Notes on some of the implementation choices in this app:
- What projects are already out there?
- Hierarchical clustering is a potentially slow process. This notebook can give us a rough sense of how the execution time increases with larger arrays.
- After trying to find a clear way of having the various cross filters work together, we ultimately decided that the last user interaction is the one that should count. This is not the usual pattern in Dash, but here's a simple demo of the scheme.
- I've come to believe that quick-startup and scalability are contradictory requirements. Here's an explanation why, and a map of some of the ways forward.