and emission probabilities.] I will have to normalize in one or more of these functions, and the PSO will have to return unnormalized vectors.
4. Add "--algorithm" as an argument on hmm_evolve.py to decide which of the two implementations to use. And actually implement the PSO call, unserializing and normalizing.
5. Test PSO implementation.
6. Test similar to integration/test_fsm.py
7. Optionally, add some documentation.

I'm going to move forward with this plan, but let me know if I'm off-track or if I should do intermediate PRs.

Also: I haven't looked into the pyswarm implementation, but I think I understand the idea of the algorithm, and it seems safe to assume that no iteration of the algorithm will create values that are not probabilities [outside 0 to 1].

[drvinceknight]

That looks right to me 👍