Skip to main content

Infer.NET user guide

Factors and Constraints

Infer.NET has a number of inbuilt factors and constraints which you can use to build your models. These are normally exposed as static methods on the Variable class, such as Variable.Discrete(). You can also add your own factors or constraints and include them in your model by using Variable<T>.Factor or Variable.Constrain.

Factors are functions that define a new variable in terms of other existing variables. A factor can be stochastic (even deterministic inputs produce an uncertain output) or deterministic (deterministic inputs will produce a deterministic output). Constraints act on existing variables and do not define a new variable. Below we mainly categorize factors in terms of the type of their output variable; constraints and experimental factors relating to undirected models are documented separately in their own sections.

Note: some factors are less mature than others. For example, their implementation may only support a limited class of models or algorithms, and they may not be as thoroughly tested as other factors. The list of factors and constraints shows an assessment of the maturity of each factor using the mechanism of quality bands. If you have an issue or question concerning an experimental factor, please raise it in the forums.