Distributions

The predict method of Golem requires to specify the type of input uncertainty via a list of probability distributions:

golem.predict(X=X, distributions=[Normal(std=0.1), Uniform(urange=0.5)])

Below are the distributions implemented and available in Golem. Note that, in addition to the distributions below, the Delta function can be used to indicate variables with no uncertainty:

golem.predict(X=X, distributions=[Delta(), Uniform(urange=0.5)])

Continuous Distributions

• Uniform
• Truncated Uniform
• Bounded Uniform
• Normal
• Truncated Normal
• Folded Normal
• Gamma
• Delta

Discrete and Categorical Distributions

• Poisson
• Discrete Laplace
• Categorical

Frozen Distributions

These are distributions that do not change location depending on the input sample. Contrary to the above classes, in the “frozen” distributions the location is fixed when instantiating the class. This may be useful when the input uncertainty is not due to a control factor one can influence, but it is caused by an environmental factor causing uncertainty in the conditions.

• Frozen Uniform
• Frozen Normal
• Frozen Gamma
• Frozen Poisson
• Frozen Discrete Laplace
• Frozen Categorical

Custom Distributions

If you would like to model input uncertainty using a distribution not available in Golem, you can provide any user-defined distribution as shown in the Custom Distributions example.