Skip to content

Latest commit

 

History

History
56 lines (37 loc) · 2.21 KB

units.md

File metadata and controls

56 lines (37 loc) · 2.21 KB
CurrentModule = Distributions  # So that non-exported names don't need to be prefixed
                               # to be referenced.

Physical Units

There is currently no common interface for units and physical dimensions in the Julia ecosystem. Therefore, we directly support the package that is the most widely used and evolved: Unitful.

Usage example

Supported distributions

Distributions that currently support unitful variates and (when appropriate) unitful parameters:

How it works, or: How to support other distributions

If you want to add unit support to a distribution that currently only supports Reals (whether this distribution is in Distributions's source, or custom made according to Create a Distribution), follow the steps below.

1. Change the type of the variate from Real to RealQuantity, wherever it appears.

# example of changes, eg `pdf(d::Gamma, x)`

RealQuantity

2. Identify distribution parameters that should have the same units as the variate, and change their type to RealQuantity, too.

# before after example

If there are no such parameters (as in e.g. LogNormal), add a "U" type parameter to your distribution (with U <: Units). Add a units keyword argument to the distribution's constructors, with a default value of NoUnits:

Unitful.NoUnits is a 0-size singleton that gets pruned away in mathematical expressions at compile-time

3. Define Base.eltype(::Type{YourDist}) to return the variate's data type.

  • This is necessary for rand(::YourDist, dims) to work correctly, as Julia's Random API asks for eltype to be implemented so that it can pre-allocate arrays with the correct element type.
  • eltype is also used by unit to determine the distribution's units.

4. Wherever the log of a RealQuantity is taken, scale and convert the argument to a Real using the provided unit method.

# also mention extrema
# and that `oneunit` is useful.

unit