liveBook · Manning

Centroid

As explained in Grokking Bayes

A cluster center is a parameter representing the location of a component in a mixture model.

A centroid is a cluster’s central location: in a mixture model, μₖ represents the center of component k. For standardized data, each center can use a standard normal prior. section 8.6

Centroids become useful when interpreting and checking mixture models:

  • Each center describes a behavior represented by one cluster, such as short or long waiting times. section 8.7
  • Symmetric mixture models can switch the labels attached to centers, making posterior traces and distributions appear scrambled. section 8.7
  • Ordering the centers makes cluster identities consistent across samples without changing the model. section 8.7

Centers for Old Faithful waiting times

For standardized eruption-waiting data, define one center per cluster with a Normal(0, 1) prior. In PyMC, declare μ with dims="cluster"; an ordered transform can keep the smaller center before the larger one. section 8.6 section 8.7

How Grokking Bayes covers it

How deep it goes
Explains center priors, dimensionality, label switching, sampling symptoms, and ordered transformations, but not centroid computation from raw points.
Examples
Uses standardized Old Faithful waiting times, short- and long-waiting-time clusters, posterior traces, and PyMC model declarations.
What it assumes
Assumes a mixture-model context and standardized observations; teaches why center labels can switch and how ordering addresses it.
What you can do after
Operate/configure PyMC mixture models by assigning center priors and ordering centers to stabilize cluster identities.
How it connects
Connects centroids to mixture models, mixture weights, cluster widths, label switching, posterior sampling, and convergence diagnostics.

Also explained in

Compare how 10 books explain centroid →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Grokking Bayes. The section links point to the parts of the book it draws on.