Uniquely identify people with nodes; use HealthService to manage hospitalization

[glwagner]

@odunbar, @agarbuno, and I propose to make the size of the contact network equal to the size of the population, omitting placeholder nodes.

The ContactSimulator will simulate all possible contacts, which includes contacts between healthcare workers and hospital beds.

Given the hospitalization state of a population, some of these edges are "phantom edges", since they represent connections between community members and people who are hospitalized, or connections between heath care workers and empty hospital beds.

In consequence:

• len(contact_network) = population
• n_contacts = population + hospital_beds

The HealthService keeps track of who is in the hospital. This mapping determines the structure of the contact network, because when a node is hospitalized, their neighbors change. The HealthService will also manage setting the weights of the contact network, since the HealthService manages the mapping of contacts (which includes "phantom edges") to "actual" edges on the contact network.

Both the KineticModel and the MasterEquationModelEnsemble will simulate an epidemic on the "actual" contact network, which does not include placeholder nodes.

There are two advantages of this approach. One is that it corresponds more closely to the "true" system --- nodes represent people, and "hospitalization" changes the contact network, not the "identity" of a node.

The second advantage is corollary to the first: clinical information attached to nodes does not need to be "copied" to placeholder nodes when somebody is hospitalized (which requires extra lines of code for the correct simulation of an epidemic with correct rates). "Hospitalization" merely implies a change in the contact network, not a change in the clinical parameters of a node. This reduces the complexity of the code; the hospitalization procedure is the sole domain of the HealthService.

[glwagner]

A few things discussed with @odunbar:

• HealthService has .hospital_beds
• there is a contact network associated with HealthService which simulates contacts between health care workers and hospital_beds. I'm not sure the best way to abstract ideas, but HealthService may need to run its own ContactSimulator. Food for thought.
• The HealthService manages connecting occupied .hospital_beds with health care workers, and ensuring that edges between patients in .hospital_beds and the community are severed.
• The HealthService needs to keep track of edges between patients in .hospital_beds and the community, so that when patients are .discharged, those edges can be restored.
• We should create a contact network for each HealthService for the given population we are considering. Basically, we need to identify health care workers in a given population (who are connected to hospital_beds, as well as specify the number of hospital_beds.
• If we could come up with a randomized / statistical way of selecting hospital workers from a population, that would probably simplify a lot of things. In other words, we can instantiate the HealthService with information about the population (mostly ages), as well as our specified number of hospital beds. This will allow us to quickly understand the effect of changing the number of hospital beds on the evolution of the epidemic.
• there's probably a lot more, but I think this is a good start.
• HealthService is essential!