This repository holds 3 notebooks that illustrates the bayesian approach and how it can be used in your project.
The basics of the bayesian approach is that the parameters of a model are not point estimates but distributions that evolves with the observations. In most cases we can divide the approach into these steps:
- Define a model for your problem
- Put a prior distribution on your model parameters
- Apply Bayes rule iteratively, for each iteration take the posterior of the previous iteration as your new prior
The Bayes rule is:
P(A | B) = P(B | A) P(A) / P(B)
Its terms are:
- P(A) is the prior
- P(B | A) is the likelihood
- P(B) is the regularization parameter. It can be computed as a sum (discrete or continuous) of likelihoods for every value of A
- P(A|B) is the posterior