liveBook · Manning

Parameter

As explained in Grokking Bayes

A parameter is a quantity that controls the behavior of a model or probability distribution.

A parameter is a number that controls how a probability distribution or model behaves. Suppose you think there is a 30% chance of rain today. A Bernoulli distribution can represent that belief with parameter p = 0.3: the probability of rain is 0.3, and the probability of no rain is 1 − p = 0.7. Changing p changes the distribution. section 1.5

A model parameter can also be something you do not know yet. In a Bayesian model, you place a prior distribution over an unknown parameter, then update it using data to obtain a posterior distribution. Keeping a distribution rather than only one chosen value expresses uncertainty about the parameter. section 10.0

The same idea appears in models of different sizes:

  • In house-price regression, the slope m controls how a home's area x relates to its price y. Observed prices help update a prior belief about m; random noise allows prices to depart from the linear relationship. section 3.13
  • In a hierarchical model, each group has a parameter drawn from a shared population distribution, rather than an entirely independent parameter. Coaching-effect estimates from eight schools motivate this setup. section 7.1
  • In a Bayesian neural network, weights and biases are parameters. Distributions over them represent multiple plausible settings and support predictions that express uncertainty. section 10.0

The practical question is what changing a parameter changes: an outcome probability, a regression relationship, or a network prediction. When that parameter is unknown, also ask how certain the data make you about its value. section 1.5 section 10.0

How Grokking Bayes covers it

How deep it goes
Defines a distribution parameter with a worked Bernoulli example and develops a regression slope through its prior and likelihood.
Examples
Illustrates parameters with rain probability, a house-price slope, school coaching effects, and neural-network weights and biases.
What you can do after
Evaluate/understand how parameter choices affect probabilities or predictions and why uncertain parameters can be represented by distributions.
How it connects
Connects parameters to Bernoulli distributions, priors, likelihoods, posterior distributions, hierarchical models, and Bayesian neural networks.

Also explained in

Compare how 41 books explain parameter →

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.