liveBook · Manning

Categorical distribution

As explained in Grokking Bayes

A probability distribution over a finite set of mutually exclusive categories with associated probabilities.

A categorical distribution represents uncertainty about one outcome among a finite set of mutually exclusive categories. Each category receives a non-negative probability, and all probabilities must sum to 1. section 1.5

For rainfall, the categories might be no rain, 0.01 inches per hour, and 0.02 inches per hour. A belief such as 80%, 10%, and 10% describes uncertainty about which outcome will occur. section 1.5

Categories do not have to be exact numerical values:

  • Use exact values when each possible outcome is represented separately, such as 0.01 or 0.02 inches of rain per hour. section 1.5
  • Use numerical ranges when outcomes are grouped, such as 0 to 0.01 inches or 0.01 to 0.02 inches. section 1.5
  • Use a categorical distribution for a discretized quantity when rounding or dividing a continuous scale into intervals. section 1.5

A discrete prior for a proportion

Suppose a proportion p can take only the values 0.0, 0.1, 0.2, …, 1.0. A categorical distribution assigns prior probability to each of these eleven possibilities. Equal probabilities express no preference; higher probabilities near 0.5 express a belief that middling values are more likely. section 2.9

Figure 1.2 An example categorical distribution for rainfall rate.

Visualizes a categorical distribution for rainfall rates, making the relative probability assigned to each category easy to compare. section 1.5

How Grokking Bayes covers it

How deep it goes
Defines categories and probability constraints with examples, but defers continuous probability distributions and greater numerical accuracy.
Examples
Uses rainfall amounts, rainfall ranges, and rounded proportion values from 0.0 through 1.0.
What it assumes
Builds from Bernoulli beliefs and establishes non-negativity and summation-to-one probability axioms.
What you can do after
Apply categorical probabilities to represent beliefs about discrete outcomes and compare uniform, moderate, or strongly weighted assignments.
How it connects
Connects Bernoulli distributions, probability axioms, prior distributions, discretization, and continuous probability distributions.

Also explained in

Compare how 5 books explain categorical distribution →

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.