Build Graph from H3

[martinfleis]

I have an implementation of building a (higher-order) contiguity Graph based on H3 index. I'd like to expose it as a builder on `Graph, so dumping the code below before I'll forget.

import h3

from libpysal import graph

data = ... # dataframe with H3 IDs as an index, like those coming from tobler h3fy
order = 5

neighbors = {}
weights = {}
for ix in data.index:
   rings = h3.hex_range_distances(ix, order)
   for i, ring in enumerate(rings):
       if i == 0:
           neighbors[ix] = []
           weights[ix] = []
       else:
           neighbors[ix].extend(list(ring))
           weights[ix].extend([i] * len(ring))

matrix = graph.Graph.from_dicts(neighbors, weights)

It encodes the order as weight now, I was casting to a binary graph and used topological inverse weighting based on that (inverse = graph.Graph(1 / matrix.adjacency)). We can figure out what the API should be once the PR is alive?

I also think that the latest beta release of h3 completely changed the API so the h3 calls will need to change a bit but it's been beta for a year.

based on https://github.com/Urban-Analytics-Technology-Platform/demoland-project/blob/main/code/04_generalisation/create_spatial_weights.ipynb

[ljwolf]

It would be interesting to try to generalise this for many of the common hierarchical indices...

[knaaptime]

I also think that the latest beta release of h3 completely changed the API so the h3 calls will need to change a bit but it's been beta for a year.

you think that's related to these failures somehow? not sure what would've changed

[martinfleis]

@knaaptime no, the version of h3 has not changed in those test envs. neither did GEOS. not sure where that comes from.