Distribution Functions

We support the following distributions:

• Cauchy Distribution
• Laplace Distribution
• Log Normal Distribution
• Normal Distribution

Connector Parameters

Each distribution connector takes in two parameters, function and value.

Function

For the function parameter, you can select "CDF" to use the cumulative density function, "PDF" to use the probability density function, or "PPF" to use the percentile point function. You can also select "SF" for the survival function, which is 1 minus the CDF.

Value

The value parameter determines the value at which to call the selected function. If "PPF" is selected, the value must be between 0.0 and 1.0.

Input Specifications

The distribution parameters - like μ and σ for Normal Distribution - enter as time series inputs. These will often be provided as outputs of the NGBoost model connector. Our supported distributions use a location and scale parameter (for Normal distribution, μ is the location parameter, and σ is the scale parameter).

Tutorial

See our Create a Price Forecast with Probabilistic Prediction tutorial for examples using the LaplaceDistribution operation connector.