Naive MCMC sample process

[thekingofkings]

Thanks to Fei's sharing MCMC procedure

The following needs to be done:

1. A sampling method. Sample a tract and permute its label.
2. Recalculating the features of all communities.
3. Evaluate the new partition (this is finished I believe).
4. Iterative step 2 and 3 alternatively.

[thekingofkings]

Tract adjacency graph

Note that two tracts are said to be adjacent only when they share at least one edge. Here is a visualization of the tract adjacency graph.

[thekingofkings]

Sampling method constraint

Notice that we should have a constraint on the CA size. Otherwise, when there are only two communities left, the MAE reach minimum, which is a trivial solution and the MAE is 0.

[thekingofkings]

Another sampling method constraint

The naive MCMC method does not guarantee the spatial continuity.