liveBook · Manning

Credible interval

As explained in Grokking Bayes

A credible interval is a range containing a specified posterior or prior probability for a random variable.

A credible interval is a range containing a specified probability for a random variable. A 95% credible interval contains 95% of the distribution’s probability. For a normal distribution, it is approximately the mean plus or minus two standard deviations. section 3.12

The interval communicates uncertainty rather than only reporting a best estimate. A house-price slope with mean $100 and a large spread might have a 95% interval from −$100 to $300, expressing substantial uncertainty about the estimate. section 3.12

Credible intervals are used to:

  • Show uncertainty around posterior predictive means with error bars. section 3.15
  • Indicate that predictions become less certain where observed data are scarce. section 10.3
  • Represent uncertainty from both the fitted model and observational noise in probabilistic predictions. section 3.15 section 10.3

House-price predictions

A probabilistic linear regression model produces a predictive mean for each test value and a corresponding 95% credible interval. Most observed points fall within the displayed intervals, while the posterior mean combines information from the data and the prior. section 3.15

Bayesian neural-network predictions

After MCMC samples the network parameters, predictions are shown as a mean curve with 95% credible intervals. The intervals cover much of the observed data and widen outside the training-data range. section 10.3

Figure 3.6 The posterior predictive distribution of individual datapoints in the test set, illustrated by the predictive mean and 95% credible intervals. Most datapoints in the training set falls within the credible intervals.

Visualizes predictive means and 95% credible intervals for individual test points. section 3.15

How Grokking Bayes covers it

How deep it goes
Defines intervals and gives the normal-distribution rule, then applies them to regression and neural-network predictions.
Examples
Uses house-price regression intervals and Bayesian neural-network intervals that widen where training data are absent.
What it assumes
Assumes familiarity with probability distributions, means, standard deviations, priors, posteriors, and predictive distributions.
What you can do after
Evaluate uncertainty by interpreting interval width and visualizing predictive intervals around model means.
How it connects
Connects prior distributions, posterior predictive distributions, normal distributions, MCMC, and uncertainty quantification.

Also explained in

Compare how 2 books explain credible interval →

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.