Draft questions for clustering and regionalization chapter #14
Conversation
|
Sorry for the blending of the cluster and choro chapters. I got confused on the docker stuff and got branches out of sync. I will separate these into different branches and prs in the next few days. |
Add questions for the weights & Local Autocorrelation notebook
|
Hey @sjsrey, I've rebased your PR and synced the jupytext representations up. This was particularly tricky, given that I've made sure that this has the questions for the weights,choro,local autocorrelation, & regionalization chapters in both markdown & ipynb representations, and believe this should be merged now that it's been rebased & synced. If you could double-check that your questions are there, I think this has been successfully merged with the changes from #16 #17 |
|
No prob, this stuff happens 😄 @darribas can you check to be sure that this looks right to you before it merges? |
|
Will look at them this week as my first task for book matters! |
There was a problem hiding this comment.
I think overall they're cool questions. My only concern is whether they're too open-ended. They don't really help solidify the concepts seen in the chapter, but they are very good at inviting the reader to think above and beyond those. This might be more of the purpose we might want to pursue.
| 3. In evaluating the quality of the solution to a regionalization problem, how might traditional measures of cluster evaluation be used? In what ways might those measures be limited and need expansion to consider the geographical dimensions of the problem? | ||
| 4. Discuss the implications for the processes of regionalization that follow from the number of *connected components* in the spatial weights matrix that would be used. | ||
| 5. True or false: The average silhouette score for a spatially constrained solution will be no larger than the average silhouette score for an unconstrained solution. Why, or why not? (add reference and or explain silhouette) | ||
| 6. Consider two possible weights matrices for use in a spatially constrained clustering problem. Both form a single connected component for all the areal units. However, they differ in the sparsity of their adjacency graphs (think Rook versus queen). How might this sparsity affect the quality of the clustering solution? |
There was a problem hiding this comment.
We could recast this question to make it more practical, along the lines @ljwolf does in other chapters. For example, on this one, it could be:
- Re-run the analysis in the chapter w/ a different set of weights.
- Compare the resulting clusters visually
- What are the key differences between the two W's?
- How do you think such differences affect the final result?
| 4. Discuss the implications for the processes of regionalization that follow from the number of *connected components* in the spatial weights matrix that would be used. | ||
| 5. True or false: The average silhouette score for a spatially constrained solution will be no larger than the average silhouette score for an unconstrained solution. Why, or why not? (add reference and or explain silhouette) | ||
| 6. Consider two possible weights matrices for use in a spatially constrained clustering problem. Both form a single connected component for all the areal units. However, they differ in the sparsity of their adjacency graphs (think Rook versus queen). How might this sparsity affect the quality of the clustering solution? | ||
| 7. What are the challenges and opportunities that spatial dependence pose for spatial cluster formation? |
There was a problem hiding this comment.
This one I think it's pretty hard for an introductory text.
| 5. True or false: The average silhouette score for a spatially constrained solution will be no larger than the average silhouette score for an unconstrained solution. Why, or why not? (add reference and or explain silhouette) | ||
| 6. Consider two possible weights matrices for use in a spatially constrained clustering problem. Both form a single connected component for all the areal units. However, they differ in the sparsity of their adjacency graphs (think Rook versus queen). How might this sparsity affect the quality of the clustering solution? | ||
| 7. What are the challenges and opportunities that spatial dependence pose for spatial cluster formation? | ||
| 8. In other areas of spatial analysis, the concept of multilevel modeling (cites) exploits the hierarchical nesting of spatial units at different levels of aggregation. How might such nesting be exploited in the implementation of regionalization algorithms? What are some possible limitations/challenges that such nesting imposes/represents in obtaining a regionalization solution. |
There was a problem hiding this comment.
I'm not sure I understand this one
|
OK, I've had a look at them and left some comments. I'd merge this anyway to avoid further conflicts and, if needed, we can open future PRs that feed into the chapter. |
No description provided.