Point Cloud Generator
Points grouped around a few centres, with the true cluster label kept alongside each one. That label is what makes the set useful for testing a clustering algorithm — you can check whether it recovered the grouping rather than just producing one. The radii are square-rooted, so each cluster is evenly dense instead of piled up at its centre, which is the same correction a uniform disc needs.
What this generator does
Shares the points out evenly between cluster centres, drawing each at a random angle and a square-rooted radius from its own centre. It then reports the mean distance from a point to its own centre and to the nearest other, which is what separability comes down to.
How to use this tool
- Choose how many points, how many clusters, and how far each cluster spreads.
- Generate, and compare the two mean distances reported.
- Copy the CSV, which includes the true cluster label.
- Widen the spread until the clusters start to overlap.
Understanding the controls
- How many points
- Up to 120, shared evenly between the clusters so no cluster is starved.
- How many clusters
- Between 1 and 6 centres, placed away from the edges so clusters are not clipped.
- Cluster spread
- How far points scatter from their own centre. Push it up and the two mean distances converge, which is exactly when clustering stops working.
- Seed
- Reproduces the same centres and the same points.
Common use cases
- Test data for k-means and other clustering algorithms, with ground truth
- Scatter plot sample data with a known structure
- Showing when clusters stop being separable as spread grows
- Teaching what within-cluster and between-cluster distance mean
- Reproducing the same cloud from a seed
How this generator works
Each point takes a random bearing and a radius of spread times the square root of a uniform value, which spreads it evenly over the disc rather than concentrating it. The check confirms every point is inside the field, claims a cluster that exists, and that the clusters are balanced to within one point.
Randomness and fairness
Centres and point positions are random; the balanced allocation and the even density are properties of the construction. Seeded clouds reproduce exactly and are therefore explicitly not cryptographically secure, and unseeded ones use the browser's cryptographically secure generator.
For how randomness is produced across the whole site, see how Generate Random works.
Limitations and good to know
- Clusters are circular and all the same size — real data is rarely so obliging.
- Two dimensions only.
- Points are clipped to the field, so a wide spread flattens clusters against the edges.
- There is no noise class of points belonging to no cluster.
- Clouds are not stored between visits.
Privacy and your data
Points and their labels are generated in your browser. Nothing about the cloud or your seed is transmitted.
Related generators
- Poisson Disc Point GeneratorPoints with a guaranteed minimum separation, placed by Bridson's algorithm and measured against a uniform scatter of the same size.
- Clustered Dataset GeneratorPoints grouped around known centres with the true group recorded, so you can check what a clustering algorithm actually recovered.
- Convex Hull GeneratorRandom points with the smallest convex boundary around them, every point checked to lie on or inside it.
- Random Coordinate GeneratorA latitude/longitude coordinate, uniformly distributed across the globe — could land anywhere, ocean included.