Seed methods

[maxhully]

We have a bunch of seed methods sitting in #248 that we should clean up, merge in, and document. But we'll also want to think about how we can present those methods in the clearest, most responsible manner.

[drdeford]

Basic outlines of the current versions: https://www.overleaf.com/read/zpmyzqmpvmnx plus some additional code for context: https://github.com/drdeford/Tree_Partitions_Preliminary

[drdeford]

This also fits into #296 as the construction of new seeds sits somewhere in between chain runs, data processing, and implementation details.

• Do we include seed building in GerryChain? (The underlying graph and data objects are the same + you can use any such method as an i.i.d. proposal)

• Is it closer to MAUP style preprocessing? (Could be a separate project taking that just also happens to make use of make graph)

• Should we offer the opportunity to generate a new seed every time you generate an ensemble? (I've used this approach on some states)

[maxhully]

I'm intrigued by the idea of making it a separate project. That would make it easier to include additional seed methods without having to worry about putting too many things in GerryChain.

Being able to generate a new seed every time you generate an ensemble would be useful. I suppose MarkovChain could optionally take a seed method as a parameter instead of an initial plan? Figuring out a standard interface for seed methods (and bipartitioning methods) would be helpful, although probably impossible.

Here's an idea I had about the bipartitioning methods. Suppose we make our constraints into callable classes instead of functions, looking something like this:

>>> pop_constraint = PopulationConstraint("TOTPOP", ideal=1250, epsilon=0.05)
>>> pop_constraint(initial_partition)
True

Then maybe we could standardize the bipartitioning methods to always be callables with the signature (graph, population constraint) -> partition?

Independent of this specific example, I think it would be good to explore using callable objects for the situations where we find ourselves using partial.

[drdeford]

Now a week 6 project for vrdi 2019.