feat: add adjacency matrix #177
Conversation
Previously adjacency is checked for every data point when calculating topographic error. Instead check adjacency for each pair of nodes in the map and use it as a lookup.
|
Hi, thanks for submitting the pull request. This has little chances to be merged. The reason is that it introduces a mechanism to check the adjacency of nodes introducing new methods, but other methods use different ways to do so (see I like the idea of using the adjacency matrix, but only if it's used consistently across the other methods. |
|
if we replaced the adjacency checks elsewhere with usage of the adjacency matrix, would that be acceptable? |
|
It seems quite a big a change and doesn't add much value. Exposing only the adjacency matrix with a public method is also acceptable. The current priority is fixing this bug first: #172 |
|
Coming back to this, i think the only places where adjacency is checked is in topographic error (both hex and rectangular would be replaced with one check to adjacency matrix) and distance map (which I believe can be simplified with a check to adjacency matrix, although unsure about efficiency impacts of indexing into weights vs iterating through the 6 possible neighbours). The other distance checks are the neig functions that all calculate distance differently anyway. Either way, happy to make the change to distance_map or to revert the change to topographic error and expose the adjacency matrix. |
Previously adjacency is checked for every data point when calculating topographic error. Instead check adjacency for each pair of nodes in the map and use it as a lookup.
Briefly discussed here: #176