Magic Lasso
Magic Lasso is a shortcut for selecting one whole cluster without tracing its boundary. You draw a rough lasso, Mantis scores every precomputed cluster by how much its membership overlaps the points you enclosed, and the selection becomes the entire best-matching cluster. It is a set-overlap heuristic over the cluster assignments already on the map, not a similarity search.
- Shortcut
- m
- Available on
- Map, Map (Legacy)
Where to find it
The Magic button sits third in the tool island, or press
M. It is on both the default Map and Map (Legacy).How it picks the cluster
For each cluster, Mantis divides the number of its points that fell inside your lasso by the size of the union of the cluster and your lasso, the Jaccard overlap. The cluster with the highest score wins, and the selection is set to every point in that cluster.
Two consequences follow, and they matter:
- Points you lassoed that do not belong to the winning cluster are dropped.
- Points in the winning cluster that you never lassoed are added, including ones off screen.
The result is a substitution, not an expansion of what you drew. If your lasso catches no point belonging to any cluster, nothing happens and the selection is left alone.
How to use it
- Activate Magic. Click the button or press M.
- Drag a rough loop through the part of the cluster you care about. A stroke through the dense core scores better than one that skims two clusters.
- Release. The selection snaps to the winning cluster. On Map (Legacy) a purple outline flashes around the cluster for about a second to show you what it picked.
Limits
- The tool can only return clusters that already exist in the space. It does not compute similarity, consult embeddings, or find relationships that are not already part of the cluster hierarchy.
- Only one cluster is ever returned. To combine two, use plain Lasso and add each to a bag.
- On the default Map the commit is capped at 25,000 points, so a very large winning cluster is truncated silently.