What is Beta Distribution?
Beta distribution is the continuous probability distribution of all unknown probabilities in a model. This distribution represents all a probability’s possible values when the probability itself is unknown. In statistical terms, beta distribution is a dynamic, continuously updated probability distribution with two parameters. Its most common use in machine learning is modeling the uncertainty about the probability of success of a given experiment.
How does Beta Distribution Work?
With a typical experiment there are only two outcomes--either success, with probability X, or failure, with probability 1-X. If X is unknown, then all its possible values are equally likely. This uncertainty is expressed by assigning X a uniform distribution across that interval.
Being a probability, X can only assume values between 0 and 1. In addition, this uniform distribution means all points in the interval have an equal probability density, so no single possible outcome is more likely than another.
As trials are run, more data comes in to refine this probability distribution. This “conditional” distribution takes into account actual observed successes and failures. The end result is a real-time beta distribution.
