How-to guides¶

Each page here is one task, as a numbered recipe. They assume you can already open a project and select points.

Getting data in¶

Create a local dataset ¶

Build a project from scratch — a worked NBLAST of hemibrain neurons, with the 3D viewer wired up. Start here if you have data but no project yet.

Load a remote dataset ¶

Open a clustering published at a URL, and what to do when it won't open.

Load part of a dataset ¶

Filter at load time with a query or an ID list, so you work with the 3,000 neurons you care about instead of 300,000.

Changing what you see¶

Recompute the embedding ¶

Swap UMAP for t-SNE, change the metric, drop feature groups — and watch the points move.

Recluster the data ¶

Run HDBSCAN, agglomerative, k-means or spectral clustering and apply the result as labels.

Find what separates two groups ¶

Rank the features that distinguish one set of neurons from another, and test whether a split is real.

Getting data out¶

Refresh the meta data ¶

Pull current annotations from neuPrint, Clio or FlyTable into your project, or merge in a local file.

Push annotations ¶

Write the types you settled on back to Clio, FlyTable or a CSV.

Export your work ¶

Meta data, cluster assignments, an interactive Plotly page, or a full project snapshot.

Working comfortably¶

Work in several views ¶

Tabs, tear-off windows, and keeping a subset of neurons in its own view.

If you are looking for what a specific control does rather than how to complete a task, that is the widget reference. If a recipe says something you don't have the background for, the concepts pages are the missing half.