Lab — Break the Clusters

Open the Clustering Playground.

Begin with k = 3. Do not name the groups. First describe only what the visualization supports: which points are grouped together under the current settings.

Then run three controlled changes:

1. change k;
2. scale the Y feature;
3. add the outlier.

A strong observation sounds like:

With k = 3 and the default representation, the points formed three visible groups. Doubling the Y scale changed at least one assignment/centroid, showing that the grouping depends on representation. This simulation does not establish three real machine types.

Evidence to save

• one baseline screenshot or written cluster description;
• the setting you changed;
• what changed in assignments or centroids;
• one useful interpretation;
• one interpretation you refuse to make.

The lab is deterministic. It is designed to make assumptions visible, not to identify real-world categories.