Concepts¶
Three pages of background. Nothing here tells you which button to press — that is what the how-to guides are for. This is the model the app is built on, and it is what makes the buttons make sense.
Embeddings, distances, features¶
The three kinds of data a project can hold, how they relate, and why what your project ships decides what the app can do. Start here.
Why data sources are directories¶
BigClust takes a directory, not a matrix. Where that idea comes from and what it buys you.
How selection works¶
Selection is the app's central verb. What it propagates to, and how growing and shrinking actually work.
The short version¶
BigClust holds one set of observations — normally neurons, one row in meta
each — and shows them three ways at once: as points in a 2D scatter plot, as
morphology in a 3D viewer, and as rows in whichever widgets you have open.
Everything else follows from that:
- The three views are joined by selection. Selecting points is how you ask every other part of the app to talk about the same neurons.
- The 2D positions come from an embedding, which is derived from distances or features. Which of those your project ships decides what can be recomputed.
- The whole thing is loaded from a directory described by an
infofile, which is why a clustering can be published at a URL.
If you read only one of these pages, read embeddings, distances, features — it is the distinction people trip over.