Blog: Machine Learning Equations by Saurabh Verma

Blog: Dirichlet Distributions.

• Tutorials

Dirichlet Distribution is a distribution over distributions. More specifically, it is a distribution over pmfs (probability mass functions). You can imagine, as if there is a bag of dices, and each dice has a corresponding pmf (related to six possible outcomes). Now picking a dice is like sampling a particular pmf from a distribution. The probability of picking a dice, which results in a pmf, comes from the Dirichlet distribution .

Let be a pmf (or a point in simplex ), where and . Here is dirichlet parameter and . Then, the probability density of is given by:

Graphical Model:

Suppose is a set of samples drawn from pmf where . For eg. could be a sequence of outputs of a dice . Let are pmfs. Then:

Let be the number of outcomes in sequence of samples that is equal to event where , and let .

Therefore,

  1. Bela A. Frigyik, Amol Kapila, and Maya R. Gupta. “Introduction to the Dirichlet Distribution and Related Processes”. [PDF]
comments powered by Disqus